Package Keras

Package ‘keras’
April 5, 2019
Type Package
Title R Interface to 'Keras'
Version 2.2.4.1
Description Interface to 'Keras' <https://keras.io>, a high-level neural
networks 'API'. 'Keras' was developed with a focus on enabling fast experimentation,
supports both convolution based networks and recurrent networks (as well as
combinations of the two), and runs seamlessly on both 'CPU' and 'GPU' devices.
Encoding UTF-8
License MIT + file LICENSE
URL https://keras.rstudio.com
BugReports https://github.com/rstudio/keras/issues
Depends R (>= 3.2)
Imports generics (>= 0.0.1), reticulate (>= 1.10), tensorflow (>=
1.13.1), tfruns (>= 1.0), magrittr, zeallot, methods, R6
Suggests ggplot2, testthat, knitr, rmarkdown
SystemRequirements Keras >= 2.0 (https://keras.io)
RoxygenNote 6.1.1
VignetteBuilder knitr
NeedsCompilation no
Author Daniel Falbel [ctb, cph, cre],
JJ Allaire [aut, cph],
François Chollet [aut, cph],
RStudio [ctb, cph, fnd],
Google [ctb, cph, fnd],
Yuan Tang [ctb, cph] (<https://orcid.org/0000-0001-5243-233X>),
Wouter Van Der Bijl [ctb, cph],
Martin Studer [ctb, cph],
Sigrid Keydana [ctb]
Maintainer Daniel Falbel <[email protected]>
Repository CRAN
Date/Publication 2019-04-05 21:30:08 UTC
1
2 R topics documented:
R topics documented:
keras-package . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
activation_relu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
application_densenet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
application_inception_resnet_v2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
application_inception_v3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
application_mobilenet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
application_mobilenet_v2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
application_nasnet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
application_resnet50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
application_vgg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
application_xception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
backend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
bidirectional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
callback_csv_logger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
callback_early_stopping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
callback_lambda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
callback_learning_rate_scheduler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
callback_model_checkpoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
callback_progbar_logger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
callback_reduce_lr_on_plateau . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
callback_remote_monitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
callback_tensorboard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
callback_terminate_on_naan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
clone_model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
compile.keras.engine.training.Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
count_params . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
create_layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
create_wrapper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
dataset_boston_housing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
dataset_cifar10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
dataset_cifar100 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
dataset_fashion_mnist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
dataset_imdb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
dataset_mnist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
dataset_reuters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
evaluate.keras.engine.training.Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
evaluate_generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
export_savedmodel.keras.engine.training.Model . . . . . . . . . . . . . . . . . . . . . . 45
fit.keras.engine.training.Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
fit_generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
fit_image_data_generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
fit_text_tokenizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
flow_images_from_data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
flow_images_from_dataframe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
flow_images_from_directory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
R topics documented: 3
freeze_weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
generator_next . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
get_config . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
get_file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
get_input_at . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
get_layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
get_weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
hdf5_matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
imagenet_decode_predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
imagenet_preprocess_input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
image_data_generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
image_load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
image_to_array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
initializer_constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
initializer_glorot_normal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
initializer_glorot_uniform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
initializer_he_normal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
initializer_he_uniform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
initializer_identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
initializer_lecun_normal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
initializer_lecun_uniform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
initializer_ones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
initializer_orthogonal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
initializer_random_normal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
initializer_random_uniform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
initializer_truncated_normal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
initializer_variance_scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
initializer_zeros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
install_keras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
is_keras_available . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
KerasCallback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
KerasConstraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
KerasLayer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
KerasWrapper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
keras_array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
keras_model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
keras_model_custom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
keras_model_sequential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
k_abs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
k_all . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
k_any . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
k_arange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
k_argmax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
k_argmin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
k_backend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
k_batch_dot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
k_batch_flatten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
k_batch_get_value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
k_batch_normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
k_batch_set_value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
k_bias_add . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
k_binary_crossentropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
k_cast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
k_cast_to_floatx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
k_categorical_crossentropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
k_clear_session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
k_clip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
k_concatenate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
k_constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
k_conv1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
k_conv2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
k_conv2d_transpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
k_conv3d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
k_conv3d_transpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
k_cos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
k_count_params . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
k_ctc_batch_cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
k_ctc_decode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
k_ctc_label_dense_to_sparse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
k_cumprod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
k_cumsum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
k_depthwise_conv2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
k_dot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
k_dropout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
k_dtype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
k_elu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
k_epsilon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
k_equal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
k_eval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
k_exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
k_expand_dims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
k_eye . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
k_flatten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
k_floatx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
k_foldl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
k_foldr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
k_function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
k_gather . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
k_get_session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
k_get_uid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
k_get_value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
k_get_variable_shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
k_gradients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
k_greater . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
k_greater_equal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
k_hard_sigmoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
k_identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
k_image_data_format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
k_int_shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
k_in_test_phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
k_in_top_k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
k_in_train_phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
k_is_keras_tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
k_is_placeholder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
k_is_sparse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
k_is_tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
k_l2_normalize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
k_learning_phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
k_less . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
k_less_equal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
k_local_conv1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
k_local_conv2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
k_log . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
k_logsumexp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
k_manual_variable_initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
k_map_fn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
k_max . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
k_maximum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
k_mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
k_min . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
k_minimum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
k_moving_average_update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
k_ndim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
k_normalize_batch_in_training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
k_not_equal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
k_ones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
k_ones_like . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
k_one_hot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
k_permute_dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
k_placeholder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
k_pool2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
k_pool3d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
k_pow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
k_print_tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
k_prod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
k_random_binomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
k_random_normal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
k_random_normal_variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
k_random_uniform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
k_random_uniform_variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
k_relu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
k_repeat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
k_repeat_elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
k_reset_uids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
k_reshape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
k_resize_images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
k_resize_volumes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
k_reverse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
k_rnn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
k_round . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
k_separable_conv2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
k_set_learning_phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
k_set_value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
k_shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
k_sigmoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
k_sign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
k_sin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
k_softmax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
k_softplus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
k_softsign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
k_sparse_categorical_crossentropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
k_spatial_2d_padding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
k_spatial_3d_padding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
k_sqrt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
k_square . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
k_squeeze . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
k_stack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
k_std . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
k_stop_gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
k_sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
k_switch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
k_tanh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
k_temporal_padding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
k_tile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
k_to_dense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
k_transpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
k_truncated_normal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
k_update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
k_update_add . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
k_update_sub . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
k_var . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
k_variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
k_zeros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
k_zeros_like . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
layer_activation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
layer_activation_elu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
layer_activation_leaky_relu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
layer_activation_parametric_relu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
layer_activation_relu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
layer_activation_selu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
layer_activation_softmax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
layer_activation_thresholded_relu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
layer_activity_regularization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
layer_add . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
layer_alpha_dropout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
layer_average . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
layer_average_pooling_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
layer_batch_normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
layer_concatenate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
layer_conv_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
layer_conv_2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
layer_conv_2d_transpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
layer_conv_3d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
layer_conv_3d_transpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
layer_conv_lstm_2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
layer_cropping_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
layer_cropping_2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
layer_cropping_3d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
layer_cudnn_gru . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
layer_cudnn_lstm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
layer_dense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
layer_depthwise_conv_2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
layer_dot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
layer_dropout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
layer_embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
layer_flatten . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
layer_gaussian_dropout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
layer_gaussian_noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
layer_global_average_pooling_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
layer_global_max_pooling_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
layer_gru . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
layer_input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
layer_lambda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
layer_locally_connected_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
layer_locally_connected_2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
layer_lstm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
layer_masking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
layer_maximum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
layer_max_pooling_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
layer_minimum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
layer_multiply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
layer_permute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
layer_repeat_vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
layer_reshape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
layer_separable_conv_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
layer_separable_conv_2d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
layer_simple_rnn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
layer_spatial_dropout_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
layer_subtract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
layer_upsampling_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
layer_zero_padding_1d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
loss_mean_squared_error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
make_sampling_table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
metric_binary_accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
model_to_json . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
model_to_yaml . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
multi_gpu_model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
normalize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
optimizer_adadelta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
optimizer_adagrad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
optimizer_adam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
optimizer_adamax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
optimizer_nadam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
optimizer_rmsprop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
optimizer_sgd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
pad_sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
plot.keras_training_history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
pop_layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
predict.keras.engine.training.Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
predict_generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
predict_on_batch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
predict_proba . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
regularizer_l1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
reset_states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
save_model_hdf5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
save_model_weights_hdf5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
save_text_tokenizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
sequences_to_matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
serialize_model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
skipgrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
summary.keras.engine.training.Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
texts_to_matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
texts_to_sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
keras-package 9
texts_to_sequences_generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
text_hashing_trick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
text_one_hot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
text_tokenizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
text_to_word_sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
timeseries_generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
time_distributed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
to_categorical . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
train_on_batch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
use_implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
with_custom_object_scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
Index 310
keras-package R interface to Keras
Description
Keras is a high-level neural networks API, developed with a focus on enabling fast experimentation.
Keras has the following key features:
Details
• Allows the same code to run on CPU or on GPU, seamlessly.
• User-friendly API which makes it easy to quickly prototype deep learning models.
• Built-in support for convolutional networks (for computer vision), recurrent networks (for
sequence processing), and any combination of both.
• Supports arbitrary network architectures: multi-input or multi-output models, layer sharing,
model sharing, etc. This means that Keras is appropriate for building essentially any deep
learning model, from a memory network to a neural Turing machine.
• Is capable of running on top of multiple back-ends including TensorFlow, CNTK, or Theano.
See the package website at https://keras.rstudio.com for complete documentation.
Author(s)
Maintainer: Daniel Falbel <[email protected]> [contributor, copyright holder]
Authors:
• JJ Allaire [copyright holder]

• François Chollet [copyright holder]
Other contributors:
• RStudio [contributor, copyright holder, funder]

• Google [contributor, copyright holder, funder]
10 activation_relu
• Yuan Tang <[email protected]> (0000-0001-5243-233X) [contributor, copyright

holder]
• Wouter Van Der Bijl [contributor, copyright holder]
• Martin Studer [contributor, copyright holder]
• Sigrid Keydana [contributor]
See Also
Useful links:
• https://keras.rstudio.com
• Report bugs at https://github.com/rstudio/keras/issues
activation_relu Activation functions
Description
Activations functions can either be used through layer_activation(), or through the activation
argument supported by all forward layers.
Usage
activation_relu(x, alpha = 0, max_value = NULL, threshold = 0)
activation_elu(x, alpha = 1)
activation_selu(x)
activation_hard_sigmoid(x)
activation_linear(x)
activation_sigmoid(x)
activation_softmax(x, axis = -1)
activation_softplus(x)
activation_softsign(x)
activation_tanh(x)
activation_exponential(x)
application_densenet 11
Arguments
x Tensor
alpha Alpha value
max_value Max value
threshold Threshold value for thresholded activation.
axis Integer, axis along which the softmax normalization is applied
Details
• activation_selu() to be used together with the initialization "lecun_normal".
• activation_selu() to be used together with the dropout variant "AlphaDropout".
Value
Tensor with the same shape and dtype as x.
References
• activation_selu(): Self-Normalizing Neural Networks
application_densenet Instantiates the DenseNet architecture.
Description
Instantiates the DenseNet architecture.
Usage
application_densenet(blocks, include_top = TRUE, weights = "imagenet",
input_tensor = NULL, input_shape = NULL, pooling = NULL,
classes = 1000)
application_densenet121(include_top = TRUE, weights = "imagenet",

classes = 1000)

classes = 1000)

classes = 1000)
densenet_preprocess_input(x, data_format = NULL)

12 application_inception_resnet_v2
Arguments
blocks numbers of building blocks for the four dense layers.
include_top whether to include the fully-connected layer at the top of the network.
weights one of NULL (random initialization), ’imagenet’ (pre-training on ImageNet), or
the path to the weights file to be loaded.
input_tensor optional Keras tensor (i.e. output of layer_input()) to use as image input for
the model.
input_shape optional shape list, only to be specified if include_top is FALSE (otherwise
the input shape has to be (224, 224, 3) (with channels_last data format) or
(3, 224, 224) (with channels_first data format). It should have exactly 3
inputs channels.
pooling optional pooling mode for feature extraction when include_top is FALSE. -
NULL means that the output of the model will be the 4D tensor output of the last
convolutional layer. - avg means that global average pooling will be applied to
the output of the last convolutional layer, and thus the output of the model will
be a 2D tensor. - max means that global max pooling will be applied.
classes optional number of classes to classify images into, only to be specified if include_top
is TRUE, and if no weights argument is specified.
x a 3D or 4D array consists of RGB values within [0, 255].
data_format data format of the image tensor.
Details
Optionally loads weights pre-trained on ImageNet. Note that when using TensorFlow, for best per-
formance you should set image_data_format='channels_last' in your Keras config at ~/.keras/keras.json.
The model and the weights are compatible with TensorFlow, Theano, and CNTK. The data format
convention used by the model is the one specified in your Keras config file.
application_inception_resnet_v2
Inception-ResNet v2 model, with weights trained on ImageNet
Description
Inception-ResNet v2 model, with weights trained on ImageNet
Usage
application_inception_resnet_v2(include_top = TRUE,
weights = "imagenet", input_tensor = NULL, input_shape = NULL,
pooling = NULL, classes = 1000)
inception_resnet_v2_preprocess_input(x)
application_inception_v3 13
Arguments
weights NULL (random initialization), imagenet (ImageNet weights), or the path to the
weights file to be loaded.
input_tensor optional Keras tensor to use as image input for the model.
input_shape optional shape list, only to be specified if include_top is FALSE (otherwise the
input shape has to be (299, 299, 3). It should have exactly 3 inputs channels,
and width and height should be no smaller than 75. E.g. (150, 150, 3) would
be one valid value.
pooling Optional pooling mode for feature extraction when include_top is FALSE.
• NULL means that the output of the model will be the 4D tensor output of the
last convolutional layer.
• avg means that global average pooling will be applied to the output of the
last convolutional layer, and thus the output of the model will be a 2D ten-
sor.
• max means that global max pooling will be applied.
x Input tensor for preprocessing
Details
Do note that the input image format for this model is different than for the VGG16 and ResNet
models (299x299 instead of 224x224).
The inception_resnet_v2_preprocess_input() function should be used for image preprocess-
ing.
Value
A Keras model instance.
Reference
• Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning(http://arxiv.org/abs/1512.00567)
application_inception_v3
Inception V3 model, with weights pre-trained on ImageNet.
Description
Inception V3 model, with weights pre-trained on ImageNet.
14 application_inception_v3
Usage
application_inception_v3(include_top = TRUE, weights = "imagenet",

classes = 1000)
inception_v3_preprocess_input(x)
Arguments
be one valid value.
sor.
Details
The inception_v3_preprocess_input() function should be used for image preprocessing.
Value
Reference
• Rethinking the Inception Architecture for Computer Vision

application_mobilenet 15
application_mobilenet MobileNet model architecture.
Description
MobileNet model architecture.
Usage
application_mobilenet(input_shape = NULL, alpha = 1,
depth_multiplier = 1, dropout = 0.001, include_top = TRUE,
weights = "imagenet", input_tensor = NULL, pooling = NULL,
classes = 1000)
mobilenet_preprocess_input(x)
mobilenet_decode_predictions(preds, top = 5)
mobilenet_load_model_hdf5(filepath)
Arguments
the input shape has to be (224, 224, 3) (with channels_last data format)
or (3, 224, 224) (with channels_first data format). It should have exactly
3 inputs channels, and width and height should be no smaller than 32. E.g.
(200, 200, 3) would be one valid value.
alpha controls the width of the network.
• If alpha < 1.0, proportionally decreases the number of filters in each layer.
• If alpha > 1.0, proportionally increases the number of filters in each layer.
• If alpha = 1, default number of filters from the paper are used at each layer.
depth_multiplier
depth multiplier for depthwise convolution (also called the resolution multiplier)
dropout dropout rate
the model.
pooling Optional pooling mode for feature extraction when include_top is FALSE. -
16 application_mobilenet_v2
x input tensor, 4D
preds Tensor encoding a batch of predictions.
top integer, how many top-guesses to return.
filepath File path
Details
The mobilenet_preprocess_input() function should be used for image preprocessing. To load
a saved instance of a MobileNet model use the mobilenet_load_model_hdf5() function. To pre-
pare image input for MobileNet use mobilenet_preprocess_input(). To decode predictions use
mobilenet_decode_predictions().
Value
application_mobilenet() and mobilenet_load_model_hdf5() return a Keras model instance.
mobilenet_preprocess_input() returns image input suitable for feeding into a mobilenet model.
mobilenet_decode_predictions() returns a list of data frames with variables class_name, class_description,
and score (one data frame per sample in batch input).
Reference
• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications.
application_mobilenet_v2
MobileNetV2 model architecture
Description
MobileNetV2 model architecture
Usage
application_mobilenet_v2(input_shape = NULL, alpha = 1,
include_top = TRUE, weights = "imagenet", input_tensor = NULL,
pooling = NULL, classes = 1000)
mobilenet_v2_preprocess_input(x)
mobilenet_v2_decode_predictions(preds, top = 5)
mobilenet_v2_load_model_hdf5(filepath)
application_mobilenet_v2 17
Arguments
the input shape has to be (224, 224, 3) (with channels_last data format)
or (3, 224, 224) (with channels_first data format). It should have exactly
3 inputs channels, and width and height should be no smaller than 32. E.g.
(200, 200, 3) would be one valid value.
alpha controls the width of the network.
• If alpha < 1.0, proportionally decreases the number of filters in each layer.
• If alpha > 1.0, proportionally increases the number of filters in each layer.
• If alpha = 1, default number of filters from the paper are used at each layer.
the model.
x input tensor, 4D
filepath File path
Value
application_mobilenet_v2() and mobilenet_v2_load_model_hdf5() return a Keras model
instance. mobilenet_v2_preprocess_input() returns image input suitable for feeding into a
mobilenet v2 model. mobilenet_v2_decode_predictions() returns a list of data frames with
variables class_name, class_description, and score (one data frame per sample in batch in-
put).
Reference
• MobileNetV2: Inverted Residuals and Linear Bottlenecks
See Also
application_mobilenet
18 application_nasnet
application_nasnet Instantiates a NASNet model.
Description
Note that only TensorFlow is supported for now, therefore it only works with the data format
image_data_format='channels_last' in your Keras config at ~/.keras/keras.json.
Usage
application_nasnet(input_shape = NULL, penultimate_filters = 4032L,
num_blocks = 6L, stem_block_filters = 96L, skip_reduction = TRUE,
filter_multiplier = 2L, include_top = TRUE, weights = NULL,
input_tensor = NULL, pooling = NULL, classes = 1000,
default_size = NULL)
application_nasnetlarge(input_shape = NULL, include_top = TRUE,

weights = NULL, input_tensor = NULL, pooling = NULL,
classes = 1000)
application_nasnetmobile(input_shape = NULL, include_top = TRUE,

weights = NULL, input_tensor = NULL, pooling = NULL,
classes = 1000)
nasnet_preprocess_input(x)
Arguments
input_shape Optional shape list, the input shape is by default (331, 331, 3) for NASNet-
Large and (224, 224, 3) for NASNetMobile It should have exactly 3 inputs
channels, and width and height should be no smaller than 32. E.g. (224, 224, 3)
would be one valid value.
penultimate_filters
Number of filters in the penultimate layer. NASNet models use the notation
NASNet (N @ P), where: - N is the number of blocks - P is the number of
penultimate filters
num_blocks Number of repeated blocks of the NASNet model. NASNet models use the
notation NASNet (N @ P), where: - N is the number of blocks - P is the number
of penultimate filters
stem_block_filters
Number of filters in the initial stem block
skip_reduction Whether to skip the reduction step at the tail end of the network. Set to FALSE
for CIFAR models.
filter_multiplier
Controls the width of the network.
application_resnet50 19
• If filter_multiplier < 1.0, proportionally decreases the number of filters

in each layer.
• If filter_multiplier > 1.0, proportionally increases the number of filters
in each layer. - If filter_multiplier = 1, default number of filters from
the paper are used at each layer.
include_top Whether to include the fully-connected layer at the top of the network.
weights NULL (random initialization) or imagenet (ImageNet weights)
input_tensor Optional Keras tensor (i.e. output of layer_input()) to use as image input for
the model.
classes Optional number of classes to classify images into, only to be specified if include_top
default_size Specifies the default image size of the model
x a 4D array consists of RGB values within [0, 255].
application_resnet50 ResNet50 model for Keras.
Description
ResNet50 model for Keras.
Usage
application_resnet50(include_top = TRUE, weights = "imagenet",
classes = 1000)
Arguments
be one valid value.
20 application_resnet50
sor.
Details
Optionally loads weights pre-trained on ImageNet.

The imagenet_preprocess_input() function should be used for image preprocessing.
Value
Reference
- Deep Residual Learning for Image Recognition
Examples
## Not run:
library(keras)
# instantiate the model

model <- application_resnet50(weights = 'imagenet')
# load the image

img_path <- "elephant.jpg"
img <- image_load(img_path, target_size = c(224,224))
x <- image_to_array(img)
# ensure we have a 4d tensor with single element in the batch dimension,

# the preprocess the input for prediction using resnet50
x <- array_reshape(x, c(1, dim(x)))
x <- imagenet_preprocess_input(x)
# make predictions then decode and print them

preds <- model %>% predict(x)
imagenet_decode_predictions(preds, top = 3)[[1]]
## End(Not run)
application_vgg 21
application_vgg VGG16 and VGG19 models for Keras.
Description
VGG16 and VGG19 models for Keras.
Usage
application_vgg16(include_top = TRUE, weights = "imagenet",
classes = 1000)
application_vgg19(include_top = TRUE, weights = "imagenet",

classes = 1000)
Arguments
include_top whether to include the 3 fully-connected layers at the top of the network.
input shape has to be (224, 224, 3) It should have exactly 3 inputs channels,
be one valid value.
sor.
Details
Optionally loads weights pre-trained on ImageNet.
The imagenet_preprocess_input() function should be used for image preprocessing.
Value
Keras model instance.
22 application_xception
Reference
- Very Deep Convolutional Networks for Large-Scale Image Recognition
Examples
## Not run:
library(keras)
model <- application_vgg16(weights = 'imagenet', include_top = FALSE)
img_path <- "elephant.jpg"

img <- image_load(img_path, target_size = c(224,224))
x <- image_to_array(img)
x <- array_reshape(x, c(1, dim(x)))
x <- imagenet_preprocess_input(x)
features <- model %>% predict(x)
## End(Not run)
application_xception Xception V1 model for Keras.
Description
Xception V1 model for Keras.
Usage
application_xception(include_top = TRUE, weights = "imagenet",
classes = 1000)
xception_preprocess_input(x)
Arguments
be one valid value.
backend 23
sor.
Details
On ImageNet, this model gets to a top-1 validation accuracy of 0.790 and a top-5 validation accuracy
of 0.945.
The xception_preprocess_input() function should be used for image preprocessing.
This application is only available when using the TensorFlow back-end.
Value
Reference
• Xception: Deep Learning with Depthwise Separable Convolutions
backend Keras backend tensor engine
Description
Obtain a reference to the keras.backend Python module used to implement tensor operations.
Usage
backend(convert = TRUE)
Arguments
convert TRUE to automatically convert Python objects to their R equivalent. If you pass
FALSE you can do manual conversion using the py_to_r() function.
Value
Reference to Keras backend python module.
24 bidirectional
Note
See the documentation here https://keras.io/backend/ for additional details on the available
functions.
bidirectional Bidirectional wrapper for RNNs.
Description
Bidirectional wrapper for RNNs.
Usage
bidirectional(object, layer, merge_mode = "concat", input_shape = NULL,
batch_input_shape = NULL, batch_size = NULL, dtype = NULL,
name = NULL, trainable = NULL, weights = NULL)
Arguments
object Model or layer object
layer Recurrent instance.
merge_mode Mode by which outputs of the forward and backward RNNs will be combined.
One of ’sum’, ’mul’, ’concat’, ’ave’, NULL. If NULL, the outputs will not be
combined, they will be returned as a list.
input_shape Dimensionality of the input (integer) not including the samples axis. This argu-
ment is required when using this layer as the first layer in a model.
batch_input_shape
Shapes, including the batch size. For instance, batch_input_shape=c(10, 32)
indicates that the expected input will be batches of 10 32-dimensional vectors.
batch_input_shape=list(NULL, 32) indicates batches of an arbitrary num-
ber of 32-dimensional vectors.
batch_size Fixed batch size for layer
dtype The data type expected by the input, as a string (float32, float64, int32...)
name An optional name string for the layer. Should be unique in a model (do not reuse
the same name twice). It will be autogenerated if it isn’t provided.
trainable Whether the layer weights will be updated during training.
weights Initial weights for layer.
See Also
Other layer wrappers: time_distributed
callback_csv_logger 25
callback_csv_logger Callback that streams epoch results to a csv file
Description
Supports all values that can be represented as a string
Usage
callback_csv_logger(filename, separator = ",", append = FALSE)
Arguments
filename filename of the csv file, e.g. ’run/log.csv’.

separator string used to separate elements in the csv file.
append TRUE: append if file exists (useful for continuing training). FALSE: overwrite
existing file,
See Also
Other callbacks: callback_early_stopping, callback_lambda, callback_learning_rate_scheduler,

callback_model_checkpoint, callback_progbar_logger, callback_reduce_lr_on_plateau,
callback_remote_monitor, callback_tensorboard, callback_terminate_on_naan
callback_early_stopping
Stop training when a monitored quantity has stopped improving.
Description
Stop training when a monitored quantity has stopped improving.
Usage
callback_early_stopping(monitor = "val_loss", min_delta = 0,

patience = 0, verbose = 0, mode = c("auto", "min", "max"),
baseline = NULL, restore_best_weights = FALSE)
26 callback_lambda
Arguments
monitor quantity to be monitored.

min_delta minimum change in the monitored quantity to qualify as an improvement, i.e.
an absolute change of less than min_delta, will count as no improvement.
patience number of epochs with no improvement after which training will be stopped.
verbose verbosity mode, 0 or 1.
mode one of "auto", "min", "max". In min mode, training will stop when the quantity
monitored has stopped decreasing; in max mode it will stop when the quantity
monitored has stopped increasing; in auto mode, the direction is automatically
inferred from the name of the monitored quantity.
baseline Baseline value for the monitored quantity to reach. Training will stop if the
model doesn’t show improvement over the baseline.
restore_best_weights
Whether to restore model weights from the epoch with the best value of the
monitored quantity. If FALSE, the model weights obtained at the last step of
training are used.
See Also
Other callbacks: callback_csv_logger, callback_lambda, callback_learning_rate_scheduler,

callback_lambda Create a custom callback
Description
This callback is constructed with anonymous functions that will be called at the appropriate time.
Note that the callbacks expects positional arguments, as:
• on_epoch_begin and on_epoch_end expect two positional arguments: epoch, logs

• on_batch_begin and on_batch_end expect two positional arguments: batch, logs
• on_train_begin and on_train_end expect one positional argument: logs
Usage
callback_lambda(on_epoch_begin = NULL, on_epoch_end = NULL,

on_batch_begin = NULL, on_batch_end = NULL, on_train_begin = NULL,
on_train_end = NULL)
callback_learning_rate_scheduler 27
Arguments
on_epoch_begin called at the beginning of every epoch.

on_epoch_end called at the end of every epoch.
on_batch_begin called at the beginning of every batch.
on_batch_end called at the end of every batch.
on_train_begin called at the beginning of model training.
on_train_end called at the end of model training.
See Also
Other callbacks: callback_csv_logger, callback_early_stopping, callback_learning_rate_scheduler,

callback_learning_rate_scheduler
Learning rate scheduler.
Description
Learning rate scheduler.
Usage
callback_learning_rate_scheduler(schedule)
Arguments
schedule a function that takes an epoch index as input (integer, indexed from 0) and cur-
rent learning rate and returns a new learning rate as output (float).
See Also
Other callbacks: callback_csv_logger, callback_early_stopping, callback_lambda, callback_model_checkpoint,

callback_progbar_logger, callback_reduce_lr_on_plateau, callback_remote_monitor, callback_tensorboard,
callback_terminate_on_naan
28 callback_model_checkpoint
callback_model_checkpoint
Save the model after every epoch.
Description
filepath can contain named formatting options, which will be filled the value of epoch and keys in
logs (passed in on_epoch_end). For example: if filepath is weights.{epoch:02d}-{val_loss:.2f}.hdf5,
then the model checkpoints will be saved with the epoch number and the validation loss in the file-
name.
Usage
callback_model_checkpoint(filepath, monitor = "val_loss", verbose = 0,
save_best_only = FALSE, save_weights_only = FALSE, mode = c("auto",
"min", "max"), period = 1)
Arguments
filepath string, path to save the model file.
monitor quantity to monitor.
save_best_only if save_best_only=TRUE, the latest best model according to the quantity mon-
itored will not be overwritten.
save_weights_only
if TRUE, then only the model’s weights will be saved (save_model_weights_hdf5(filepath)),
else the full model is saved (save_model_hdf5(filepath)).
mode one of "auto", "min", "max". If save_best_only=TRUE, the decision to over-
write the current save file is made based on either the maximization or the mini-
mization of the monitored quantity. For val_acc, this should be max, for val_loss
this should be min, etc. In auto mode, the direction is automatically inferred
from the name of the monitored quantity.
period Interval (number of epochs) between checkpoints.
For example
if filepath is weights.{epoch:02d}-{val_loss:.2f}.hdf5,: then the model checkpoints will
be saved with the epoch number and the validation loss in the filename.
See Also
Other callbacks: callback_csv_logger, callback_early_stopping, callback_lambda, callback_learning_rate_sch
callback_progbar_logger, callback_reduce_lr_on_plateau, callback_remote_monitor, callback_tensorboard,
callback_progbar_logger 29
callback_progbar_logger
Callback that prints metrics to stdout.
Description
Callback that prints metrics to stdout.
Usage
callback_progbar_logger(count_mode = "samples",
stateful_metrics = NULL)
Arguments
count_mode One of "steps" or "samples". Whether the progress bar should count samples
seens or steps (batches) seen.
stateful_metrics
List of metric names that should not be averaged onver an epoch. Metrics in
this list will be logged as-is in on_epoch_end. All others will be averaged in
on_epoch_end.
See Also
callback_model_checkpoint, callback_reduce_lr_on_plateau, callback_remote_monitor,
callback_tensorboard, callback_terminate_on_naan
callback_reduce_lr_on_plateau
Reduce learning rate when a metric has stopped improving.
Description
Models often benefit from reducing the learning rate by a factor of 2-10 once learning stagnates.
This callback monitors a quantity and if no improvement is seen for a ’patience’ number of epochs,
the learning rate is reduced.
Usage
callback_reduce_lr_on_plateau(monitor = "val_loss", factor = 0.1,
patience = 10, verbose = 0, mode = c("auto", "min", "max"),
min_delta = 1e-04, cooldown = 0, min_lr = 0)
30 callback_remote_monitor
Arguments
monitor quantity to be monitored.
factor factor by which the learning rate will be reduced. new_lr = lr
• factor
patience number of epochs with no improvement after which learning rate will be re-
duced.
verbose int. 0: quiet, 1: update messages.
mode one of "auto", "min", "max". In min mode, lr will be reduced when the quan-
tity monitored has stopped decreasing; in max mode it will be reduced when
the quantity monitored has stopped increasing; in auto mode, the direction is
automatically inferred from the name of the monitored quantity.
min_delta threshold for measuring the new optimum, to only focus on significant changes.
cooldown number of epochs to wait before resuming normal operation after lr has been
reduced.
min_lr lower bound on the learning rate.
See Also
callback_model_checkpoint, callback_progbar_logger, callback_remote_monitor, callback_tensorboard,
callback_remote_monitor
Callback used to stream events to a server.
Description
Callback used to stream events to a server.
Usage
callback_remote_monitor(root = "http://localhost:9000",
path = "/publish/epoch/end/", field = "data", headers = NULL,
send_as_json = FALSE)
Arguments
root root url of the target server.
path path relative to root to which the events will be sent.
field JSON field under which the data will be stored.
headers Optional named list of custom HTTP headers. Defaults to: list(Accept = "application/json", Cont
Type = "application/json")
send_as_json Whether the request should be sent as application/json.
callback_tensorboard 31
Details
Events are sent to root + '/publish/epoch/end/' by default. Calls are HTTP POST, with a
data argument which is a JSON-encoded dictionary of event data. If send_as_json is set to True,
the content type of the request will be application/json. Otherwise the serialized JSON will be send
within a form
See Also
callback_tensorboard, callback_terminate_on_naan
callback_tensorboard TensorBoard basic visualizations
Description
This callback writes a log for TensorBoard, which allows you to visualize dynamic graphs of your
training and test metrics, as well as activation histograms for the different layers in your model.
Usage
callback_tensorboard(log_dir = NULL, histogram_freq = 0,
batch_size = 32, write_graph = TRUE, write_grads = FALSE,
write_images = FALSE, embeddings_freq = 0,
embeddings_layer_names = NULL, embeddings_metadata = NULL,
embeddings_data = NULL, update_freq = "epoch")
Arguments
log_dir The path of the directory where to save the log files to be parsed by Tensorboard.
The default is NULL, which will use the active run directory (if available) and
otherwise will use "logs".
histogram_freq frequency (in epochs) at which to compute activation histograms for the layers
of the model. If set to 0, histograms won’t be computed.
batch_size size of batch of inputs to feed to the network for histograms computation.
write_graph whether to visualize the graph in Tensorboard. The log file can become quite
large when write_graph is set to TRUE
write_grads whether to visualize gradient histograms in TensorBoard. histogram_freq
must be greater than 0.
write_images whether to write model weights to visualize as image in Tensorboard.
embeddings_freq
frequency (in epochs) at which selected embedding layers will be saved.
embeddings_layer_names
a list of names of layers to keep eye on. If NULL or empty list all the embedding
layers will be watched.
32 callback_terminate_on_naan
embeddings_metadata
a named list which maps layer name to a file name in which metadata for this
embedding layer is saved. See the details about the metadata file format. In case
if the same metadata file is used for all embedding layers, string can be passed.
embeddings_data
Data to be embedded at layers specified in embeddings_layer_names. Array (if
the model has a single input) or list of arrays (if the model has multiple inputs).
Learn more about embeddings
update_freq 'batch' or 'epoch' or integer. When using 'batch', writes the losses and met-
rics to TensorBoard after each batch. The same applies for 'epoch'. If using
an integer, let’s say 10000, the callback will write the metrics and losses to Ten-
sorBoard every 10000 samples. Note that writing too frequently to TensorBoard
can slow down your training.
Details
TensorBoard is a visualization tool provided with TensorFlow.
You can find more information about TensorBoard here.
When using a backend other than TensorFlow, TensorBoard will still work (if you have TensorFlow
installed), but the only feature available will be the display of the losses and metrics plots.
See Also
callback_remote_monitor, callback_terminate_on_naan
Callback that terminates training when a NaN loss is encountered.
Description
Callback that terminates training when a NaN loss is encountered.
Usage
callback_terminate_on_naan()
See Also
callback_remote_monitor, callback_tensorboard
clone_model 33
clone_model Clone a model instance.
Description
Model cloning is similar to calling a model on new inputs, except that it creates new layers (and
thus new weights) instead of sharing the weights of the existing layers.
Usage
clone_model(model, input_tensors = NULL)
Arguments
model Instance of Keras model (could be a functional model or a Sequential model).
input_tensors Optional list of input tensors to build the model upon. If not provided, place-
holders will be created.
compile.keras.engine.training.Model
Configure a Keras model for training
Description
Configure a Keras model for training
Usage
## S3 method for class 'keras.engine.training.Model'
compile(object, optimizer, loss,
metrics = NULL, loss_weights = NULL, sample_weight_mode = NULL,
weighted_metrics = NULL, target_tensors = NULL, ...)
Arguments
object Model object to compile.
optimizer Name of optimizer or optimizer instance.
loss Name of objective function or objective function. If the model has multiple
outputs, you can use a different loss on each output by passing a dictionary or a
list of objectives. The loss value that will be minimized by the model will then
be the sum of all individual losses.
metrics List of metrics to be evaluated by the model during training and testing. Typi-
cally you will use metrics='accuracy'. To specify different metrics for dif-
ferent outputs of a multi-output model, you could also pass a named list such as
metrics=list(output_a = 'accuracy').
34 constraints
loss_weights Optional list specifying scalar coefficients to weight the loss contributions of
different model outputs. The loss value that will be minimized by the model will
then be the weighted sum of all indvidual losses, weighted by the loss_weights
coefficients.
sample_weight_mode
If you need to do timestep-wise sample weighting (2D weights), set this to "tem-
poral". NULL defaults to sample-wise weights (1D). If the model has multiple
outputs, you can use a different sample_weight_mode on each output by pass-
ing a list of modes.
weighted_metrics
List of metrics to be evaluated and weighted by sample_weight or class_weight
during training and testing
target_tensors By default, Keras will create a placeholder for the model’s target, which will be
fed with the target data during training. If instead you would like to use your
own target tensor (in turn, Keras will not expect external data for these targets
at training time), you can specify them via the target_tensors argument. It
should be a single tensor (for a single-output sequential model),
... When using the Theano/CNTK backends, these arguments are passed into K.function.
When using the TensorFlow backend, these arguments are passed into tf$Session()$run.
See Also
Other model functions: evaluate.keras.engine.training.Model, evaluate_generator, fit.keras.engine.training
fit_generator, get_config, get_layer, keras_model_sequential, keras_model, multi_gpu_model,
pop_layer, predict.keras.engine.training.Model, predict_generator, predict_on_batch,
predict_proba, summary.keras.engine.training.Model, train_on_batch
constraints Weight constraints
Description
Functions that impose constraints on weight values.
Usage
constraint_maxnorm(max_value = 2, axis = 0)
constraint_nonneg()
constraint_unitnorm(axis = 0)
constraint_minmaxnorm(min_value = 0, max_value = 1, rate = 1,

axis = 0)
constraints 35
Arguments
max_value The maximum norm for the incoming weights.
axis The axis along which to calculate weight norms. For instance, in a dense layer
the weight matrix has shape input_dim, output_dim, set axis to 0 to con-
strain each weight vector of length input_dim,. In a convolution 2D layer with
dim_ordering="tf", the weight tensor has shape rows, cols, input_depth, output_depth,
set axis to c(0, 1, 2) to constrain the weights of each filter tensor of size
rows, cols, input_depth.
min_value The minimum norm for the incoming weights.
rate The rate for enforcing the constraint: weights will be rescaled to yield (1 - rate) *
norm + rate * norm.clip(low, high). Effectively, this means that rate=1.0 stands
for strict enforcement of the constraint, while rate<1.0 means that weights will
be rescaled at each step to slowly move towards a value inside the desired inter-
val.
Details
• constraint_maxnorm() constrains the weights incident to each hidden unit to have a norm
less than or equal to a desired value.
• constraint_nonneg() constraints the weights to be non-negative
• constraint_unitnorm() constrains the weights incident to each hidden unit to have unit
norm.
• constraint_minmaxnorm() constrains the weights incident to each hidden unit to have the
norm between a lower bound and an upper bound.
Custom constraints
You can implement your own constraint functions in R. A custom constraint is an R function that
takes weights (w) as input and returns modified weights. Note that keras backend() tensor func-
tions (e.g. k_greater_equal()) should be used in the implementation of custom constraints. For
example:
nonneg_constraint <- function(w) {

w * k_cast(k_greater_equal(w, 0), k_floatx())
}
layer_dense(units = 32, input_shape = c(784),

kernel_constraint = nonneg_constraint)
Note that models which use custom constraints cannot be serialized using save_model_hdf5().
Rather, the weights of the model should be saved and restored using save_model_weights_hdf5().
See Also
Dropout: A Simple Way to Prevent Neural Networks from Overfitting Srivastava, Hinton, et al.
2014
KerasConstraint
36 create_layer
count_params Count the total number of scalars composing the weights.
Description
Count the total number of scalars composing the weights.
Usage
count_params(object)
Arguments
object Layer or model object
Value
An integer count
See Also
Other layer methods: get_config, get_input_at, get_weights, reset_states
create_layer Create a Keras Layer
Description
Create a Keras Layer
Usage
create_layer(layer_class, object, args = list())
Arguments
layer_class Python layer class or R6 class of type KerasLayer
object Object to compose layer with. This is either a keras_model_sequential() to
add the layer to, or another Layer which this layer will call.
args List of arguments to layer constructor function
Value
A Keras layer
create_wrapper 37
Note
The object parameter can be missing, in which case the layer is created without a connection to an
existing graph.
create_wrapper Create a Keras Wrapper
Description
Create a Keras Wrapper
Usage
create_wrapper(wrapper_class, object, args = list())
Arguments
wrapper_class R6 class of type KerasWrapper
object Object to compose layer with. This is either a keras_model_sequential() to
add the layer to, or another Layer which this layer will call.
args List of arguments to layer constructor function
Value
A Keras wrapper
Note
The object parameter can be missing, in which case the layer is created without a connection to an
existing graph.
dataset_boston_housing
Boston housing price regression dataset
Description
Dataset taken from the StatLib library which is maintained at Carnegie Mellon University.
Usage
dataset_boston_housing(path = "boston_housing.npz", test_split = 0.2,
seed = 113L)
38 dataset_cifar10
Arguments
path Path where to cache the dataset locally (relative to ~/.keras/datasets).

test_split fraction of the data to reserve as test set.
seed Random seed for shuffling the data before computing the test split.
Value
Lists of training and test data: train$x, train$y, test$x, test$y.

Samples contain 13 attributes of houses at different locations around the Boston suburbs in the late
1970s. Targets are the median values of the houses at a location (in k$).
See Also
Other datasets: dataset_cifar100, dataset_cifar10, dataset_fashion_mnist, dataset_imdb,

dataset_mnist, dataset_reuters
dataset_cifar10 CIFAR10 small image classification
Description
Dataset of 50,000 32x32 color training images, labeled over 10 categories, and 10,000 test images.
Usage
dataset_cifar10()
Value

The x data is an array of RGB image data with shape (num_samples, 3, 32, 32).
The y data is an array of category labels (integers in range 0-9) with shape (num_samples).
See Also
Other datasets: dataset_boston_housing, dataset_cifar100, dataset_fashion_mnist, dataset_imdb,

dataset_cifar100 39
dataset_cifar100 CIFAR100 small image classification
Description
Dataset of 50,000 32x32 color training images, labeled over 100 categories, and 10,000 test images.
Usage
dataset_cifar100(label_mode = c("fine", "coarse"))
Arguments
label_mode one of "fine", "coarse".
Value

The x data is an array of RGB image data with shape (num_samples, 3, 32, 32).
The y data is an array of category labels with shape (num_samples).
See Also
Other datasets: dataset_boston_housing, dataset_cifar10, dataset_fashion_mnist, dataset_imdb,

dataset_fashion_mnist Fashion-MNIST database of fashion articles
Description
Dataset of 60,000 28x28 grayscale images of the 10 fashion article classes, along with a test set
of 10,000 images. This dataset can be used as a drop-in replacement for MNIST. The class labels
are encoded as integers from 0-9 which correspond to T-shirt/top, Trouser, Pullover, Dress, Coat,
Sandal, Shirt,
Usage
dataset_fashion_mnist()
40 dataset_imdb
Details
Dataset of 60,000 28x28 grayscale images of 10 fashion categories, along with a test set of 10,000
images. This dataset can be used as a drop-in replacement for MNIST. The class labels are:
• 0 - T-shirt/top
• 1 - Trouser
• 2 - Pullover
• 3 - Dress
• 4 - Coat
• 5 - Sandal
• 6 - Shirt
• 7 - Sneaker
• 8 - Bag
• 9 - Ankle boot
Value
Lists of training and test data: train$x, train$y, test$x, test$y, where x is an array of
grayscale image data with shape (num_samples, 28, 28) and y is an array of article labels (integers
in range 0-9) with shape (num_samples).
See Also
Other datasets: dataset_boston_housing, dataset_cifar100, dataset_cifar10, dataset_imdb,
dataset_imdb IMDB Movie reviews sentiment classification
Description
Dataset of 25,000 movies reviews from IMDB, labeled by sentiment (positive/negative). Reviews
have been preprocessed, and each review is encoded as a sequence of word indexes (integers). For
convenience, words are indexed by overall frequency in the dataset, so that for instance the integer
"3" encodes the 3rd most frequent word in the data. This allows for quick filtering operations such
as: "only consider the top 10,000 most common words, but eliminate the top 20 most common
words".
Usage
dataset_imdb(path = "imdb.npz", num_words = NULL, skip_top = 0L,
maxlen = NULL, seed = 113L, start_char = 1L, oov_char = 2L,
index_from = 3L)
dataset_imdb_word_index(path = "imdb_word_index.json")
dataset_mnist 41
Arguments
path Where to cache the data (relative to ~/.keras/dataset).
num_words Max number of words to include. Words are ranked by how often they occur (in
the training set) and only the most frequent words are kept
skip_top Skip the top N most frequently occuring words (which may not be informative).
maxlen sequences longer than this will be filtered out.
seed random seed for sample shuffling.
start_char The start of a sequence will be marked with this character. Set to 1 because 0 is
usually the padding character.
oov_char Words that were cut out because of the num_words or skip_top limit will be
replaced with this character.
index_from Index actual words with this index and higher.
Details
As a convention, "0" does not stand for a specific word, but instead is used to encode any unknown
word.
Value
The x data includes integer sequences. If the num_words argument was specific, the maximum pos-
sible index value is num_words-1. If the maxlen`` argument was specified, the largest possible sequence length
The y data includes a set of integer labels (0 or 1).
The dataset_imdb_word_index() function returns a list where the names are words and the values
are integer.
See Also
Other datasets: dataset_boston_housing, dataset_cifar100, dataset_cifar10, dataset_fashion_mnist,
dataset_mnist MNIST database of handwritten digits
Description
Dataset of 60,000 28x28 grayscale images of the 10 digits, along with a test set of 10,000 images.
Usage
dataset_mnist(path = "mnist.npz")
42 dataset_reuters
Arguments
path Path where to cache the dataset locally (relative to ~/.keras/datasets).
Value
Lists of training and test data: train$x, train$y, test$x, test$y, where x is an array of
grayscale image data with shape (num_samples, 28, 28) and y is an array of digit labels (integers in
range 0-9) with shape (num_samples).
See Also
dataset_imdb, dataset_reuters
dataset_reuters Reuters newswire topics classification
Description
Dataset of 11,228 newswires from Reuters, labeled over 46 topics. As with dataset_imdb() , each
wire is encoded as a sequence of word indexes (same conventions).
Usage
dataset_reuters(path = "reuters.npz", num_words = NULL,
skip_top = 0L, maxlen = NULL, test_split = 0.2, seed = 113L,
start_char = 1L, oov_char = 2L, index_from = 3L)
dataset_reuters_word_index(path = "reuters_word_index.pkl")
Arguments
path Where to cache the data (relative to ~/.keras/dataset).
num_words Max number of words to include. Words are ranked by how often they occur (in
the training set) and only the most frequent words are kept
skip_top Skip the top N most frequently occuring words (which may not be informative).
maxlen Truncate sequences after this length.
test_split Fraction of the dataset to be used as test data.
seed Random seed for sample shuffling.
start_char The start of a sequence will be marked with this character. Set to 1 because 0 is
usually the padding character.
oov_char words that were cut out because of the num_words or skip_top limit will be
replaced with this character.
index_from index actual words with this index and higher.
evaluate.keras.engine.training.Model 43
Value
Lists of training and test data: train$x, train$y, test$x, test$y with same format as
dataset_imdb(). The dataset_reuters_word_index() function returns a list where the names
are words and the values are integer. e.g. word_index[["giraffe"]] might return 1234.
See Also
dataset_imdb, dataset_mnist
evaluate.keras.engine.training.Model
Evaluate a Keras model
Description
Evaluate a Keras model
Usage
evaluate(object, x = NULL,
y = NULL, batch_size = NULL, verbose = 1, sample_weight = NULL,
steps = NULL, ...)
Arguments
object Model object to evaluate
x Vector, matrix, or array of test data (or list if the model has multiple inputs). If
all inputs in the model are named, you can also pass a list mapping input names
to data. x can be NULL (default) if feeding from framework-native tensors (e.g.
TensorFlow data tensors).
y Vector, matrix, or array of target (label) data (or list if the model has multiple
outputs). If all outputs in the model are named, you can also pass a list mapping
output names to data. y can be NULL (default) if feeding from framework-native
tensors (e.g. TensorFlow data tensors).
batch_size Integer or NULL. Number of samples per gradient update. If unspecified, batch_size
will default to 32.
verbose Verbosity mode (0 = silent, 1 = progress bar, 2 = one line per epoch).
sample_weight Optional array of the same length as x, containing weights to apply to the
model’s loss for each sample. In the case of temporal data, you can pass a
2D array with shape (samples, sequence_length), to apply a different weight to
every timestep of every sample. In this case you should make sure to specify
sample_weight_mode="temporal" in compile().
steps Total number of steps (batches of samples) before declaring the evaluation round
finished. Ignored with the default value of NULL.
... Unused
44 evaluate_generator
Value
Named list of model test loss (or losses for models with multiple outputs) and model metrics.
See Also
Other model functions: compile.keras.engine.training.Model, evaluate_generator, fit.keras.engine.training.
fit_generator, get_config, get_layer, keras_model_sequential, keras_model, multi_gpu_model,
pop_layer, predict.keras.engine.training.Model, predict_generator, predict_on_batch,
predict_proba, summary.keras.engine.training.Model, train_on_batch
evaluate_generator Evaluates the model on a data generator.
Description
The generator should return the same kind of data as accepted by test_on_batch().
Usage
evaluate_generator(object, generator, steps, max_queue_size = 10,
workers = 1)
Arguments
object Model object to evaluate
generator Generator yielding lists (inputs, targets) or (inputs, targets, sample_weights)
steps Total number of steps (batches of samples) to yield from generator before
stopping.
max_queue_size Maximum size for the generator queue. If unspecified, max_queue_size will
default to 10.
workers Maximum number of threads to use for parallel processing. Note that parallel
processing will only be performed for native Keras generators (e.g. flow_images_from_directory())
as R based generators must run on the main thread.
Value
Named list of model test loss (or losses for models with multiple outputs) and model metrics.
See Also
Other model functions: compile.keras.engine.training.Model, evaluate.keras.engine.training.Model,
fit.keras.engine.training.Model, fit_generator, get_config, get_layer, keras_model_sequential,
keras_model, multi_gpu_model, pop_layer, predict.keras.engine.training.Model, predict_generator,
predict_on_batch, predict_proba, summary.keras.engine.training.Model, train_on_batch
export_savedmodel.keras.engine.training.Model 45
export_savedmodel.keras.engine.training.Model
Export a Saved Model
Description
Serialize a model to disk.
Usage

export_savedmodel(object,
export_dir_base, overwrite = TRUE, versioned = !overwrite,
remove_learning_phase = TRUE, as_text = FALSE, ...)
Arguments
object An R object.
export_dir_base
A string containing a directory in which to export the SavedModel.
overwrite Should the export_dir_base directory be overwritten?
versioned Should the model be exported under a versioned subdirectory?
remove_learning_phase
Should the learning phase be removed by saving and reloading the model? De-
faults to TRUE.
as_text Whether to write the SavedModel in text format.
... Unused
Value
The path to the exported directory, as a string.
fit.keras.engine.training.Model
Train a Keras model
Description
Trains the model for a fixed number of epochs (iterations on a dataset).

46 fit.keras.engine.training.Model
Usage
fit(object, x = NULL, y = NULL,
batch_size = NULL, epochs = 10,
verbose = getOption("keras.fit_verbose", default = 1),
callbacks = NULL, view_metrics = getOption("keras.view_metrics",
default = "auto"), validation_split = 0, validation_data = NULL,
shuffle = TRUE, class_weight = NULL, sample_weight = NULL,
initial_epoch = 0, steps_per_epoch = NULL, validation_steps = NULL,
...)
Arguments
object Model to train.
x Vector, matrix, or array of training data (or list if the model has multiple inputs).
If all inputs in the model are named, you can also pass a list mapping input
names to data. x can be NULL (default) if feeding from framework-native tensors
(e.g. TensorFlow data tensors).
y Vector, matrix, or array of target (label) data (or list if the model has multiple
outputs). If all outputs in the model are named, you can also pass a list mapping
output names to data. y can be NULL (default) if feeding from framework-native
tensors (e.g. TensorFlow data tensors).
batch_size Integer or NULL. Number of samples per gradient update. If unspecified, batch_size
will default to 32.
epochs Number of epochs to train the model. Note that in conjunction with initial_epoch,
epochs is to be understood as "final epoch". The model is not trained for a num-
ber of iterations given by epochs, but merely until the epoch of index epochs is
reached.
callbacks List of callbacks to be called during training.
view_metrics View realtime plot of training metrics (by epoch). The default ("auto") will
display the plot when running within RStudio, metrics were specified during
model compile(), epochs > 1 and verbose > 0. Use the global keras.view_metrics
option to establish a different default.
validation_split
Float between 0 and 1. Fraction of the training data to be used as validation
data. The model will set apart this fraction of the training data, will not train on
it, and will evaluate the loss and any model metrics on this data at the end of
each epoch. The validation data is selected from the last samples in the x and y
data provided, before shuffling.
validation_data
Data on which to evaluate the loss and any model metrics at the end of each
epoch. The model will not be trained on this data. This could be a list (x_val,
y_val) or a list (x_val, y_val, val_sample_weights). validation_data will
override validation_split.
fit_generator 47
shuffle shuffle: Logical (whether to shuffle the training data before each epoch) or string
(for "batch"). "batch" is a special option for dealing with the limitations of HDF5
data; it shuffles in batch-sized chunks. Has no effect when steps_per_epoch is
not NULL.
class_weight Optional named list mapping indices (integers) to a weight (float) value, used
for weighting the loss function (during training only). This can be useful to tell
the model to "pay more attention" to samples from an under-represented class.
sample_weight Optional array of the same length as x, containing weights to apply to the
model’s loss for each sample. In the case of temporal data, you can pass a
2D array with shape (samples, sequence_length), to apply a different weight to
every timestep of every sample. In this case you should make sure to specify
sample_weight_mode="temporal" in compile().
initial_epoch Integer, Epoch at which to start training (useful for resuming a previous training
run).
steps_per_epoch
Total number of steps (batches of samples) before declaring one epoch finished
and starting the next epoch. When training with input tensors such as Tensor-
Flow data tensors, the default NULL is equal to the number of samples in your
dataset divided by the batch size, or 1 if that cannot be determined.
validation_steps
Only relevant if steps_per_epoch is specified. Total number of steps (batches
of samples) to validate before stopping.
... Unused
Value
A history object that contains all information collected during training.
See Also

evaluate_generator, fit_generator, get_config, get_layer, keras_model_sequential, keras_model,
multi_gpu_model, pop_layer, predict.keras.engine.training.Model, predict_generator,
fit_generator Fits the model on data yielded batch-by-batch by a generator.
Description
The generator is run in parallel to the model, for efficiency. For instance, this allows you to do
real-time data augmentation on images on CPU in parallel to training your model on GPU.
48 fit_generator
Usage
fit_generator(object, generator, steps_per_epoch, epochs = 1,
verbose = getOption("keras.fit_verbose", default = 1),
callbacks = NULL, view_metrics = getOption("keras.view_metrics",
default = "auto"), validation_data = NULL, validation_steps = NULL,
class_weight = NULL, max_queue_size = 10, workers = 1,
initial_epoch = 0)
Arguments
object Keras model object
generator A generator (e.g. like the one provided by flow_images_from_directory()
or a custom R generator function).
The output of the generator must be a list of one of these forms:
- (inputs, targets)
- (inputs, targets, sample_weights)
This list (a single output of the generator) makes a single batch. Therefore, all
arrays in this list must have the same length (equal to the size of this batch). Dif-
ferent batches may have different sizes. For example, the last batch of the epoch
is commonly smaller than the others, if the size of the dataset is not divisible by
the batch size. The generator is expected to loop over its data indefinitely. An
epoch finishes when steps_per_epoch batches have been seen by the model.
steps_per_epoch
Total number of steps (batches of samples) to yield from generator before
declaring one epoch finished and starting the next epoch. It should typically be
equal to the number of samples if your dataset divided by the batch size.
epochs Integer. Number of epochs to train the model. An epoch is an iteration over the
entire data provided, as defined by steps_per_epoch. Note that in conjunction
with initial_epoch, epochs is to be understood as "final epoch". The model
is not trained for a number of iterations given by epochs, but merely until the
epoch of index epochs is reached.
callbacks List of callbacks to apply during training.
view_metrics View realtime plot of training metrics (by epoch). The default ("auto") will
display the plot when running within RStudio, metrics were specified during
model compile(), epochs > 1 and verbose > 0. Use the global keras.view_metrics
option to establish a different default.
validation_data
this can be either:
• a generator for the validation data
• a list (inputs, targets)
• a list (inputs, targets, sample_weights). on which to evaluate the loss and
any model metrics at the end of each epoch. The model will not be trained
on this data.
fit_image_data_generator 49
validation_steps
Only relevant if validation_data is a generator. Total number of steps (batches
of samples) to yield from generator before stopping at the end of every epoch.
It should typically be equal to the number of samples of your validation dataset
divided by the batch size.
class_weight Optional named list mapping class indices (integer) to a weight (float) value,
used for weighting the loss function (during training only). This can be useful
to tell the model to "pay more attention" to samples from an under-represented
class.
default to 10.
initial_epoch epoch at which to start training (useful for resuming a previous training run)
Value
Training history object (invisibly)
See Also
evaluate_generator, fit.keras.engine.training.Model, get_config, get_layer, keras_model_sequential,
fit_image_data_generator
Fit image data generator internal statistics to some sample data.
Description
Required for featurewise_center, featurewise_std_normalization and zca_whitening.
Usage
fit_image_data_generator(object, x, augment = FALSE, rounds = 1,
seed = NULL)
Arguments
object image_data_generator()
x array, the data to fit on (should have rank 4). In case of grayscale data, the
channels axis should have value 1, and in case of RGB data, it should have
value 3.
50 fit_text_tokenizer
augment Whether to fit on randomly augmented samples

rounds If augment, how many augmentation passes to do over the data
seed random seed.
See Also
Other image preprocessing: flow_images_from_dataframe, flow_images_from_data, flow_images_from_directory,

image_load, image_to_array
fit_text_tokenizer Update tokenizer internal vocabulary based on a list of texts or list of

sequences.
Description
Update tokenizer internal vocabulary based on a list of texts or list of sequences.
Usage
fit_text_tokenizer(object, x)
Arguments
object Tokenizer returned by text_tokenizer()

x Vector/list of strings, or a generator of strings (for memory-efficiency); Alterna-
tively a list of "sequence" (a sequence is a list of integer word indices).
Note
Required before using texts_to_sequences(), texts_to_matrix(), or sequences_to_matrix().
See Also
Other text tokenization: save_text_tokenizer, sequences_to_matrix, text_tokenizer, texts_to_matrix,

texts_to_sequences_generator, texts_to_sequences
flow_images_from_data 51
flow_images_from_data Generates batches of augmented/normalized data from image data

and labels
Description
Generates batches of augmented/normalized data from image data and labels
Usage
flow_images_from_data(x, y = NULL, generator = image_data_generator(),
batch_size = 32, shuffle = TRUE, sample_weight = NULL,
seed = NULL, save_to_dir = NULL, save_prefix = "",
save_format = "png", subset = NULL)
Arguments
x data. Should have rank 4. In case of grayscale data, the channels axis should
have value 1, and in case of RGB data, it should have value 3.
y labels (can be NULL if no labels are required)
generator Image data generator to use for augmenting/normalizing image data.
batch_size int (default: 32).
shuffle boolean (defaut: TRUE).
sample_weight Sample weights.
seed int (default: NULL).
save_to_dir NULL or str (default: NULL). This allows you to optionally specify a directory
to which to save the augmented pictures being generated (useful for visualizing
what you are doing).
save_prefix str (default: ”). Prefix to use for filenames of saved pictures (only relevant if
save_to_dir is set).
save_format one of "png", "jpeg" (only relevant if save_to_dir is set). Default: "png".
subset Subset of data ("training" or "validation") if validation_split is set in
image_data_generator().
Details
Yields batches indefinitely, in an infinite loop.
Yields
(x, y) where x is an array of image data and y is a array of corresponding labels. The generator
loops indefinitely.
52 flow_images_from_dataframe
See Also
Other image preprocessing: fit_image_data_generator, flow_images_from_dataframe, flow_images_from_director
flow_images_from_dataframe
Takes the dataframe and the path to a directory and generates batches
of augmented/normalized data.
Description
Takes the dataframe and the path to a directory and generates batches of augmented/normalized
data.
Usage
flow_images_from_dataframe(dataframe, directory = NULL,
x_col = "filename", y_col = "class",
generator = image_data_generator(), target_size = c(256, 256),
color_mode = "rgb", classes = NULL, class_mode = "categorical",
batch_size = 32, shuffle = TRUE, seed = NULL, save_to_dir = NULL,
save_prefix = "", save_format = "png", subset = NULL,
interpolation = "nearest", drop_duplicates = TRUE)
Arguments
dataframe data.frame containing the filepaths relative to directory (or absolute paths if
directory is NULL) of the images in a character column. It should include other
column/s depending on the class_mode:
• if class_mode is "categorical" (default value) it must include the y_col col-
umn with the class/es of each image. Values in column can be character/list
if a single class or list if multiple classes.
• if class_mode is "binary" or "sparse" it must include the given y_col col-
umn with class values as strings.
• if class_mode is "other" it should contain the columns specified in y_col.
• if class_mode is "input" or NULL no extra column is needed.
directory character, path to the directory to read images from. If NULL, data in x_col
column should be absolute paths.
x_col character, column in dataframe that contains the filenames (or absolute paths if
directory is NULL).
y_col string or list, column/s in dataframe that has the target data.
generator Image data generator to use for augmenting/normalizing image data.
target_size Either NULL (default to original size) or integer vector (img_height, img_width).
flow_images_from_dataframe 53
color_mode one of "grayscale", "rgb". Default: "rgb". Whether the images will be converted
to have 1 or 3 color channels.
classes optional list of classes (e.g. c('dogs', 'cats'). Default: NULL If not provided,
the list of classes will be automatically inferred from the y_col, which will
map to the label indices, will be alphanumeric). The dictionary containing the
mapping from class names to class indices can be obtained via the attribute
class_indices.
class_mode one of "categorical", "binary", "sparse", "input", "other" or None. Default: "cat-
egorical". Mode for yielding the targets:
• "binary": 1D array of binary labels,
• "categorical": 2D array of one-hot encoded labels. Supports multi-label
output.
• "sparse": 1D array of integer labels,
• "input": images identical to input images (mainly used to work with au-
toencoders),
• "other": array of y_col data, NULL, no targets are returned (the generator will
only yield batches of image data, which is useful to use in predict_generator()).
interpolation Interpolation method used to resample the image if the target size is different
from that of the loaded image. Supported methods are "nearest", "bilinear", and
"bicubic". If PIL version 1.1.3 or newer is installed, "lanczos" is also supported.
If PIL version 3.4.0 or newer is installed, "box" and "hamming" are also sup-
ported. By default, "nearest" is used.
drop_duplicates
Boolean, whether to drop duplicate rows based on filename.
Details
Yields
loops indefinitely.
54 flow_images_from_directory
Note
This functions requires that pandas (python module) is installed in the same environment as tensorflow
and keras.
If you are using r-tensorflow (the default environment) you can install pandas by running reticulate::virtualenv_inst
or reticulate::conda_install("pandas", envname = "r-tensorflow") depending on the
kind of environment you are using.
See Also
Other image preprocessing: fit_image_data_generator, flow_images_from_data, flow_images_from_directory,
flow_images_from_directory
Generates batches of data from images in a directory (with optional
augmented/normalized data)
Description
Generates batches of data from images in a directory (with optional augmented/normalized data)
Usage
flow_images_from_directory(directory, generator = image_data_generator(),
target_size = c(256, 256), color_mode = "rgb", classes = NULL,
class_mode = "categorical", batch_size = 32, shuffle = TRUE,
seed = NULL, save_to_dir = NULL, save_prefix = "",
save_format = "png", follow_links = FALSE, subset = NULL,
interpolation = "nearest")
Arguments
directory path to the target directory. It should contain one subdirectory per class. Any
PNG, JPG, BMP, PPM, or TIF images inside each of the subdirectories directory
tree will be included in the generator. See this script for more details.
generator Image data generator (default generator does no data augmentation/normalization
transformations)
target_size integer vector, default: c(256, 256). The dimensions to which all images found
will be resized.
color_mode one of "grayscale", "rbg". Default: "rgb". Whether the images will be converted
to have 1 or 3 color channels.
classes optional list of class subdirectories (e.g. c('dogs', 'cats')). Default: NULL,
If not provided, the list of classes will be automatically inferred (and the order
of the classes, which will map to the label indices, will be alphanumeric).
flow_images_from_directory 55
class_mode one of "categorical", "binary", "sparse" or NULL. Default: "categorical". Deter-

mines the type of label arrays that are returned: "categorical" will be 2D one-hot
encoded labels, "binary" will be 1D binary labels, "sparse" will be 1D integer la-
bels. If NULL, no labels are returned (the generator will only yield batches of im-
age data, which is useful to use predict_generator(), evaluate_generator(),
etc.).
follow_links whether to follow symlinks inside class subdirectories (default: FALSE)
Details
Yields
loops indefinitely.
See Also
Other image preprocessing: fit_image_data_generator, flow_images_from_dataframe, flow_images_from_data,

56 freeze_weights
freeze_weights Freeze and unfreeze weights
Description
Freeze weights in a model or layer so that they are no longer trainable.
Usage
freeze_weights(object, from = NULL, to = NULL)
unfreeze_weights(object, from = NULL, to = NULL)
Arguments
object Keras model or layer object
from Layer instance, layer name, or layer index within model
to Layer instance, layer name, or layer index within model
Note
The from and to layer arguments are both inclusive.
When applied to a model, the freeze or unfreeze is a global operation over all layers in the model
(i.e. layers not within the specified range will be set to the opposite value, e.g. unfrozen for a call
to freeze).
Models must be compiled again after weights are frozen or unfrozen.
Examples
## Not run:
# instantiate a VGG16 model
conv_base <- application_vgg16(
weights = "imagenet",
include_top = FALSE,
input_shape = c(150, 150, 3)
)
# freeze it's weights

freeze_weights(conv_base)
# create a composite model that includes the base + more layers

model <- keras_model_sequential() %>%
conv_base %>%
layer_flatten() %>%
layer_dense(units = 256, activation = "relu") %>%
layer_dense(units = 1, activation = "sigmoid")
# compile
generator_next 57
model %>% compile(

loss = "binary_crossentropy",
optimizer = optimizer_rmsprop(lr = 2e-5),
metrics = c("accuracy")
)
# unfreeze weights from "block5_conv1" on

unfreeze_weights(conv_base, from = "block5_conv1")
# compile again since we froze or unfroze weights

model %>% compile(
optimizer = optimizer_rmsprop(lr = 2e-5),
metrics = c("accuracy")
)
## End(Not run)
generator_next Retrieve the next item from a generator
Description
Use to retrieve items from generators (e.g. image_data_generator()). Will return either the next
item or NULL if there are no more items.
Usage
generator_next(generator, completed = NULL)
Arguments
generator Generator
completed Sentinel value to return from generator_next() if the iteration completes (de-
faults to NULL but can be any R value you specify).
get_config Layer/Model configuration
Description
A layer config is an object returned from get_config() that contains the configuration of a layer
or model. The same layer or model can be reinstantiated later (without its trained weights) from this
configuration using from_config(). The config does not include connectivity information, nor the
class name (those are handled externally).
58 get_file
Usage
get_config(object)
from_config(config)
Arguments
config Object with layer or model configuration
Value
get_config() returns an object with the configuration, from_config() returns a re-instantation of
hte object.
Note
Objects returned from get_config() are not serializable. Therefore, if you want to save and re-
store a model across sessions, you can use the model_to_json() or model_to_yaml() functions
(for model configuration only, not weights) or the save_model_hdf5() function to save the model
configuration and weights to a file.
See Also
evaluate_generator, fit.keras.engine.training.Model, fit_generator, get_layer, keras_model_sequential,
Other layer methods: count_params, get_input_at, get_weights, reset_states
get_file Downloads a file from a URL if it not already in the cache.
Description
Passing the MD5 hash will verify the file after download as well as if it is already present in the
cache.
Usage
get_file(fname, origin, file_hash = NULL, cache_subdir = "datasets",
hash_algorithm = "auto", extract = FALSE, archive_format = "auto",
cache_dir = NULL)
get_input_at 59
Arguments
fname Name of the file. If an absolute path /path/to/file.txt is specified the file
will be saved at that location.
origin Original URL of the file.
file_hash The expected hash string of the file after download. The sha256 and md5 hash
algorithms are both supported.
cache_subdir Subdirectory under the Keras cache dir where the file is saved. If an absolute
path /path/to/folder is specified the file will be saved at that location.
hash_algorithm Select the hash algorithm to verify the file. options are ’md5’, ’sha256’, and
’auto’. The default ’auto’ detects the hash algorithm in use.
extract True tries extracting the file as an Archive, like tar or zip.
archive_format Archive format to try for extracting the file. Options are ’auto’, ’tar’, ’zip’, and
None. ’tar’ includes tar, tar.gz, and tar.bz files. The default ’auto’ is (’tar’, ’zip’).
None or an empty list will return no matches found.
cache_dir Location to store cached files, when NULL it defaults to the Keras configuration
directory.
Value
Path to the downloaded file
get_input_at Retrieve tensors for layers with multiple nodes
Description
Whenever you are calling a layer on some input, you are creating a new tensor (the output of the
layer), and you are adding a "node" to the layer, linking the input tensor to the output tensor. When
you are calling the same layer multiple times, that layer owns multiple nodes indexed as 1, 2, 3.
These functions enable you to retrieve various tensor properties of layers with multiple nodes.
Usage
get_input_at(object, node_index)
get_output_at(object, node_index)
get_input_shape_at(object, node_index)
get_output_shape_at(object, node_index)
get_input_mask_at(object, node_index)
get_output_mask_at(object, node_index)
60 get_layer
Arguments
node_index Integer, index of the node from which to retrieve the attribute. E.g. node_index = 1
will correspond to the first time the layer was called.
Value
A tensor (or list of tensors if the layer has multiple inputs/outputs).
See Also
Other layer methods: count_params, get_config, get_weights, reset_states
get_layer Retrieves a layer based on either its name (unique) or index.
Description
Indices are based on order of horizontal graph traversal (bottom-up) and are 1-based. If name and
index are both provided, index will take precedence.
Usage
get_layer(object, name = NULL, index = NULL)
Arguments
name String, name of layer.
index Integer, index of layer (0-based)
Value
A layer instance.
See Also
evaluate_generator, fit.keras.engine.training.Model, fit_generator, get_config, keras_model_sequential,
get_weights 61
get_weights Layer/Model weights as R arrays
Description
Layer/Model weights as R arrays
Usage
get_weights(object)
set_weights(object, weights)
Arguments
weights Weights as R array
See Also
Other model persistence: model_to_json, model_to_yaml, save_model_hdf5, save_model_weights_hdf5,
serialize_model
Other layer methods: count_params, get_config, get_input_at, reset_states
hdf5_matrix Representation of HDF5 dataset to be used instead of an R array
Description
Representation of HDF5 dataset to be used instead of an R array
Usage
hdf5_matrix(datapath, dataset, start = 0, end = NULL,
normalizer = NULL)
Arguments
datapath string, path to a HDF5 file
dataset string, name of the HDF5 dataset in the file specified in datapath
start int, start of desired slice of the specified dataset
end int, end of desired slice of the specified dataset
normalizer function to be called on data when retrieved
62 imagenet_preprocess_input
Details
Providing start and end allows use of a slice of the dataset.
Optionally, a normalizer function (or lambda) can be given. This will be called on every slice of
data retrieved.
Value
An array-like HDF5 dataset.
imagenet_decode_predictions
Decodes the prediction of an ImageNet model.
Description
Decodes the prediction of an ImageNet model.
Usage
imagenet_decode_predictions(preds, top = 5)
Arguments
Value
List of data frames with variables class_name, class_description, and score (one data frame
per sample in batch input).
imagenet_preprocess_input
Preprocesses a tensor or array encoding a batch of images.
Description
Preprocesses a tensor or array encoding a batch of images.
Usage
imagenet_preprocess_input(x, data_format = NULL, mode = "caffe")
image_data_generator 63
Arguments
x Input Numpy or symbolic tensor, 3D or 4D.
data_format Data format of the image tensor/array.
mode One of "caffe", "tf", or "torch"
• caffe: will convert the images from RGB to BGR, then will zero-center
each color channel with respect to the ImageNet dataset, without scaling.
• tf: will scale pixels between -1 and 1, sample-wise.
• torch: will scale pixels between 0 and 1 and then will normalize each chan-
nel with respect to the ImageNet dataset.
Value
Preprocessed tensor or array.
image_data_generator Generate batches of image data with real-time data augmentation. The
data will be looped over (in batches).
Description
Generate batches of image data with real-time data augmentation. The data will be looped over (in
batches).
Usage
image_data_generator(featurewise_center = FALSE,
samplewise_center = FALSE, featurewise_std_normalization = FALSE,
samplewise_std_normalization = FALSE, zca_whitening = FALSE,
zca_epsilon = 1e-06, rotation_range = 0, width_shift_range = 0,
height_shift_range = 0, brightness_range = NULL, shear_range = 0,
zoom_range = 0, channel_shift_range = 0, fill_mode = "nearest",
cval = 0, horizontal_flip = FALSE, vertical_flip = FALSE,
rescale = NULL, preprocessing_function = NULL, data_format = NULL,
validation_split = 0)
Arguments
featurewise_center
Set input mean to 0 over the dataset, feature-wise.
samplewise_center
Boolean. Set each sample mean to 0.
featurewise_std_normalization
Divide inputs by std of the dataset, feature-wise.
samplewise_std_normalization
Divide each input by its std.
64 image_data_generator
zca_whitening apply ZCA whitening.

zca_epsilon Epsilon for ZCA whitening. Default is 1e-6.
rotation_range degrees (0 to 180).
width_shift_range
fraction of total width.
height_shift_range
fraction of total height.
brightness_range
the range of brightness to apply
shear_range shear intensity (shear angle in radians).
zoom_range amount of zoom. if scalar z, zoom will be randomly picked in the range [1-z, 1+z].
A sequence of two can be passed instead to select this range.
channel_shift_range
shift range for each channels.
fill_mode One of "constant", "nearest", "reflect" or "wrap". Points outside the boundaries
of the input are filled according to the given mode:
• "constant": kkkkkkkk|abcd|kkkkkkkk (cval=k)
• "nearest": aaaaaaaa|abcd|dddddddd
• "reflect": abcddcba|abcd|dcbaabcd
• "wrap": abcdabcd|abcd|abcdabcd
cval value used for points outside the boundaries when fill_mode is ’constant’. De-
fault is 0.
horizontal_flip
whether to randomly flip images horizontally.
vertical_flip whether to randomly flip images vertically.
rescale rescaling factor. If NULL or 0, no rescaling is applied, otherwise we multiply
the data by the value provided (before applying any other transformation).
preprocessing_function
function that will be implied on each input. The function will run before any
other modification on it. The function should take one argument: one image
(tensor with rank 3), and should output a tensor with the same shape.
data_format ’channels_first’ or ’channels_last’. In ’channels_first’ mode, the channels di-
mension (the depth) is at index 1, in ’channels_last’ mode it is at index 3. It
defaults to the image_data_format value found in your Keras config file at
~/.keras/keras.json. If you never set it, then it will be "channels_last".
validation_split
fraction of images reserved for validation (strictly between 0 and 1).
image_load 65
image_load Loads an image into PIL format.
Description
Loads an image into PIL format.
Usage
image_load(path, grayscale = FALSE, target_size = NULL,

interpolation = "nearest")
Arguments
path Path to image file

grayscale Boolean, whether to load the image as grayscale.
target_size Either NULL (default to original size) or integer vector (img_height, img_width).
Value
A PIL Image instance.
See Also

flow_images_from_directory, image_to_array
image_to_array 3D array representation of images
Description
3D array that represents an image with dimensions (height,width,channels) or (channels,height,width)

depending on the data_format.
66 implementation
Usage
image_to_array(img, data_format = c("channels_last", "channels_first"))
image_array_resize(img, height, width, data_format = c("channels_last",

"channels_first"))
image_array_save(img, path, data_format = NULL, file_format = NULL,

scale = TRUE)
Arguments
img Image
data_format Image data format ("channels_last" or "channels_first")
height Height to resize to
width Width to resize to
path Path to save image to
file_format Optional file format override. If omitted, the format to use is determined from
the filename extension. If a file object was used instead of a filename, this
parameter should always be used.
scale Whether to rescale image values to be within 0,255
See Also
flow_images_from_directory, image_load
implementation Keras implementation
Description
Obtain a reference to the Python module used for the implementation of Keras.
Usage
implementation()
Details
There are currently two Python modules which implement Keras:
• keras ("keras")
• tensorflow.keras ("tensorflow")
This function returns a reference to the implementation being currently used by the keras package.
The default implementation is "keras". You can override this by setting the KERAS_IMPLEMENTATION
environment variable to "tensorflow".
initializer_constant 67
Value
Reference to the Python module used for the implementation of Keras.
initializer_constant Initializer that generates tensors initialized to a constant value.
Description
Initializer that generates tensors initialized to a constant value.
Usage
initializer_constant(value = 0)
Arguments
value float; the value of the generator tensors.
See Also
Other initializers: initializer_glorot_normal, initializer_glorot_uniform, initializer_he_normal,
initializer_he_uniform, initializer_identity, initializer_lecun_normal, initializer_lecun_uniform,
initializer_ones, initializer_orthogonal, initializer_random_normal, initializer_random_uniform,
initializer_truncated_normal, initializer_variance_scaling, initializer_zeros
initializer_glorot_normal
Glorot normal initializer, also called Xavier normal initializer.
Description
It draws samples from a truncated normal distribution centered on 0 with stddev = sqrt(2 / (fan_in + fan_out))
where fan_in is the number of input units in the weight tensor and fan_out is the number of output
units in the weight tensor.
Usage
initializer_glorot_normal(seed = NULL)
Arguments
seed Integer used to seed the random generator.
68 initializer_glorot_uniform
References
Glorot & Bengio, AISTATS 2010 http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.

pdf
See Also
Other initializers: initializer_constant, initializer_glorot_uniform, initializer_he_normal,

initializer_glorot_uniform
Glorot uniform initializer, also called Xavier uniform initializer.
Description
It draws samples from a uniform distribution within -limit, limit where limit is sqrt(6 / (fan_in + fan_out))
where fan_in is the number of input units in the weight tensor and fan_out is the number of output
units in the weight tensor.
Usage
initializer_glorot_uniform(seed = NULL)
Arguments
References
Glorot & Bengio, AISTATS 2010 http://jmlr.org/proceedings/papers/v9/glorot10a/glorot10a.pdf
See Also
Other initializers: initializer_constant, initializer_glorot_normal, initializer_he_normal,

initializer_he_normal 69
initializer_he_normal He normal initializer.
Description
It draws samples from a truncated normal distribution centered on 0 with stddev = sqrt(2 / fan_in)
where fan_in is the number of input units in the weight tensor.
Usage
initializer_he_normal(seed = NULL)
Arguments
References
He et al., http://arxiv.org/abs/1502.01852
See Also
Other initializers: initializer_constant, initializer_glorot_normal, initializer_glorot_uniform,
initializer_he_uniform
He uniform variance scaling initializer.
Description
It draws samples from a uniform distribution within -limit, limit where limit`` is sqrt(6 /
fan_in)where fan_in‘ is the number of input units in the weight tensor.
Usage
initializer_he_uniform(seed = NULL)
Arguments
References
He et al., http://arxiv.org/abs/1502.01852
70 initializer_lecun_normal
See Also
initializer_he_normal, initializer_identity, initializer_lecun_normal, initializer_lecun_uniform,
initializer_identity Initializer that generates the identity matrix.
Description
Only use for square 2D matrices.
Usage
initializer_identity(gain = 1)
Arguments
gain Multiplicative factor to apply to the identity matrix
See Also
initializer_he_normal, initializer_he_uniform, initializer_lecun_normal, initializer_lecun_uniform,
initializer_lecun_normal
LeCun normal initializer.
Description
It draws samples from a truncated normal distribution centered on 0 with stddev <- sqrt(1 / fan_in)
where fan_in is the number of input units in the weight tensor..
Usage
initializer_lecun_normal(seed = NULL)
Arguments
seed A Python integer. Used to seed the random generator.
initializer_lecun_uniform 71
References
• Self-Normalizing Neural Networks

• Efficient Backprop
See Also

initializer_he_normal, initializer_he_uniform, initializer_identity, initializer_lecun_uniform,
initializer_lecun_uniform
LeCun uniform initializer.
Description
It draws samples from a uniform distribution within -limit, limit where limit is sqrt(3 / fan_in)
where fan_in is the number of input units in the weight tensor.
Usage
initializer_lecun_uniform(seed = NULL)
Arguments
References
LeCun 98, Efficient Backprop, http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf
See Also

initializer_he_normal, initializer_he_uniform, initializer_identity, initializer_lecun_normal,
72 initializer_orthogonal
initializer_ones Initializer that generates tensors initialized to 1.
Description
Initializer that generates tensors initialized to 1.
Usage
initializer_ones()
See Also
initializer_lecun_uniform, initializer_orthogonal, initializer_random_normal, initializer_random_unifor
initializer_orthogonal
Initializer that generates a random orthogonal matrix.
Description
Initializer that generates a random orthogonal matrix.
Usage
initializer_orthogonal(gain = 1, seed = NULL)
Arguments
gain Multiplicative factor to apply to the orthogonal matrix.
References
Saxe et al., http://arxiv.org/abs/1312.6120
See Also
initializer_lecun_uniform, initializer_ones, initializer_random_normal, initializer_random_uniform,
initializer_random_normal 73
initializer_random_normal
Initializer that generates tensors with a normal distribution.
Description
Initializer that generates tensors with a normal distribution.
Usage
initializer_random_normal(mean = 0, stddev = 0.05, seed = NULL)
Arguments
mean Mean of the random values to generate.
stddev Standard deviation of the random values to generate.
See Also
initializer_lecun_uniform, initializer_ones, initializer_orthogonal, initializer_random_uniform,
initializer_random_uniform
Initializer that generates tensors with a uniform distribution.
Description
Initializer that generates tensors with a uniform distribution.
Usage
initializer_random_uniform(minval = -0.05, maxval = 0.05,
seed = NULL)
Arguments
minval Lower bound of the range of random values to generate.
maxval Upper bound of the range of random values to generate. Defaults to 1 for float
types.
seed seed
74 initializer_variance_scaling
See Also
initializer_lecun_uniform, initializer_ones, initializer_orthogonal, initializer_random_normal,
initializer_truncated_normal
Initializer that generates a truncated normal distribution.
Description
These values are similar to values from an initializer_random_normal() except that values
more than two standard deviations from the mean are discarded and re-drawn. This is the recom-
mended initializer for neural network weights and filters.
Usage
initializer_truncated_normal(mean = 0, stddev = 0.05, seed = NULL)
Arguments
mean Mean of the random values to generate.
stddev Standard deviation of the random values to generate.
See Also
initializer_random_uniform, initializer_variance_scaling, initializer_zeros
initializer_variance_scaling
Initializer capable of adapting its scale to the shape of weights.
Description
With distribution="normal", samples are drawn from a truncated normal distribution centered
on zero, with stddev = sqrt(scale / n) where n is:
• number of input units in the weight tensor, if mode = "fan_in"
• number of output units, if mode = "fan_out"
• average of the numbers of input and output units, if mode = "fan_avg"
initializer_zeros 75
Usage
initializer_variance_scaling(scale = 1, mode = c("fan_in", "fan_out",

"fan_avg"), distribution = c("normal", "uniform"), seed = NULL)
Arguments
scale Scaling factor (positive float).

mode One of "fan_in", "fan_out", "fan_avg".
distribution One of "normal", "uniform"
Details
With distribution="uniform", samples are drawn from a uniform distribution within -limit, limit,
with limit = sqrt(3 * scale / n).
See Also

initializer_random_uniform, initializer_truncated_normal, initializer_zeros
initializer_zeros Initializer that generates tensors initialized to 0.
Description
Initializer that generates tensors initialized to 0.
Usage
initializer_zeros()
See Also

initializer_random_uniform, initializer_truncated_normal, initializer_variance_scaling
76 install_keras
install_keras Install Keras and the TensorFlow backend
Description
Keras and TensorFlow will be installed into an "r-tensorflow" virtual or conda environment. Note
that "virtualenv" is not available on Windows (as this isn’t supported by TensorFlow).
Usage
install_keras(method = c("auto", "virtualenv", "conda"),
conda = "auto", version = "default", tensorflow = "default",
extra_packages = c("tensorflow-hub"))
Arguments
method Installation method ("virtualenv" or "conda")
conda Path to conda executable (or "auto" to find conda using the PATH and other
conventional install locations).
version Version of Keras to install. Specify "default" to install the latest release. Other-
wise specify an alternate version (e.g. "2.2.2").
tensorflow TensorFlow version to install. Specify "default" to install the CPU version of
the latest release. Specify "gpu" to install the GPU version of the latest release.
You can also provide a full major.minor.patch specification (e.g. "1.1.0"), ap-
pending "-gpu" if you want the GPU version (e.g. "1.1.0-gpu").
Alternatively, you can provide the full URL to an installer binary (e.g. for a
nightly binary).
extra_packages Additional PyPI packages to install along with Keras and TensorFlow.
GPU Installation
Keras and TensorFlow can be configured to run on either CPUs or GPUs. The CPU version is much
easier to install and configure so is the best starting place especially when you are first learning how
to use Keras. Here’s the guidance on CPU vs. GPU versions from the TensorFlow website:
• TensorFlow with CPU support only. If your system does not have a NVIDIA® GPU, you must
install this version. Note that this version of TensorFlow is typically much easier to install, so
even if you have an NVIDIA GPU, we recommend installing this version first.
• TensorFlow with GPU support. TensorFlow programs typically run significantly faster on a
GPU than on a CPU. Therefore, if your system has a NVIDIA® GPU meeting all prerequi-
sites and you need to run performance-critical applications, you should ultimately install this
version.
To install the GPU version:
1. Ensure that you have met all installation prerequisites including installation of the CUDA and
cuDNN libraries as described in TensorFlow GPU Prerequistes.
install_keras 77
2. Pass tensorflow = "gpu" to install_keras(). For example:
install_keras(tensorflow = "gpu")
Windows Installation
The only supported installation method on Windows is "conda". This means that you should install
Anaconda 3.x for Windows prior to installing Keras.
Custom Installation
Installing Keras and TensorFlow using install_keras() isn’t required to use the Keras R package.
You can do a custom installation of Keras (and desired backend) as described on the Keras website
and the Keras R package will find and use that version.
See the documentation on custom installations for additional information on how version of Keras
and TensorFlow are located by the Keras package.
Additional Packages
If you wish to add additional PyPI packages to your Keras / TensorFlow environment you can either
specify the packages in the extra_packages argument of install_keras(), or alternatively install
them into an existing environment using the reticulate::py_install() function.
Examples
## Not run:
# default installation
library(keras)
install_keras()
# install using a conda environment (default is virtualenv)

install_keras(method = "conda")
# install with GPU version of TensorFlow

# (NOTE: only do this if you have an NVIDIA GPU + CUDA!)
install_keras(tensorflow = "gpu")
# install a specific version of TensorFlow

install_keras(tensorflow = "1.2.1")
install_keras(tensorflow = "1.2.1-gpu")
## End(Not run)
78 KerasCallback
is_keras_available Check if Keras is Available
Description
Probe to see whether the Keras python package is available in the current system environment.
Usage
is_keras_available(version = NULL)
Arguments
version Minimum required version of Keras (defaults to NULL, no required version).
Value
Logical indicating whether Keras (or the specified minimum version of Keras) is available.
Examples
## Not run:
# testthat utilty for skipping tests when Keras isn't available
skip_if_no_keras <- function(version = NULL) {
if (!is_keras_available(version))
skip("Required keras version not available for testing")
}
# use the function within a test

test_that("keras function works correctly", {
skip_if_no_keras()
# test code here
})
## End(Not run)
KerasCallback Base R6 class for Keras callbacks
Description
Base R6 class for Keras callbacks
Usage
KerasCallback
KerasCallback 79
Format
An R6Class generator object
Details
The logs named list that callback methods take as argument will contain keys for quantities relevant
to the current batch or epoch.
Currently, the fit.keras.engine.training.Model() method for sequential models will include
the following quantities in the logs that it passes to its callbacks:
• on_epoch_end: logs include acc and loss, and optionally include val_loss (if validation is
enabled in fit), and val_acc (if validation and accuracy monitoring are enabled).
• on_batch_begin: logs include size, the number of samples in the current batch.
• on_batch_end: logs include loss, and optionally acc (if accuracy monitoring is enabled).
Value
KerasCallback.
Fields
params Named list with training parameters (eg. verbosity, batch size, number of epochs...).
model Reference to the Keras model being trained.
Methods
on_epoch_begin(epoch, logs) Called at the beginning of each epoch.
on_epoch_end(epoch, logs) Called at the end of each epoch.
on_batch_begin(batch, logs) Called at the beginning of each batch.
on_batch_end(batch, logs) Called at the end of each batch.
on_train_begin(logs) Called at the beginning of training.
on_train_end(logs) Called at the end of training.
Examples
## Not run:
library(keras)
LossHistory <- R6::R6Class("LossHistory",

inherit = KerasCallback,
public = list(
losses = NULL,
on_batch_end = function(batch, logs = list()) {

self$losses <- c(self$losses, logs[["loss"]])
}
80 KerasConstraint
)
)
## End(Not run)
KerasConstraint Base R6 class for Keras constraints
Description
Base R6 class for Keras constraints
Usage
KerasConstraint
Format
Details
You can implement a custom constraint either by creating an R function that accepts a weights (w)
parameter, or by creating an R6 class that derives from KerasConstraint and implements a call
method.
Methods
call(w) Constrain the specified weights.
Note
Models which use custom constraints cannot be serialized using save_model_hdf5(). Rather, the
weights of the model should be saved and restored using save_model_weights_hdf5().
See Also
constraints
Examples
## Not run:
CustomNonNegConstraint <- R6::R6Class(
"CustomNonNegConstraint",
inherit = KerasConstraint,
public = list(
call = function(x) {
w * k_cast(k_greater_equal(w, 0), k_floatx())
}
KerasLayer 81
)
)
layer_dense(units = 32, input_shape = c(784),

kernel_constraint = CustomNonNegConstraint$new())
## End(Not run)
KerasLayer Base R6 class for Keras layers
Description
Base R6 class for Keras layers
Usage
KerasLayer
Format
An R6Class generator object #’
Value
KerasLayer.
Methods
build(input_shape) Creates the layer weights (must be implemented by all layers that have
weights)
call(inputs,mask) Call the layer on an input tensor.
compute_output_shape(input_shape) Compute the output shape for the layer.
add_loss(losses, inputs) Add losses to the layer.
add_weight(name,shape,dtype,initializer,regularizer,trainable,constraint) Adds a
weight variable to the layer.
82 keras_array
KerasWrapper Base R6 class for Keras wrappers
Description
Base R6 class for Keras wrappers
Usage
KerasWrapper
Format
Value
KerasWrapper.
Methods
build(input_shape) Builds the wrapped layer. Subclasses can extend this to perform custom
operations on that layer.
call(inputs,mask) Calls the wrapped layer on an input tensor.
compute_output_shape(input_shape) Computes the output shape for the wrapped layer.
add_loss(losses, inputs) Subclasses can use this to add losses to the wrapped layer.
add_weight(name,shape,dtype,initializer,regularizer,trainable,constraint) Subclasses
can use this to add weights to the wrapped layer.
keras_array Keras array object
Description
Convert an R vector, matrix, or array object to an array that has the optimal in-memory layout and
floating point data type for the current Keras backend.
Usage
keras_array(x, dtype = NULL)
Arguments
x Object or list of objects to convert
dtype NumPy data type (e.g. float32, float64). If this is unspecified then R doubles will
be converted to the default floating point type for the current Keras backend.
keras_model 83
Details
Keras does frequent row-oriented access to arrays (for shuffling and drawing batches) so the order
of arrays created by this function is always row-oriented ("C" as opposed to "Fortran" ordering,
which is the default for R arrays).
If the passed array is already a NumPy array with the desired dtype and "C" order then it is returned
unmodified (no additional copies are made).
Value
NumPy array with the specified dtype (or list of NumPy arrays if a list was passed for x).
keras_model Keras Model
Description
A model is a directed acyclic graph of layers.
Usage
keras_model(inputs, outputs = NULL)
Arguments
inputs Input layer
outputs Output layer
See Also
evaluate_generator, fit.keras.engine.training.Model, fit_generator, get_config, get_layer,
keras_model_sequential, multi_gpu_model, pop_layer, predict.keras.engine.training.Model,
predict_generator, predict_on_batch, predict_proba, summary.keras.engine.training.Model,
train_on_batch
Examples
## Not run:
library(keras)
# input layer
inputs <- layer_input(shape = c(784))
# outputs compose input + dense layers

predictions <- inputs %>%
layer_dense(units = 64, activation = 'relu') %>%
layer_dense(units = 64, activation = 'relu') %>%
layer_dense(units = 10, activation = 'softmax')
84 keras_model_sequential
# create and compile model

model <- keras_model(inputs = inputs, outputs = predictions)
model %>% compile(
optimizer = 'rmsprop',
loss = 'categorical_crossentropy',
metrics = c('accuracy')
)
## End(Not run)
keras_model_custom Create a Keras custom model
Description
Create a Keras custom model
Usage
keras_model_custom(model_fn, name = NULL)
Arguments
model_fn Function that returns an R custom model
name Optional name for model
Details
For documentation on using custom models, see https://keras.rstudio.com/articles/custom_
models.html.
Value
A Keras model
keras_model_sequential
Keras Model composed of a linear stack of layers
Description
Keras Model composed of a linear stack of layers
Usage
keras_model_sequential(layers = NULL, name = NULL)
k_abs 85
Arguments
layers List of layers to add to the model
name Name of model
Note
The first layer passed to a Sequential model should have a defined input shape. What that means is
that it should have received an input_shape or batch_input_shape argument, or for some type
of layers (recurrent, Dense...) an input_dim argument.
See Also
Examples
## Not run:
library(keras)
model <- keras_model_sequential()

model %>%
layer_dense(units = 32, input_shape = c(784)) %>%
layer_activation('relu') %>%
layer_dense(units = 10) %>%
layer_activation('softmax')
model %>% compile(

optimizer = 'rmsprop',
metrics = c('accuracy')
)
## End(Not run)
k_abs Element-wise absolute value.
Description
Element-wise absolute value.
Usage
k_abs(x)
86 k_all
Arguments
x Tensor or variable.
Value
A tensor.
Keras Backend
This function is part of a set of Keras backend functions that enable lower level access to the core
operations of the backend tensor engine (e.g. TensorFlow, CNTK, Theano, etc.).
You can see a list of all available backend functions here: https://keras.rstudio.com/articles/
backend.html#backend-functions.
k_all Bitwise reduction (logical AND).
Description
Bitwise reduction (logical AND).
Usage
k_all(x, axis = NULL, keepdims = FALSE)
Arguments
axis Axis along which to perform the reduction (axis indexes are 1-based).
keepdims whether the drop or broadcast the reduction axes.
Value
A uint8 tensor (0s and 1s).
Keras Backend
k_any 87
k_any Bitwise reduction (logical OR).
Description
Bitwise reduction (logical OR).
Usage
k_any(x, axis = NULL, keepdims = FALSE)
Arguments
axis Axis along which to perform the reduction (axis indexes are 1-based).
keepdims whether the drop or broadcast the reduction axes.
Value
A uint8 tensor (0s and 1s).
Keras Backend
k_arange Creates a 1D tensor containing a sequence of integers.
Description
The function arguments use the same convention as Theano’s arange: if only one argument is
provided, it is in fact the "stop" argument. The default type of the returned tensor is 'int32' to
match TensorFlow’s default.
Usage
k_arange(start, stop = NULL, step = 1, dtype = "int32")

88 k_argmax
Arguments
start Start value.
stop Stop value.
step Difference between two successive values.
dtype Integer dtype to use.
Value
An integer tensor.
Keras Backend
k_argmax Returns the index of the maximum value along an axis.
Description
Returns the index of the maximum value along an axis.
Usage
k_argmax(x, axis = -1)
Arguments
axis Axis along which to perform the reduction (axis indexes are 1-based). Pass -1
(the default) to select the last axis.
Value
A tensor.
Keras Backend
k_argmin 89
k_argmin Returns the index of the minimum value along an axis.
Description
Returns the index of the minimum value along an axis.
Usage
k_argmin(x, axis = -1)
Arguments
axis Axis along which to perform the reduction (axis indexes are 1-based). Pass -1
(the default) to select the last axis.
Value
A tensor.
Keras Backend
k_backend Active Keras backend
Description
Active Keras backend
Usage
k_backend()
Value
The name of the backend Keras is currently using.
90 k_batch_dot
Keras Backend
k_batch_dot Batchwise dot product.
Description
batch_dot is used to compute dot product of x and y when x and y are data in batch, i.e. in a shape
of (batch_size). batch_dot results in a tensor or variable with less dimensions than the input. If
the number of dimensions is reduced to 1, we use expand_dims to make sure that ndim is at least 2.
Usage
k_batch_dot(x, y, axes)
Arguments
x Keras tensor or variable with 2 more more axes.

y Keras tensor or variable with 2 or more axes
axes List of (or single) integer with target dimensions (axis indexes are 1-based). The
lengths of axes[[1]] and axes[[2]] should be the same.
Value
A tensor with shape equal to the concatenation of x’s shape (less the dimension that was summed
over) and y’s shape (less the batch dimension and the dimension that was summed over). If the final
rank is 1, we reshape it to (batch_size, 1).
Keras Backend
k_batch_flatten 91
k_batch_flatten Turn a nD tensor into a 2D tensor with same 1st dimension.
Description
In other words, it flattens each data samples of a batch.
Usage
k_batch_flatten(x)
Arguments
x A tensor or variable.
Value
A tensor.
Keras Backend
k_batch_get_value Returns the value of more than one tensor variable.
Description
Returns the value of more than one tensor variable.
Usage
k_batch_get_value(ops)
Arguments
ops List of ops to evaluate.
Value
A list of arrays.
92 k_batch_normalization
Keras Backend
See Also
k_batch_set_value()
k_batch_normalization Applies batch normalization on x given mean, var, beta and gamma.
Description
i.e. returns output <- (x - mean) / (sqrt(var) + epsilon) * gamma + beta
Usage
k_batch_normalization(x, mean, var, beta, gamma, axis = -1,
epsilon = 0.001)
Arguments
x Input tensor or variable.
mean Mean of batch.
var Variance of batch.
beta Tensor with which to center the input.
gamma Tensor by which to scale the input.
axis Axis (axis indexes are 1-based). Pass -1 (the default) to select the last axis.
epsilon Fuzz factor.
Value
A tensor.
Keras Backend
k_batch_set_value 93
k_batch_set_value Sets the values of many tensor variables at once.
Description
Sets the values of many tensor variables at once.
Usage
k_batch_set_value(lists)
Arguments
lists a list of lists (tensor, value). value should be an R array.
Keras Backend
See Also
k_batch_get_value()
k_bias_add Adds a bias vector to a tensor.
Description
Adds a bias vector to a tensor.
Usage
k_bias_add(x, bias, data_format = NULL)
Arguments
bias Bias tensor to add.
data_format string, "channels_last" or "channels_first".
Value
Output tensor.
94 k_binary_crossentropy
Keras Backend
k_binary_crossentropy Binary crossentropy between an output tensor and a target tensor.
Description
Binary crossentropy between an output tensor and a target tensor.
Usage
k_binary_crossentropy(target, output, from_logits = FALSE)
Arguments
target A tensor with the same shape as output.

output A tensor.
from_logits Whether output is expected to be a logits tensor. By default, we consider that
output encodes a probability distribution.
Value
A tensor.
Keras Backend
k_cast 95
k_cast Casts a tensor to a different dtype and returns it.
Description
You can cast a Keras variable but it still returns a Keras tensor.
Usage
k_cast(x, dtype)
Arguments
x Keras tensor (or variable).
dtype String, either ('float16', 'float32', or 'float64').
Value
Keras tensor with dtype dtype.
Keras Backend
k_cast_to_floatx Cast an array to the default Keras float type.
Description
Cast an array to the default Keras float type.
Usage
k_cast_to_floatx(x)
Arguments
x Array.
Value
The same array, cast to its new type.
96 k_categorical_crossentropy
Keras Backend
k_categorical_crossentropy
Categorical crossentropy between an output tensor and a target tensor.
Description
Categorical crossentropy between an output tensor and a target tensor.
Usage
k_categorical_crossentropy(target, output, from_logits = FALSE,

axis = -1)
Arguments
target A tensor of the same shape as output.

output A tensor resulting from a softmax (unless from_logits is TRUE, in which case
output is expected to be the logits).
from_logits Logical, whether output is the result of a softmax, or is a tensor of logits.
Value
Output tensor.
Keras Backend
k_clear_session 97
k_clear_session Destroys the current TF graph and creates a new one.
Description
Useful to avoid clutter from old models / layers.
Usage
k_clear_session()
Keras Backend
k_clip Element-wise value clipping.
Description
Element-wise value clipping.
Usage
k_clip(x, min_value, max_value)
Arguments
min_value Float or integer.
max_value Float or integer.
Value
A tensor.
Keras Backend
98 k_constant
k_concatenate Concatenates a list of tensors alongside the specified axis.
Description
Concatenates a list of tensors alongside the specified axis.
Usage
k_concatenate(tensors, axis = -1)
Arguments
tensors list of tensors to concatenate.
axis concatenation axis (axis indexes are 1-based). Pass -1 (the default) to select the
last axis.
Value
A tensor.
Keras Backend
k_constant Creates a constant tensor.
Description
Creates a constant tensor.
Usage
k_constant(value, dtype = NULL, shape = NULL, name = NULL)
Arguments
value A constant value
dtype The type of the elements of the resulting tensor.
shape Optional dimensions of resulting tensor.
name Optional name for the tensor.
k_conv1d 99
Value
A Constant Tensor.
Keras Backend
k_conv1d 1D convolution.
Description
1D convolution.
Usage
k_conv1d(x, kernel, strides = 1, padding = "valid",
data_format = NULL, dilation_rate = 1)
Arguments
kernel kernel tensor.
strides stride integer.
padding string, "same", "causal" or "valid".
dilation_rate integer dilate rate.
Value
A tensor, result of 1D convolution.
Keras Backend
100 k_conv2d_transpose
Description
2D convolution.
Usage
k_conv2d(x, kernel, strides = c(1, 1), padding = "valid",
data_format = NULL, dilation_rate = c(1, 1))
Arguments
strides strides
padding string, "same" or "valid".
data_format string, "channels_last" or "channels_first". Whether to use Theano or
TensorFlow/CNTK data format for inputs/kernels/outputs.
dilation_rate vector of 2 integers.
Value
Keras Backend
k_conv2d_transpose 2D deconvolution (i.e. transposed convolution).
Description
2D deconvolution (i.e. transposed convolution).
Usage
k_conv2d_transpose(x, kernel, output_shape, strides = c(1, 1),
padding = "valid", data_format = NULL)
k_conv3d 101
Arguments
output_shape 1D int tensor for the output shape.
strides strides list.
Value
A tensor, result of transposed 2D convolution.
Keras Backend
Description
3D convolution.
Usage
k_conv3d(x, kernel, strides = c(1, 1, 1), padding = "valid",
data_format = NULL, dilation_rate = c(1, 1, 1))
Arguments
strides strides
dilation_rate list of 3 integers.
Value
102 k_conv3d_transpose
Keras Backend
k_conv3d_transpose 3D deconvolution (i.e. transposed convolution).
Description
3D deconvolution (i.e. transposed convolution).
Usage
k_conv3d_transpose(x, kernel, output_shape, strides = c(1, 1, 1),

padding = "valid", data_format = NULL)
Arguments
x input tensor.
output_shape 1D int tensor for the output shape.
strides strides
Value
A tensor, result of transposed 3D convolution.
Keras Backend
k_cos 103
k_cos Computes cos of x element-wise.
Description
Computes cos of x element-wise.
Usage
k_cos(x)
Arguments
Value
A tensor.
Keras Backend
k_count_params Returns the static number of elements in a Keras variable or tensor.
Description
Returns the static number of elements in a Keras variable or tensor.
Usage
k_count_params(x)
Arguments
x Keras variable or tensor.
Value
Integer, the number of elements in x, i.e., the product of the array’s static dimensions.
104 k_ctc_batch_cost
Keras Backend
k_ctc_batch_cost Runs CTC loss algorithm on each batch element.
Description
Runs CTC loss algorithm on each batch element.
Usage
k_ctc_batch_cost(y_true, y_pred, input_length, label_length)
Arguments
y_true tensor (samples, max_string_length) containing the truth labels.

y_pred tensor (samples, time_steps, num_categories) containing the prediction,
or output of the softmax.
input_length tensor (samples, 1) containing the sequence length for each batch item in
y_pred.
label_length tensor (samples, 1) containing the sequence length for each batch item in
y_true.
Value
Tensor with shape (samples,1) containing the CTC loss of each element.
Keras Backend
k_ctc_decode 105
k_ctc_decode Decodes the output of a softmax.
Description
Can use either greedy search (also known as best path) or a constrained dictionary search.
Usage
k_ctc_decode(y_pred, input_length, greedy = TRUE, beam_width = 100L,

top_paths = 1)
Arguments
y_pred tensor (samples, time_steps, num_categories) containing the prediction,

or output of the softmax.
input_length tensor (samples, ) containing the sequence length for each batch item in
y_pred.
greedy perform much faster best-path search if TRUE. This does not use a dictionary.
beam_width if greedy is FALSE: a beam search decoder will be used with a beam of this
width.
top_paths if greedy is FALSE, how many of the most probable paths will be returned.
Value
If greedy is TRUE, returns a list of one element that contains the decoded sequence. If FALSE,
returns the top_paths most probable decoded sequences. Important: blank labels are returned as
-1. Tensor (top_paths) that contains the log probability of each decoded sequence.
Keras Backend
106 k_cumprod
k_ctc_label_dense_to_sparse
Converts CTC labels from dense to sparse.
Description
Converts CTC labels from dense to sparse.
Usage
k_ctc_label_dense_to_sparse(labels, label_lengths)
Arguments
labels dense CTC labels.
label_lengths length of the labels.
Value
A sparse tensor representation of the labels.
Keras Backend
k_cumprod Cumulative product of the values in a tensor, alongside the specified

axis.
Description
Cumulative product of the values in a tensor, alongside the specified axis.
Usage
k_cumprod(x, axis = 1)
Arguments
axis An integer, the axis to compute the product (axis indexes are 1-based).
k_cumsum 107
Value
A tensor of the cumulative product of values of x along axis.
Keras Backend
k_cumsum Cumulative sum of the values in a tensor, alongside the specified axis.
Description
Cumulative sum of the values in a tensor, alongside the specified axis.
Usage
k_cumsum(x, axis = 1)
Arguments
axis An integer, the axis to compute the sum (axis indexes are 1-based).
Value
A tensor of the cumulative sum of values of x along axis.
Keras Backend
108 k_dot
k_depthwise_conv2d Depthwise 2D convolution with separable filters.
Description
Depthwise 2D convolution with separable filters.
Usage
k_depthwise_conv2d(x, depthwise_kernel, strides = c(1, 1),
padding = "valid", data_format = NULL, dilation_rate = c(1, 1))
Arguments
x input tensor
depthwise_kernel
convolution kernel for the depthwise convolution.
strides strides (length 2).
dilation_rate vector of integers, dilation rates for the separable convolution.
Value
Output tensor.
Keras Backend
k_dot Multiplies 2 tensors (and/or variables) and returns a tensor.
Description
When attempting to multiply a nD tensor with a nD tensor, it reproduces the Theano behavior. (e.g.
(2, 3) * (4, 3, 5) -> (2, 4, 5))
Usage
k_dot(x, y)
k_dropout 109
Arguments
y Tensor or variable.
Value
A tensor, dot product of x and y.
Keras Backend
k_dropout Sets entries in x to zero at random, while scaling the entire tensor.
Description
Sets entries in x to zero at random, while scaling the entire tensor.
Usage
k_dropout(x, level, noise_shape = NULL, seed = NULL)
Arguments
x tensor
level fraction of the entries in the tensor that will be set to 0.
noise_shape shape for randomly generated keep/drop flags, must be broadcastable to the
shape of x
seed random seed to ensure determinism.
Value
A tensor.
Keras Backend
110 k_elu
k_dtype Returns the dtype of a Keras tensor or variable, as a string.
Description
Returns the dtype of a Keras tensor or variable, as a string.
Usage
k_dtype(x)
Arguments
Value
String, dtype of x.
Keras Backend
k_elu Exponential linear unit.
Description
Exponential linear unit.
Usage
k_elu(x, alpha = 1)
Arguments
x A tensor or variable to compute the activation function for.
alpha A scalar, slope of negative section.
Value
A tensor.
k_epsilon 111
Keras Backend
k_epsilon Fuzz factor used in numeric expressions.
Description
Fuzz factor used in numeric expressions.
Usage
k_epsilon()
k_set_epsilon(e)
Arguments
e float. New value of epsilon.
Keras Backend
k_equal Element-wise equality between two tensors.
Description
Element-wise equality between two tensors.
Usage
k_equal(x, y)
Arguments
112 k_eval
Value
A bool tensor.
Keras Backend
k_eval Evaluates the value of a variable.
Description
Evaluates the value of a variable.
Usage
k_eval(x)
Arguments
x A variable.
Value
An R array.
Keras Backend
k_exp 113
k_exp Element-wise exponential.
Description
Element-wise exponential.
Usage
k_exp(x)
Arguments
Value
A tensor.
Keras Backend
k_expand_dims Adds a 1-sized dimension at index axis.
Description
Adds a 1-sized dimension at index axis.
Usage
k_expand_dims(x, axis = -1)
Arguments
axis Position where to add a new axis (axis indexes are 1-based). Pass -1 (the default)
to select the last axis.
Value
A tensor with expanded dimensions.
114 k_eye
Keras Backend
k_eye Instantiate an identity matrix and returns it.
Description
Instantiate an identity matrix and returns it.
Usage
k_eye(size, dtype = NULL, name = NULL)
Arguments
size Integer, number of rows/columns.

dtype String, data type of returned Keras variable.
name String, name of returned Keras variable.
Value
A Keras variable, an identity matrix.
Keras Backend
k_flatten 115
k_flatten Flatten a tensor.
Description
Flatten a tensor.
Usage
k_flatten(x)
Arguments
Value
A tensor, reshaped into 1-D
Keras Backend
k_floatx Default float type
Description
Default float type
Usage
k_floatx()
k_set_floatx(floatx)
Arguments
floatx String, ’float16’, ’float32’, or ’float64’.
116 k_foldl
Keras Backend
k_foldl Reduce elems using fn to combine them from left to right.
Description
Reduce elems using fn to combine them from left to right.
Usage
k_foldl(fn, elems, initializer = NULL, name = NULL)
Arguments
fn Function that will be called upon each element in elems and an accumulator
elems tensor
initializer The first value used (first element of elems in case of ‘NULL“)
name A string name for the foldl node in the graph
Value
Tensor with same type and shape as initializer.
Keras Backend
k_foldr 117
k_foldr Reduce elems using fn to combine them from right to left.
Description
Reduce elems using fn to combine them from right to left.
Usage
k_foldr(fn, elems, initializer = NULL, name = NULL)
Arguments
fn Function that will be called upon each element in elems and an accumulator
elems tensor
initializer The first value used (last element of elems in case of NULL)
name A string name for the foldr node in the graph
Value
Tensor with same type and shape as initializer.
Keras Backend
k_function Instantiates a Keras function
Description
Instantiates a Keras function
Usage
k_function(inputs, outputs, updates = NULL, ...)
Arguments
inputs List of placeholder tensors.
outputs List of output tensors.
updates List of update ops.
... Named arguments passed to tf$Session$run.
118 k_gather
Value
Output values as R arrays.
Keras Backend
k_gather Retrieves the elements of indices indices in the tensor reference.
Description
Retrieves the elements of indices indices in the tensor reference.
Usage
k_gather(reference, indices)
Arguments
reference A tensor.
indices Indices. Dimension indices are 1-based. Note however that if you pass a tensor
for indices they will be passed as-is, in which case indices will be 0 based
because no normalizing of R 1-based axes to Python 0-based axes is performed.
Value
A tensor of same type as reference.
Keras Backend
k_get_session 119
k_get_session TF session to be used by the backend.
Description
If a default TensorFlow session is available, we will return it. Else, we will return the global Keras
session. If no global Keras session exists at this point: we will create a new global session. Note
that you can manually set the global session via k_set_session().
Usage
k_get_session()
k_set_session(session)
Arguments
session A TensorFlow Session.
Value
A TensorFlow session
Keras Backend
k_get_uid Get the uid for the default graph.
Description
Get the uid for the default graph.
Usage
k_get_uid(prefix = "")
Arguments
prefix An optional prefix of the graph.
120 k_get_value
Value
A unique identifier for the graph.
Keras Backend
k_get_value Returns the value of a variable.
Description
Returns the value of a variable.
Usage
k_get_value(x)
Arguments
x input variable.
Value
An R array.
Keras Backend
k_get_variable_shape 121
k_get_variable_shape Returns the shape of a variable.
Description
Returns the shape of a variable.
Usage
k_get_variable_shape(x)
Arguments
x A variable.
Value
A vector of integers.
Keras Backend
k_gradients Returns the gradients of variables w.r.t. loss.
Description
Returns the gradients of variables w.r.t. loss.
Usage
k_gradients(loss, variables)
Arguments
loss Scalar tensor to minimize.
variables List of variables.
Value
A gradients tensor.
122 k_greater_equal
Keras Backend
k_greater Element-wise truth value of (x > y).
Description
Element-wise truth value of (x > y).
Usage
k_greater(x, y)
Arguments
Value
A bool tensor.
Keras Backend
k_greater_equal Element-wise truth value of (x >= y).
Description
Element-wise truth value of (x >= y).
Usage
k_greater_equal(x, y)
k_hard_sigmoid 123
Arguments
Value
A bool tensor.
Keras Backend
k_hard_sigmoid Segment-wise linear approximation of sigmoid.
Description
Faster than sigmoid. Returns 0. if x < -2.5, 1. if x > 2.5. In -2.5 <= x <= 2.5, returns
0.2 * x + 0.5.
Usage
k_hard_sigmoid(x)
Arguments
Value
A tensor.
Keras Backend
124 k_image_data_format
k_identity Returns a tensor with the same content as the input tensor.
Description
Returns a tensor with the same content as the input tensor.
Usage
k_identity(x, name = NULL)
Arguments
x The input tensor.
name String, name for the variable to create.
Value
A tensor of the same shape, type and content.
Keras Backend
k_image_data_format Default image data format convention (’channels_first’ or ’chan-

nels_last’).
Description
Default image data format convention (’channels_first’ or ’channels_last’).
Usage
k_image_data_format()
k_set_image_data_format(data_format)
Arguments
data_format string. 'channels_first' or 'channels_last'.
k_int_shape 125
Keras Backend
k_int_shape Returns the shape of tensor or variable as a list of int or NULL entries.
Description
Returns the shape of tensor or variable as a list of int or NULL entries.
Usage
k_int_shape(x)
Arguments
Value
A list of integers (or NULL entries).
Keras Backend
k_in_test_phase Selects x in test phase, and alt otherwise.
Description
Note that alt should have the same shape as x.
Usage
k_in_test_phase(x, alt, training = NULL)
126 k_in_top_k
Arguments
x What to return in test phase (tensor or function that returns a tensor).
alt What to return otherwise (tensor or function that returns a tensor).
training Optional scalar tensor (or R logical or integer) specifying the learning phase.
Value
Either x or alt based on k_learning_phase().
Keras Backend
k_in_top_k Returns whether the targets are in the top k predictions.
Description
Returns whether the targets are in the top k predictions.
Usage
k_in_top_k(predictions, targets, k)
Arguments
predictions A tensor of shape (batch_size, classes) and type float32.
targets A 1D tensor of length batch_size and type int32 or int64.
k An int, number of top elements to consider.
Value
A 1D tensor of length batch_size and type bool. output[[i]] is TRUE if predictions[i, targets[[i]]
is within top-k values of predictions[[i]].
Keras Backend
k_in_train_phase 127
k_in_train_phase Selects x in train phase, and alt otherwise.
Description
Note that alt should have the same shape as x.
Usage
k_in_train_phase(x, alt, training = NULL)
Arguments
x What to return in train phase (tensor or function that returns a tensor).
alt What to return otherwise (tensor or function that returns a tensor).
training Optional scalar tensor (or R logical or integer) specifying the learning phase.
Value
Either x or alt based on the training flag. the training flag defaults to k_learning_phase().
Keras Backend
k_is_keras_tensor Returns whether x is a Keras tensor.
Description
A "Keras tensor" is a tensor that was returned by a Keras layer
Usage
k_is_keras_tensor(x)
Arguments
x A candidate tensor.
Value
A logical: Whether the argument is a Keras tensor.
128 k_is_sparse
Keras Backend
k_is_placeholder Returns whether x is a placeholder.
Description
Returns whether x is a placeholder.
Usage
k_is_placeholder(x)
Arguments
x A candidate placeholder.
Value
A logical
Keras Backend
k_is_sparse Returns whether a tensor is a sparse tensor.
Description
Returns whether a tensor is a sparse tensor.
Usage
k_is_sparse(tensor)
Arguments
tensor A tensor instance.
k_is_tensor 129
Value
A logical
Keras Backend
k_is_tensor Returns whether x is a symbolic tensor.
Description
Returns whether x is a symbolic tensor.
Usage
k_is_tensor(x)
Arguments
x A candidate tensor.
Value
A logical: Whether the argument is a symbolic tensor.
Keras Backend
130 k_learning_phase
k_l2_normalize Normalizes a tensor wrt the L2 norm alongside the specified axis.
Description
Normalizes a tensor wrt the L2 norm alongside the specified axis.
Usage
k_l2_normalize(x, axis = NULL)
Arguments
axis Axis along which to perform normalization (axis indexes are 1-based)
Value
A tensor.
Keras Backend
k_learning_phase Returns the learning phase flag.
Description
The learning phase flag is a bool tensor (0 = test, 1 = train) to be passed as input to any Keras
function that uses a different behavior at train time and test time.
Usage
k_learning_phase()
Value
Learning phase (scalar integer tensor or R integer).
k_less 131
Keras Backend
k_less Element-wise truth value of (x < y).
Description
Element-wise truth value of (x < y).
Usage
k_less(x, y)
Arguments
Value
A bool tensor.
Keras Backend
k_less_equal Element-wise truth value of (x <= y).
Description
Element-wise truth value of (x <= y).
Usage
k_less_equal(x, y)
132 k_local_conv1d
Arguments
Value
A bool tensor.
Keras Backend
k_local_conv1d Apply 1D conv with un-shared weights.
Description
Apply 1D conv with un-shared weights.
Usage
k_local_conv1d(inputs, kernel, kernel_size, strides, data_format = NULL)
Arguments
inputs 3D tensor with shape: (batch_size, steps, input_dim)
kernel the unshared weight for convolution, with shape (output_length, feature_dim,
filters)
kernel_size a list of a single integer, specifying the length of the 1D convolution window
strides a list of a single integer, specifying the stride length of the convolution
data_format the data format, channels_first or channels_last
Value
the tensor after 1d conv with un-shared weights, with shape (batch_size, output_length, filters)
Keras Backend
k_local_conv2d 133
k_local_conv2d Apply 2D conv with un-shared weights.
Description
Apply 2D conv with un-shared weights.
Usage
k_local_conv2d(inputs, kernel, kernel_size, strides, output_shape,

data_format = NULL)
Arguments
inputs 4D tensor with shape: (batch_size, filters, new_rows, new_cols) if data_format=’channels_first’

or 4D tensor with shape: (batch_size, new_rows, new_cols, filters) if data_format=’channels_last’.
kernel the unshared weight for convolution, with shape (output_items, feature_dim,
filters)
kernel_size a list of 2 integers, specifying the width and height of the 2D convolution win-
dow.
strides a list of 2 integers, specifying the strides of the convolution along the width and
height.
output_shape a list with (output_row, output_col)
data_format the data format, channels_first or channels_last
Value
A 4d tensor with shape: (batch_size, filters, new_rows, new_cols) if data_format=’channels_first’

or 4D tensor with shape: (batch_size, new_rows, new_cols, filters) if data_format=’channels_last’.
Keras Backend
134 k_logsumexp
k_log Element-wise log.
Description
Element-wise log.
Usage
k_log(x)
Arguments
Value
A tensor.
Keras Backend
k_logsumexp Computes log(sum(exp(elements across dimensions of a tensor))).
Description
This function is more numerically stable than log(sum(exp(x))). It avoids overflows caused by
taking the exp of large inputs and underflows caused by taking the log of small inputs.
Usage
k_logsumexp(x, axis = NULL, keepdims = FALSE)
Arguments
axis An integer, the axis to reduce over (axis indexes are 1-based).
keepdims A boolean, whether to keep the dimensions or not. If keepdims is FALSE, the
rank of the tensor is reduced by 1. If keepdims is TRUE, the reduced dimension
is retained with length 1.
k_manual_variable_initialization 135
Value
The reduced tensor.
Keras Backend
k_manual_variable_initialization
Sets the manual variable initialization flag.
Description
This boolean flag determines whether variables should be initialized as they are instantiated (de-
fault), or if the user should handle the initialization (e.g. via tf$initialize_all_variables()).
Usage
k_manual_variable_initialization(value)
Arguments
value Logical
Keras Backend
136 k_max
k_map_fn Map the function fn over the elements elems and return the outputs.
Description
Map the function fn over the elements elems and return the outputs.
Usage
k_map_fn(fn, elems, name = NULL, dtype = NULL)
Arguments
fn Function that will be called upon each element in elems

elems tensor
name A string name for the map node in the graph
dtype Output data type.
Value
Tensor with dtype dtype.
Keras Backend
k_max Maximum value in a tensor.
Description
Maximum value in a tensor.
Usage
k_max(x, axis = NULL, keepdims = FALSE)

k_maximum 137
Arguments
axis An integer, the axis to find maximum values (axis indexes are 1-based).
Value
A tensor with maximum values of x.
Keras Backend
k_maximum Element-wise maximum of two tensors.
Description
Element-wise maximum of two tensors.
Usage
k_maximum(x, y)
Arguments
Value
A tensor.
Keras Backend
138 k_min
k_mean Mean of a tensor, alongside the specified axis.
Description
Mean of a tensor, alongside the specified axis.
Usage
k_mean(x, axis = NULL, keepdims = FALSE)
Arguments
axis A list of axes to compute the mean over (axis indexes are 1-based).
rank of the tensor is reduced by 1 for each entry in axis. If keep_dims is TRUE,
the reduced dimensions are retained with length 1.
Value
A tensor with the mean of elements of x.
Keras Backend
k_min Minimum value in a tensor.
Description
Minimum value in a tensor.
Usage
k_min(x, axis = NULL, keepdims = FALSE)

k_minimum 139
Arguments
axis An integer, axis to find minimum values (axis indexes are 1-based).
Value
A tensor with miminum values of x.
Keras Backend
k_minimum Element-wise minimum of two tensors.
Description
Element-wise minimum of two tensors.
Usage
k_minimum(x, y)
Arguments
Value
A tensor.
Keras Backend
140 k_ndim
k_moving_average_update
Compute the moving average of a variable.
Description
Compute the moving average of a variable.
Usage
k_moving_average_update(x, value, momentum)
Arguments
x A Variable.
value A tensor with the same shape as x.
momentum The moving average momentum.
Value
An operation to update the variable.
Keras Backend
k_ndim Returns the number of axes in a tensor, as an integer.
Description
Returns the number of axes in a tensor, as an integer.
Usage
k_ndim(x)
Arguments
k_normalize_batch_in_training 141
Value
Integer (scalar), number of axes.
Keras Backend
k_normalize_batch_in_training
Computes mean and std for batch then apply batch_normalization on
batch.
Description
Computes mean and std for batch then apply batch_normalization on batch.
Usage
k_normalize_batch_in_training(x, gamma, beta, reduction_axes,
epsilon = 0.001)
Arguments
x Input tensor or variable.
gamma Tensor by which to scale the input.
beta Tensor with which to center the input.
reduction_axes iterable of integers, axes over which to normalize.
epsilon Fuzz factor.
Value
A list length of 3, (normalized_tensor, mean, variance).
Keras Backend
142 k_ones
k_not_equal Element-wise inequality between two tensors.
Description
Element-wise inequality between two tensors.
Usage
k_not_equal(x, y)
Arguments
Value
A bool tensor.
Keras Backend
k_ones Instantiates an all-ones tensor variable and returns it.
Description
Instantiates an all-ones tensor variable and returns it.
Usage
k_ones(shape, dtype = NULL, name = NULL)
Arguments
shape Tuple of integers, shape of returned Keras variable.
dtype String, data type of returned Keras variable.
k_ones_like 143
Value
A Keras variable, filled with 1.0.
Keras Backend
k_ones_like Instantiates an all-ones variable of the same shape as another tensor.
Description
Instantiates an all-ones variable of the same shape as another tensor.
Usage
k_ones_like(x, dtype = NULL, name = NULL)
Arguments
x Keras variable or tensor.

dtype String, dtype of returned Keras variable. NULL uses the dtype of x.
Value
A Keras variable with the shape of x filled with ones.
Keras Backend
144 k_permute_dimensions
k_one_hot Computes the one-hot representation of an integer tensor.
Description
Computes the one-hot representation of an integer tensor.
Usage
k_one_hot(indices, num_classes)
Arguments
indices nD integer tensor of shape (batch_size, dim1, dim2, ... dim(n-1))
num_classes Integer, number of classes to consider.
Value
(n + 1)D one hot representation of the input with shape (batch_size, dim1, dim2, ... dim(n-1), num_classes)
Keras Backend
k_permute_dimensions Permutes axes in a tensor.
Description
Permutes axes in a tensor.
Usage
k_permute_dimensions(x, pattern)
Arguments
pattern A list of dimension indices, e.g. (1, 3, 2). Dimension indices are 1-based.
Value
A tensor.
k_placeholder 145
Keras Backend
k_placeholder Instantiates a placeholder tensor and returns it.
Description
Instantiates a placeholder tensor and returns it.
Usage
k_placeholder(shape = NULL, ndim = NULL, dtype = NULL,

sparse = FALSE, name = NULL)
Arguments
shape Shape of the placeholder (integer list, may include NULL entries).
ndim Number of axes of the tensor. At least one of shape, ndim must be specified. If
both are specified, shape is used.
dtype Placeholder type.
sparse Logical, whether the placeholder should have a sparse type.
name Optional name string for the placeholder.
Value
Tensor instance (with Keras metadata included).
Keras Backend
146 k_pool3d
k_pool2d 2D Pooling.
Description
2D Pooling.
Usage
k_pool2d(x, pool_size, strides = c(1, 1), padding = "valid",
data_format = NULL, pool_mode = "max")
Arguments
pool_size list of 2 integers.
strides list of 2 integers.
pool_mode string, "max" or "avg".
Value
A tensor, result of 2D pooling.
Keras Backend
k_pool3d 3D Pooling.
Description
3D Pooling.
Usage
k_pool3d(x, pool_size, strides = c(1, 1, 1), padding = "valid",
data_format = NULL, pool_mode = "max")
k_pow 147
Arguments
pool_size list of 3 integers.
strides list of 3 integers.
pool_mode string, "max" or "avg".
Value
A tensor, result of 3D pooling.
Keras Backend
k_pow Element-wise exponentiation.
Description
Element-wise exponentiation.
Usage
k_pow(x, a)
Arguments
a R integer.
Value
A tensor.
Keras Backend
148 k_prod
k_print_tensor Prints message and the tensor value when evaluated.
Description
Note that print_tensor returns a new tensor identical to x which should be used in the following
code. Otherwise the print operation is not taken into account during evaluation.
Usage
k_print_tensor(x, message = "")
Arguments
x Tensor to print.
message Message to print jointly with the tensor.
Value
The same tensor x, unchanged.
Keras Backend
k_prod Multiplies the values in a tensor, alongside the specified axis.
Description
Multiplies the values in a tensor, alongside the specified axis.
Usage
k_prod(x, axis = NULL, keepdims = FALSE)
Arguments
axis An integer, axis to compute the product over (axis indexes are 1-based).
k_random_binomial 149
Value
A tensor with the product of elements of x.
Keras Backend
k_random_binomial Returns a tensor with random binomial distribution of values.
Description
Returns a tensor with random binomial distribution of values.
Usage
k_random_binomial(shape, p = 0, dtype = NULL, seed = NULL)
Arguments
shape A list of integers, the shape of tensor to create.

p A float, 0. <= p <= 1, probability of binomial distribution.
dtype String, dtype of returned tensor.
seed Integer, random seed.
Value
A tensor.
Keras Backend
150 k_random_normal_variable
k_random_normal Returns a tensor with normal distribution of values.
Description
Returns a tensor with normal distribution of values.
Usage
k_random_normal(shape, mean = 0, stddev = 1, dtype = NULL,
seed = NULL)
Arguments
mean A float, mean of the normal distribution to draw samples.
stddev A float, standard deviation of the normal distribution to draw samples.
Value
A tensor.
Keras Backend
k_random_normal_variable
Instantiates a variable with values drawn from a normal distribution.
Description
Instantiates a variable with values drawn from a normal distribution.
Usage
k_random_normal_variable(shape, mean, scale, dtype = NULL, name = NULL,
seed = NULL)
k_random_uniform 151
Arguments
mean Float, mean of the normal distribution.
scale Float, standard deviation of the normal distribution.
dtype String, dtype of returned Keras variable.
Value
A Keras variable, filled with drawn samples.
Keras Backend
k_random_uniform Returns a tensor with uniform distribution of values.
Description
Returns a tensor with uniform distribution of values.
Usage
k_random_uniform(shape, minval = 0, maxval = 1, dtype = NULL,
seed = NULL)
Arguments
minval A float, lower boundary of the uniform distribution to draw samples.
maxval A float, upper boundary of the uniform distribution to draw samples.
Value
A tensor.
152 k_random_uniform_variable
Keras Backend
k_random_uniform_variable
Instantiates a variable with values drawn from a uniform distribution.
Description
Instantiates a variable with values drawn from a uniform distribution.
Usage
k_random_uniform_variable(shape, low, high, dtype = NULL, name = NULL,

seed = NULL)
Arguments

low Float, lower boundary of the output interval.
high Float, upper boundary of the output interval.
dtype String, dtype of returned Keras variable.
Value
A Keras variable, filled with drawn samples.
Keras Backend
k_relu 153
k_relu Rectified linear unit.
Description
With default values, it returns element-wise max(x, 0).
Usage
k_relu(x, alpha = 0, max_value = NULL)
Arguments
alpha A scalar, slope of negative section (default=0.).
max_value Saturation threshold.
Value
A tensor.
Keras Backend
k_repeat Repeats a 2D tensor.
Description
If x has shape (samples, dim) and n is 2, the output will have shape (samples, 2, dim).
Usage
k_repeat(x, n)
Arguments
n Integer, number of times to repeat.
154 k_repeat_elements
Value
A tensor
Keras Backend
k_repeat_elements Repeats the elements of a tensor along an axis.
Description
If x has shape (s1, s2, s3) and axis is 2, the output will have shape (s1, s2 * rep, s3).
Usage
k_repeat_elements(x, rep, axis)
Arguments
rep Integer, number of times to repeat.
axis Axis along which to repeat (axis indexes are 1-based)
Value
A tensor.
Keras Backend
k_reset_uids 155
k_reset_uids Reset graph identifiers.
Description
Reset graph identifiers.
Usage
k_reset_uids()
Keras Backend
k_reshape Reshapes a tensor to the specified shape.
Description
Reshapes a tensor to the specified shape.
Usage
k_reshape(x, shape)
Arguments
shape Target shape list.
Value
A tensor.
Keras Backend
156 k_resize_volumes
k_resize_images Resizes the images contained in a 4D tensor.
Description
Resizes the images contained in a 4D tensor.
Usage
k_resize_images(x, height_factor, width_factor, data_format)
Arguments
x Tensor or variable to resize.

height_factor Positive integer.
width_factor Positive integer.
Value
A tensor.
Keras Backend
k_resize_volumes Resizes the volume contained in a 5D tensor.
Description
Resizes the volume contained in a 5D tensor.
Usage
k_resize_volumes(x, depth_factor, height_factor, width_factor, data_format)

k_reverse 157
Arguments
x Tensor or variable to resize.
depth_factor Positive integer.
height_factor Positive integer.
width_factor Positive integer.
Value
A tensor.
Keras Backend
k_reverse Reverse a tensor along the specified axes.
Description
Reverse a tensor along the specified axes.
Usage
k_reverse(x, axes)
Arguments
x Tensor to reverse.
axes Integer or list of integers of axes to reverse (axis indexes are 1-based).
Value
A tensor.
Keras Backend
158 k_rnn
k_rnn Iterates over the time dimension of a tensor
Description
Iterates over the time dimension of a tensor
Usage
k_rnn(step_function, inputs, initial_states, go_backwards = FALSE,
mask = NULL, constants = NULL, unroll = FALSE,
input_length = NULL)
Arguments
step_function RNN step function.
inputs Tensor with shape (samples, ...) (no time dimension), representing input for the
batch of samples at a certain time step.
initial_states Tensor with shape (samples, output_dim) (no time dimension), containing the
initial values for the states used in the step function.
go_backwards Logical If TRUE, do the iteration over the time dimension in reverse order and
return the reversed sequence.
mask Binary tensor with shape (samples, time, 1), with a zero for every element that
is masked.
constants A list of constant values passed at each step.
unroll Whether to unroll the RNN or to use a symbolic loop (while_loop or scan de-
pending on backend).
input_length Not relevant in the TensorFlow implementation. Must be specified if using un-
rolling with Theano.
Value
A list with:
• last_output: the latest output of the rnn, of shape (samples, ...)
• outputs: tensor with shape (samples, time, ...) where each entry outputs[s, t] is the output
of the step function at time t for sample s.
• new_states: list of tensors, latest states returned by the step function, of shape (samples, ...).
Keras Backend
k_round 159
k_round Element-wise rounding to the closest integer.
Description
In case of tie, the rounding mode used is "half to even".
Usage
k_round(x)
Arguments
Value
A tensor.
Keras Backend
k_separable_conv2d 2D convolution with separable filters.
Description
2D convolution with separable filters.
Usage
k_separable_conv2d(x, depthwise_kernel, pointwise_kernel, strides = c(1,

1), padding = "valid", data_format = NULL, dilation_rate = c(1, 1))
160 k_set_learning_phase
Arguments
x input tensor
depthwise_kernel
convolution kernel for the depthwise convolution.
pointwise_kernel
kernel for the 1x1 convolution.
strides strides list (length 2).
dilation_rate list of integers, dilation rates for the separable convolution.
Value
Output tensor.
Keras Backend
k_set_learning_phase Sets the learning phase to a fixed value.
Description
Sets the learning phase to a fixed value.
Usage
k_set_learning_phase(value)
Arguments
value Learning phase value, either 0 or 1 (integers).
Keras Backend
k_set_value 161
k_set_value Sets the value of a variable, from an R array.
Description
Sets the value of a variable, from an R array.
Usage
k_set_value(x, value)
Arguments
x Tensor to set to a new value.
value Value to set the tensor to, as an R array (of the same shape).
Keras Backend
k_shape Returns the symbolic shape of a tensor or variable.
Description
Returns the symbolic shape of a tensor or variable.
Usage
k_shape(x)
Arguments
Value
A symbolic shape (which is itself a tensor).
162 k_sign
Keras Backend
k_sigmoid Element-wise sigmoid.
Description
Element-wise sigmoid.
Usage
k_sigmoid(x)
Arguments
Value
A tensor.
Keras Backend
k_sign Element-wise sign.
Description
Element-wise sign.
Usage
k_sign(x)
Arguments
k_sin 163
Value
A tensor.
Keras Backend
k_sin Computes sin of x element-wise.
Description
Computes sin of x element-wise.
Usage
k_sin(x)
Arguments
Value
A tensor.
Keras Backend
164 k_softplus
k_softmax Softmax of a tensor.
Description
Softmax of a tensor.
Usage
k_softmax(x, axis = -1)
Arguments
axis The dimension softmax would be performed on. The default is -1 which indi-
cates the last dimension.
Value
A tensor.
Keras Backend
k_softplus Softplus of a tensor.
Description
Softplus of a tensor.
Usage
k_softplus(x)
Arguments
Value
A tensor.
k_softsign 165
Keras Backend
k_softsign Softsign of a tensor.
Description
Softsign of a tensor.
Usage
k_softsign(x)
Arguments
Value
A tensor.
Keras Backend
k_sparse_categorical_crossentropy
Categorical crossentropy with integer targets.
Description
Categorical crossentropy with integer targets.
Usage
k_sparse_categorical_crossentropy(target, output, from_logits = FALSE,
axis = -1)
166 k_spatial_2d_padding
Arguments
target An integer tensor.
output A tensor resulting from a softmax (unless from_logits is TRUE, in which case
output is expected to be the logits).
from_logits Boolean, whether output is the result of a softmax, or is a tensor of logits.
Value
Output tensor.
Keras Backend
k_spatial_2d_padding Pads the 2nd and 3rd dimensions of a 4D tensor.
Description
Pads the 2nd and 3rd dimensions of a 4D tensor.
Usage
k_spatial_2d_padding(x, padding = list(list(1, 1), list(1, 1)),
data_format = NULL)
Arguments
padding Tuple of 2 lists, padding pattern.
Value
A padded 4D tensor.
Keras Backend
k_spatial_3d_padding 167
k_spatial_3d_padding Pads 5D tensor with zeros along the depth, height, width dimensions.
Description
Pads these dimensions with respectively padding[[1]], padding[[2]], and padding[[3]] zeros
left and right. For ’channels_last’ data_format, the 2nd, 3rd and 4th dimension will be padded. For
’channels_first’ data_format, the 3rd, 4th and 5th dimension will be padded.
Usage
k_spatial_3d_padding(x, padding = list(list(1, 1), list(1, 1), list(1,
1)), data_format = NULL)
Arguments
padding List of 3 lists, padding pattern.
Value
A padded 5D tensor.
Keras Backend
k_sqrt Element-wise square root.
Description
Element-wise square root.
Usage
k_sqrt(x)
Arguments
168 k_square
Value
A tensor.
Keras Backend
k_square Element-wise square.
Description
Element-wise square.
Usage
k_square(x)
Arguments
Value
A tensor.
Keras Backend
k_squeeze 169
k_squeeze Removes a 1-dimension from the tensor at index axis.
Description
Removes a 1-dimension from the tensor at index axis.
Usage
k_squeeze(x, axis)
Arguments
axis Axis to drop (axis indexes are 1-based).
Value
A tensor with the same data as x but reduced dimensions.
Keras Backend
k_stack Stacks a list of rank R tensors into a rank R+1 tensor.
Description
Stacks a list of rank R tensors into a rank R+1 tensor.
Usage
k_stack(x, axis = 1)
Arguments
x List of tensors.
axis Axis along which to perform stacking (axis indexes are 1-based).
Value
A tensor.
170 k_std
Keras Backend
k_std Standard deviation of a tensor, alongside the specified axis.
Description
Standard deviation of a tensor, alongside the specified axis.
Usage
k_std(x, axis = NULL, keepdims = FALSE)
Arguments
axis An integer, the axis to compute the standard deviation over (axis indexes are
1-based).
Value
A tensor with the standard deviation of elements of x.
Keras Backend
k_stop_gradient 171
k_stop_gradient Returns variables but with zero gradient w.r.t. every other variable.
Description
Returns variables but with zero gradient w.r.t. every other variable.
Usage
k_stop_gradient(variables)
Arguments
variables tensor or list of tensors to consider constant with respect to any other variable.
Value
A single tensor or a list of tensors (depending on the passed argument) that has constant gradient
with respect to any other variable.
Keras Backend
k_sum Sum of the values in a tensor, alongside the specified axis.
Description
Sum of the values in a tensor, alongside the specified axis.
Usage
k_sum(x, axis = NULL, keepdims = FALSE)
Arguments
axis An integer, the axis to sum over (axis indexes are 1-based).
172 k_switch
Value
A tensor with sum of x.
Keras Backend
k_switch Switches between two operations depending on a scalar value.
Description
Note that both then_expression and else_expression should be symbolic tensors of the same
shape.
Usage
k_switch(condition, then_expression, else_expression)
Arguments
condition tensor (int or bool).

then_expression
either a tensor, or a function that returns a tensor.
else_expression
either a tensor, or a function that returns a tensor.
Value
The selected tensor.
Keras Backend
k_tanh 173
k_tanh Element-wise tanh.
Description
Element-wise tanh.
Usage
k_tanh(x)
Arguments
Value
A tensor.
Keras Backend
k_temporal_padding Pads the middle dimension of a 3D tensor.
Description
Pads the middle dimension of a 3D tensor.
Usage
k_temporal_padding(x, padding = c(1, 1))
Arguments
padding List of 2 integers, how many zeros to add at the start and end of dim 1.
Value
A padded 3D tensor.
174 k_tile
Keras Backend
k_tile Creates a tensor by tiling x by n.
Description
Creates a tensor by tiling x by n.
Usage
k_tile(x, n)
Arguments
x A tensor or variable
n A list of integers. The length must be the same as the number of dimensions in
x.
Value
A tiled tensor.
Keras Backend
k_to_dense 175
k_to_dense Converts a sparse tensor into a dense tensor and returns it.
Description
Converts a sparse tensor into a dense tensor and returns it.
Usage
k_to_dense(tensor)
Arguments
tensor A tensor instance (potentially sparse).
Value
A dense tensor.
Keras Backend
k_transpose Transposes a tensor and returns it.
Description
Transposes a tensor and returns it.
Usage
k_transpose(x)
Arguments
Value
A tensor.
176 k_truncated_normal
Keras Backend
k_truncated_normal Returns a tensor with truncated random normal distribution of values.
Description
The generated values follow a normal distribution with specified mean and standard deviation, ex-
cept that values whose magnitude is more than two standard deviations from the mean are dropped
and re-picked.
Usage
k_truncated_normal(shape, mean = 0, stddev = 1, dtype = NULL,

seed = NULL)
Arguments

mean Mean of the values.
stddev Standard deviation of the values.
Value
A tensor.
Keras Backend
k_update 177
k_update Update the value of x to new_x.
Description
Update the value of x to new_x.
Usage
k_update(x, new_x)
Arguments
x A Variable.
new_x A tensor of same shape as x.
Value
The variable x updated.
Keras Backend
k_update_add Update the value of x by adding increment.
Description
Update the value of x by adding increment.
Usage
k_update_add(x, increment)
Arguments
x A Variable.
increment A tensor of same shape as x.
Value
178 k_var
Keras Backend
k_update_sub Update the value of x by subtracting decrement.
Description
Update the value of x by subtracting decrement.
Usage
k_update_sub(x, decrement)
Arguments
x A Variable.
decrement A tensor of same shape as x.
Value
Keras Backend
k_var Variance of a tensor, alongside the specified axis.
Description
Variance of a tensor, alongside the specified axis.
Usage
k_var(x, axis = NULL, keepdims = FALSE)
k_variable 179
Arguments
axis An integer, the axis to compute the variance over (axis indexes are 1-based).
Value
A tensor with the variance of elements of x.
Keras Backend
k_variable Instantiates a variable and returns it.
Description
Instantiates a variable and returns it.
Usage
k_variable(value, dtype = NULL, name = NULL, constraint = NULL)
Arguments
value Numpy array, initial value of the tensor.
dtype Tensor type.
name Optional name string for the tensor.
constraint Optional projection function to be applied to the variable after an optimizer up-
date.
Value
A variable instance (with Keras metadata included).
Keras Backend
180 k_zeros_like
k_zeros Instantiates an all-zeros variable and returns it.
Description
Instantiates an all-zeros variable and returns it.
Usage
k_zeros(shape, dtype = NULL, name = NULL)
Arguments
shape Tuple of integers, shape of returned Keras variable
dtype String, data type of returned Keras variable
name String, name of returned Keras variable
Value
A variable (including Keras metadata), filled with 0.0.
Keras Backend
k_zeros_like Instantiates an all-zeros variable of the same shape as another tensor.
Description
Instantiates an all-zeros variable of the same shape as another tensor.
Usage
k_zeros_like(x, dtype = NULL, name = NULL)
Arguments
x Keras variable or Keras tensor.
dtype String, dtype of returned Keras variable. NULL uses the dtype of x.
layer_activation 181
Value
A Keras variable with the shape of x filled with zeros.
Keras Backend
layer_activation Apply an activation function to an output.
Description
Apply an activation function to an output.
Usage
layer_activation(object, activation, input_shape = NULL,
Arguments
activation Name of activation function to use. If you don’t specify anything, no activation
is applied (ie. "linear" activation: a(x) = x).
input_shape Input shape (list of integers, does not include the samples axis) which is required
when using this layer as the first layer in a model.
batch_input_shape
182 layer_activation_elu
See Also
Other core layers: layer_activity_regularization, layer_dense, layer_dropout, layer_flatten,
layer_input, layer_lambda, layer_masking, layer_permute, layer_repeat_vector, layer_reshape
Other activation layers: layer_activation_elu, layer_activation_leaky_relu, layer_activation_parametric_relu
layer_activation_relu, layer_activation_selu, layer_activation_softmax, layer_activation_thresholded_re
layer_activation_elu Exponential Linear Unit.
Description
It follows: f(x) = alpha * (exp(x) - 1.0) for x < 0, f(x) = x for ‘x
= 0‘.
Usage
layer_activation_elu(object, alpha = 1, input_shape = NULL,
Arguments
alpha Scale for the negative factor.
batch_input_shape
See Also
Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs).
Other activation layers: layer_activation_leaky_relu, layer_activation_parametric_relu,
layer_activation_relu, layer_activation_selu, layer_activation_softmax, layer_activation_thresholded_re
layer_activation
layer_activation_leaky_relu 183
layer_activation_leaky_relu
Leaky version of a Rectified Linear Unit.
Description
Allows a small gradient when the unit is not active: f(x) = alpha * x for x < 0, f(x) = x for
x >= 0.
Usage
layer_activation_leaky_relu(object, alpha = 0.3, input_shape = NULL,

Arguments

alpha float >= 0. Negative slope coefficient.
batch_input_shape
See Also
Rectifier Nonlinearities Improve Neural Network Acoustic Models.

Other activation layers: layer_activation_elu, layer_activation_parametric_relu, layer_activation_relu,
layer_activation_selu, layer_activation_softmax, layer_activation_thresholded_relu,
layer_activation
184 layer_activation_parametric_relu
layer_activation_parametric_relu
Parametric Rectified Linear Unit.
Description
It follows: f(x) = alpha * x`` for x < 0, f(x) = xforx >= 0‘, where alpha is a learned array
with the same shape as x.
Usage
layer_activation_parametric_relu(object, alpha_initializer = "zeros",
alpha_regularizer = NULL, alpha_constraint = NULL,
shared_axes = NULL, input_shape = NULL, batch_input_shape = NULL,
batch_size = NULL, dtype = NULL, name = NULL, trainable = NULL,
weights = NULL)
Arguments
alpha_initializer
Initializer function for the weights.
alpha_regularizer
Regularizer for the weights.
alpha_constraint
Constraint for the weights.
shared_axes The axes along which to share learnable parameters for the activation func-
tion. For example, if the incoming feature maps are from a 2D convolution
with output shape (batch, height, width, channels), and you wish to share pa-
rameters across space so that each filter only has one set of parameters, set
shared_axes=c(1, 2).
batch_input_shape
layer_activation_relu 185
See Also
Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification.
Other activation layers: layer_activation_elu, layer_activation_leaky_relu, layer_activation_relu,
layer_activation
layer_activation_relu Rectified Linear Unit activation function
Description
Rectified Linear Unit activation function
Usage
layer_activation_relu(object, max_value = NULL, negative_slope = 0,
threshold = 0, input_shape = NULL, batch_input_shape = NULL,
weights = NULL)
Arguments
max_value loat, the maximum output value.
negative_slope float >= 0 Negative slope coefficient.
threshold float. Threshold value for thresholded activation.
batch_input_shape
See Also
layer_activation
186 layer_activation_selu
layer_activation_selu Scaled Exponential Linear Unit.
Description
SELU is equal to: scale * elu(x, alpha), where alpha and scale are pre-defined constants.
Usage
layer_activation_selu(object, input_shape = NULL,
Arguments
batch_input_shape
Details
The values of alpha and scale are chosen so that the mean and variance of the inputs are pre-
served between two consecutive layers as long as the weights are initialized correctly (see initial-
izer_lecun_normal) and the number of inputs is "large enough" (see article for more information).
Note:
• To be used together with the initialization "lecun_normal".
• To be used together with the dropout variant "AlphaDropout".
See Also
Self-Normalizing Neural Networks, initializer_lecun_normal, layer_alpha_dropout
layer_activation_relu, layer_activation_softmax, layer_activation_thresholded_relu,
layer_activation
layer_activation_softmax 187
layer_activation_softmax
Softmax activation function.
Description
It follows: f(x) = alpha * (exp(x) - 1.0) for x < 0, f(x) = x for ‘x

= 0‘.
Usage
layer_activation_softmax(object, axis = -1, input_shape = NULL,

Arguments

axis Integer, axis along which the softmax normalization is applied.
batch_input_shape
See Also

layer_activation_relu, layer_activation_selu, layer_activation_thresholded_relu, layer_activation
188 layer_activation_thresholded_relu
layer_activation_thresholded_relu
Thresholded Rectified Linear Unit.
Description
It follows: f(x) = x for x > theta, f(x) = 0 otherwise.
Usage
layer_activation_thresholded_relu(object, theta = 1,
input_shape = NULL, batch_input_shape = NULL, batch_size = NULL,
dtype = NULL, name = NULL, trainable = NULL, weights = NULL)
Arguments

theta float >= 0. Threshold location of activation.
batch_input_shape
See Also
Zero-bias autoencoders and the benefits of co-adapting features.

layer_activation_relu, layer_activation_selu, layer_activation_softmax, layer_activation
layer_activity_regularization 189
layer_activity_regularization
Layer that applies an update to the cost function based input activity.
Description
Layer that applies an update to the cost function based input activity.
Usage
layer_activity_regularization(object, l1 = 0, l2 = 0,
Arguments
l1 L1 regularization factor (positive float).
l2 L2 regularization factor (positive float).
batch_input_shape
Input shape
Arbitrary. Use the keyword argument input_shape (list of integers, does not include the samples
axis) when using this layer as the first layer in a model.
Output shape
Same shape as input.
See Also
Other core layers: layer_activation, layer_dense, layer_dropout, layer_flatten, layer_input,
layer_lambda, layer_masking, layer_permute, layer_repeat_vector, layer_reshape
190 layer_alpha_dropout
layer_add Layer that adds a list of inputs.
Description
It takes as input a list of tensors, all of the same shape, and returns a single tensor (also of the same
shape).
Usage
layer_add(inputs, batch_size = NULL, dtype = NULL, name = NULL,
trainable = NULL, weights = NULL)
Arguments
inputs A list of input tensors (at least 2).
Value
A tensor, the sum of the inputs.
See Also
Other merge layers: layer_average, layer_concatenate, layer_dot, layer_maximum, layer_minimum,
layer_multiply, layer_subtract
layer_alpha_dropout Applies Alpha Dropout to the input.
Description
Alpha Dropout is a dropout that keeps mean and variance of inputs to their original values, in order
to ensure the self-normalizing property even after this dropout.
Usage
layer_alpha_dropout(object, rate, noise_shape = NULL, seed = NULL,
layer_alpha_dropout 191
Arguments
rate float, drop probability (as with layer_dropout()). The multiplicative noise
will have standard deviation sqrt(rate / (1 - rate)).
noise_shape Noise shape
seed An integer to use as random seed.
batch_input_shape
Details
Alpha Dropout fits well to Scaled Exponential Linear Units by randomly setting activations to the
negative saturation value.
Input shape
Output shape
References
• Self-Normalizing Neural Networks
See Also
Other noise layers: layer_gaussian_dropout, layer_gaussian_noise
192 layer_average_pooling_1d
layer_average Layer that averages a list of inputs.
Description
shape).
Usage
layer_average(inputs, batch_size = NULL, dtype = NULL, name = NULL,
Arguments
Value
A tensor, the average of the inputs.
See Also
Other merge layers: layer_add, layer_concatenate, layer_dot, layer_maximum, layer_minimum,
layer_average_pooling_1d
Average pooling for temporal data.
Description
Average pooling for temporal data.
Usage
layer_average_pooling_1d(object, pool_size = 2L, strides = NULL,
padding = "valid", data_format = "channels_last",
batch_size = NULL, name = NULL, trainable = NULL, weights = NULL)
layer_average_pooling_2d 193
Arguments
pool_size Integer, size of the average pooling windows.
strides Integer, or NULL. Factor by which to downscale. E.g. 2 will halve the input. If
NULL, it will default to pool_size.
padding One of "valid" or "same" (case-insensitive).
data_format One of channels_last (default) or channels_first. The ordering of the di-
mensions in the inputs.
Input shape
3D tensor with shape: (batch_size, steps, features).
Output shape
3D tensor with shape: (batch_size, downsampled_steps, features).
See Also
Other pooling layers: layer_average_pooling_2d, layer_average_pooling_3d, layer_global_average_pooling_1d,
layer_global_average_pooling_2d, layer_global_average_pooling_3d, layer_global_max_pooling_1d,
layer_global_max_pooling_2d, layer_global_max_pooling_3d, layer_max_pooling_1d, layer_max_pooling_2d,
layer_max_pooling_3d
Average pooling operation for spatial data.
Description
Average pooling operation for spatial data.
Usage
layer_average_pooling_2d(object, pool_size = c(2L, 2L), strides = NULL,
padding = "valid", data_format = NULL, batch_size = NULL,
194 layer_average_pooling_2d
Arguments

pool_size integer or list of 2 integers, factors by which to downscale (vertical, horizontal).
(2, 2) will halve the input in both spatial dimension. If only one integer is
specified, the same window length will be used for both dimensions.
strides Integer, list of 2 integers, or NULL. Strides values. If NULL, it will default to
pool_size.
data_format A string, one of channels_last (default) or channels_first. The ordering
of the dimensions in the inputs. channels_last corresponds to inputs with
shape (batch, height, width, channels) while channels_first cor-
responds to inputs with shape (batch, channels, height, width). It
Input shape
• If data_format='channels_last': 4D tensor with shape: (batch_size, rows, cols, channels)

• If data_format='channels_first': 4D tensor with shape: (batch_size, channels, rows, cols)
Output shape
• If data_format='channels_last': 4D tensor with shape: (batch_size, pooled_rows, pooled_cols, channels)

• If data_format='channels_first': 4D tensor with shape: (batch_size, channels, pooled_rows, pooled_cols
See Also

layer_average_pooling_3d 195
Average pooling operation for 3D data (spatial or spatio-temporal).
Description
Average pooling operation for 3D data (spatial or spatio-temporal).
Usage
layer_average_pooling_3d(object, pool_size = c(2L, 2L, 2L),
strides = NULL, padding = "valid", data_format = NULL,
Arguments
pool_size list of 3 integers, factors by which to downscale (dim1, dim2, dim3). (2, 2, 2)
will halve the size of the 3D input in each dimension.
strides list of 3 integers, or NULL. Strides values.
data_format A string, one of channels_last (default) or channels_first. The ordering of
the dimensions in the inputs. channels_last corresponds to inputs with shape
(batch, spatial_dim1, spatial_dim2, spatial_dim3, channels) while
channels_first corresponds to inputs with shape (batch, channels, spatial_dim1, spatial_dim2
It defaults to the image_data_format value found in your Keras config file at
Input shape
• If data_format='channels_last': 5D tensor with shape: (batch_size, spatial_dim1, spatial_dim2, spatial
• If data_format='channels_first': 5D tensor with shape: (batch_size, channels, spatial_dim1, spatial_di
Output shape
• If data_format='channels_last': 5D tensor with shape: (batch_size, pooled_dim1, pooled_dim2, pooled_di
• If data_format='channels_first': 5D tensor with shape: (batch_size, channels, pooled_dim1, pooled_dim2
196 layer_batch_normalization
See Also
layer_batch_normalization
Batch normalization layer (Ioffe and Szegedy, 2014).
Description
Normalize the activations of the previous layer at each batch, i.e. applies a transformation that
maintains the mean activation close to 0 and the activation standard deviation close to 1.
Usage
layer_batch_normalization(object, axis = -1L, momentum = 0.99,
epsilon = 0.001, center = TRUE, scale = TRUE,
beta_initializer = "zeros", gamma_initializer = "ones",
moving_mean_initializer = "zeros",
moving_variance_initializer = "ones", beta_regularizer = NULL,
gamma_regularizer = NULL, beta_constraint = NULL,
gamma_constraint = NULL, input_shape = NULL,
Arguments
axis Integer, the axis that should be normalized (typically the features axis). For in-
stance, after a Conv2D layer with data_format="channels_first", set axis=1
in BatchNormalization.
momentum Momentum for the moving mean and the moving variance.
epsilon Small float added to variance to avoid dividing by zero.
center If TRUE, add offset of beta to normalized tensor. If FALSE, beta is ignored.
scale If TRUE, multiply by gamma. If FALSE, gamma is not used. When the next layer
is linear (also e.g. nn.relu), this can be disabled since the scaling will be done
by the next layer.
beta_initializer
Initializer for the beta weight.
gamma_initializer
Initializer for the gamma weight.
moving_mean_initializer
Initializer for the moving mean.
layer_batch_normalization 197
moving_variance_initializer
Initializer for the moving variance.
beta_regularizer
Optional regularizer for the beta weight.
gamma_regularizer
Optional regularizer for the gamma weight.
beta_constraint
Optional constraint for the beta weight.
gamma_constraint
Optional constraint for the gamma weight.
batch_input_shape
Input shape
Output shape
References
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate

Shift
198 layer_conv_1d
layer_concatenate Layer that concatenates a list of inputs.
Description
It takes as input a list of tensors, all of the same shape expect for the concatenation axis, and returns
a single tensor, the concatenation of all inputs.
Usage
layer_concatenate(inputs, axis = -1, batch_size = NULL, dtype = NULL,
Arguments
axis Concatenation axis.
Value
A tensor, the concatenation of the inputs alongside axis axis.
See Also
Other merge layers: layer_add, layer_average, layer_dot, layer_maximum, layer_minimum,
layer_conv_1d 1D convolution layer (e.g. temporal convolution).
Description
This layer creates a convolution kernel that is convolved with the layer input over a single spatial
(or temporal) dimension to produce a tensor of outputs. If use_bias is TRUE, a bias vector is
created and added to the outputs. Finally, if activation is not NULL, it is applied to the outputs as
well. When using this layer as the first layer in a model, provide an input_shape argument (list
of integers or NULL , e.g. (10, 128) for sequences of 10 vectors of 128-dimensional vectors, or
(NULL, 128) for variable-length sequences of 128-dimensional vectors.
layer_conv_1d 199
Usage
layer_conv_1d(object, filters, kernel_size, strides = 1L,
dilation_rate = 1L, activation = NULL, use_bias = TRUE,
kernel_initializer = "glorot_uniform", bias_initializer = "zeros",
kernel_regularizer = NULL, bias_regularizer = NULL,
activity_regularizer = NULL, kernel_constraint = NULL,
bias_constraint = NULL, input_shape = NULL,
Arguments
filters Integer, the dimensionality of the output space (i.e. the number of output filters
in the convolution).
kernel_size An integer or list of a single integer, specifying the length of the 1D convolution
window.
strides An integer or list of a single integer, specifying the stride length of the con-
volution. Specifying any stride value != 1 is incompatible with specifying any
dilation_rate value != 1.
padding One of "valid", "causal" or "same" (case-insensitive). "valid" means "no
padding". "same" results in padding the input such that the output has the same
length as the original input. "causal" results in causal (dilated) convolutions,
e.g. output[t] does not depend on input[t+1:]. Useful when modeling tem-
poral data where the model should not violate the temporal order. See WaveNet:
A Generative Model for Raw Audio, section 2.1.
data_format A string, one of "channels_last" (default) or "channels_first". The or-
dering of the dimensions in the inputs. "channels_last" corresponds to in-
puts with shape (batch, length, channels) (default format for tempo-
ral data in Keras) while "channels_first" corresponds to inputs with shape
(batch, channels, length).
dilation_rate an integer or list of a single integer, specifying the dilation rate to use for dilated
convolution. Currently, specifying any dilation_rate value != 1 is incompat-
ible with specifying any strides value != 1.
activation Activation function to use. If you don’t specify anything, no activation is applied
(ie. "linear" activation: a(x) = x).
use_bias Boolean, whether the layer uses a bias vector.
kernel_initializer
Initializer for the kernel weights matrix.
bias_initializer
Initializer for the bias vector.
kernel_regularizer
Regularizer function applied to the kernel weights matrix.
bias_regularizer
Regularizer function applied to the bias vector.
200 layer_conv_2d
activity_regularizer
Regularizer function applied to the output of the layer (its "activation")..
kernel_constraint
Constraint function applied to the kernel matrix.
bias_constraint
Constraint function applied to the bias vector.
batch_input_shape
Input shape
3D tensor with shape: (batch_size, steps, input_dim)
Output shape
3D tensor with shape: (batch_size, new_steps, filters) steps value might have changed
due to padding or strides.
See Also
Other convolutional layers: layer_conv_2d_transpose, layer_conv_2d, layer_conv_3d_transpose,
layer_conv_3d, layer_conv_lstm_2d, layer_cropping_1d, layer_cropping_2d, layer_cropping_3d,
layer_depthwise_conv_2d, layer_separable_conv_1d, layer_separable_conv_2d, layer_upsampling_1d,
layer_upsampling_2d, layer_upsampling_3d, layer_zero_padding_1d, layer_zero_padding_2d,
layer_zero_padding_3d
layer_conv_2d 2D convolution layer (e.g. spatial convolution over images).
Description
This layer creates a convolution kernel that is convolved with the layer input to produce a ten-
sor of outputs. If use_bias is TRUE, a bias vector is created and added to the outputs. Fi-
nally, if activation is not NULL, it is applied to the outputs as well. When using this layer
as the first layer in a model, provide the keyword argument input_shape (list of integers, does
not include the sample axis), e.g. input_shape=c(128, 128, 3) for 128x128 RGB pictures in
data_format="channels_last".
layer_conv_2d 201
Usage
layer_conv_2d(object, filters, kernel_size, strides = c(1L, 1L),
padding = "valid", data_format = NULL, dilation_rate = c(1L, 1L),
activation = NULL, use_bias = TRUE,
Arguments
kernel_size An integer or list of 2 integers, specifying the width and height of the 2D convo-
lution window. Can be a single integer to specify the same value for all spatial
dimensions.
strides An integer or list of 2 integers, specifying the strides of the convolution along
the width and height. Can be a single integer to specify the same value for
all spatial dimensions. Specifying any stride value != 1 is incompatible with
specifying any dilation_rate value != 1.
padding one of "valid" or "same" (case-insensitive). Note that "same" is slightly in-
consistent across backends with strides != 1, as described here
dilation_rate an integer or list of 2 integers, specifying the dilation rate to use for dilated
convolution. Can be a single integer to specify the same value for all spatial di-
mensions. Currently, specifying any dilation_rate value != 1 is incompatible
with specifying any stride value != 1.
kernel_initializer
bias_initializer
kernel_regularizer
bias_regularizer
202 layer_conv_2d
kernel_constraint
bias_constraint
batch_input_shape
Input shape
4D tensor with shape: (samples, channels, rows, cols) if data_format=’channels_first’ or 4D

tensor with shape: (samples, rows, cols, channels) if data_format=’channels_last’.
Output shape
4D tensor with shape: (samples, filters, new_rows, new_cols) if data_format=’channels_first’

or 4D tensor with shape: (samples, new_rows, new_cols, filters) if data_format=’channels_last’.
rows and cols values might have changed due to padding.
See Also
Other convolutional layers: layer_conv_1d, layer_conv_2d_transpose, layer_conv_3d_transpose,

layer_upsampling_2d, layer_upsampling_3d, layer_zero_padding_1d, layer_zero_padding_2d,
layer_zero_padding_3d
layer_conv_2d_transpose 203
layer_conv_2d_transpose
Transposed 2D convolution layer (sometimes called Deconvolution).
Description
The need for transposed convolutions generally arises from the desire to use a transformation go-
ing in the opposite direction of a normal convolution, i.e., from something that has the shape
of the output of some convolution to something that has the shape of its input while maintain-
ing a connectivity pattern that is compatible with said convolution. When using this layer as the
first layer in a model, provide the keyword argument input_shape (list of integers, does not in-
clude the sample axis), e.g. input_shape=c(128L, 128L, 3L) for 128x128 RGB pictures in
data_format="channels_last".
Usage
layer_conv_2d_transpose(object, filters, kernel_size, strides = c(1, 1),
padding = "valid", output_padding = NULL, data_format = NULL,
dilation_rate = c(1, 1), activation = NULL, use_bias = TRUE,
Arguments
dimensions.
padding one of "valid" or "same" (case-insensitive).
output_padding An integer or list of 2 integers, specifying the amount of padding along the
height and width of the output tensor. Can be a single integer to specify the
same value for all spatial dimensions. The amount of output padding along a
given dimension must be lower than the stride along that same dimension. If set
to NULL (default), the output shape is inferred.
204 layer_conv_2d_transpose

dilation_rate Dialation rate.
kernel_initializer
bias_initializer
kernel_regularizer
bias_regularizer
kernel_constraint
bias_constraint
batch_input_shape
Input shape
4D tensor with shape: (batch, channels, rows, cols) if data_format=’channels_first’ or 4D
tensor with shape: (batch, rows, cols, channels) if data_format=’channels_last’.
layer_conv_3d 205
Output shape
4D tensor with shape: (batch, filters, new_rows, new_cols) if data_format=’channels_first’
or 4D tensor with shape: (batch, new_rows, new_cols, filters) if data_format=’channels_last’.
References
• A guide to convolution arithmetic for deep learning
• Deconvolutional Networks
See Also
Other convolutional layers: layer_conv_1d, layer_conv_2d, layer_conv_3d_transpose, layer_conv_3d,
layer_conv_lstm_2d, layer_cropping_1d, layer_cropping_2d, layer_cropping_3d, layer_depthwise_conv_2d,
layer_separable_conv_1d, layer_separable_conv_2d, layer_upsampling_1d, layer_upsampling_2d,
layer_upsampling_3d, layer_zero_padding_1d, layer_zero_padding_2d, layer_zero_padding_3d
layer_conv_3d 3D convolution layer (e.g. spatial convolution over volumes).
Description
This layer creates a convolution kernel that is convolved with the layer input to produce a tensor
of outputs. If use_bias is TRUE, a bias vector is created and added to the outputs. Finally, if
activation is not NULL, it is applied to the outputs as well. When using this layer as the first
layer in a model, provide the keyword argument input_shape (list of integers, does not include the
sample axis), e.g. input_shape=c(128L, 128L, 128L, 3L) for 128x128x128 volumes with a
single channel, in data_format="channels_last".
Usage
layer_conv_3d(object, filters, kernel_size, strides = c(1L, 1L, 1L),
padding = "valid", data_format = NULL, dilation_rate = c(1L, 1L,
1L), activation = NULL, use_bias = TRUE,
Arguments
206 layer_conv_3d
kernel_size An integer or list of 3 integers, specifying the depth, height, and width of the 3D
convolution window. Can be a single integer to specify the same value for all
spatial dimensions.
each spatial dimension. Can be a single integer to specify the same value for
kernel_initializer
bias_initializer
kernel_regularizer
bias_regularizer
kernel_constraint
bias_constraint
batch_input_shape
layer_conv_3d_transpose 207
Input shape
5D tensor with shape: (samples, channels, conv_dim1, conv_dim2, conv_dim3) if data_format=’channels_first’
or 5D tensor with shape: (samples, conv_dim1, conv_dim2, conv_dim3, channels) if
data_format=’channels_last’.
Output shape
5D tensor with shape: (samples, filters, new_conv_dim1, new_conv_dim2, new_conv_dim3)
if data_format=’channels_first’ or 5D tensor with shape: (samples, new_conv_dim1, new_conv_dim2, new_conv_dim3,
if data_format=’channels_last’. new_conv_dim1, new_conv_dim2 and new_conv_dim3 values might
have changed due to padding.
See Also
Other convolutional layers: layer_conv_1d, layer_conv_2d_transpose, layer_conv_2d, layer_conv_3d_transpose,
layer_conv_3d_transpose
Transposed 3D convolution layer (sometimes called Deconvolution).
Description
The need for transposed convolutions generally arises from the desire to use a transformation go-
ing in the opposite direction of a normal convolution, i.e., from something that has the shape of
the output of some convolution to something that has the shape of its input while maintaining a
connectivity pattern that is compatible with said convolution.
Usage
layer_conv_3d_transpose(object, filters, kernel_size, strides = c(1, 1,
1), padding = "valid", output_padding = NULL, data_format = NULL,
208 layer_conv_3d_transpose
Arguments
kernel_size An integer or list of 3 integers, specifying the depth, height, and width of the 3D
convolution window. Can be a single integer to specify the same value for all
spatial dimensions.
the depth, height and width.. Can be a single integer to specify the same value
for all spatial dimensions. Specifying any stride value != 1 is incompatible with
output_padding An integer or list of 3 integers, specifying the amount of padding along the
depth, height, and width of the output tensor. Can be a single integer to specify
the same value for all spatial dimensions. The amount of output padding along
a given dimension must be lower than the stride along that same dimension. If
set to NULL (default), the output shape is inferred.
(batch, depth, height, width, channels) while channels_first corre-
sponds to inputs with shape (batch, channels, depth, height, width).
kernel_initializer
bias_initializer
kernel_regularizer
Regularizer function applied to the kernel weights matrix,
bias_regularizer
Regularizer function applied to the output of the layer (its "activation").
kernel_constraint
bias_constraint
layer_conv_lstm_2d 209
batch_input_shape
Details
When using this layer as the first layer in a model, provide the keyword argument input_shape (list
of integers, does not include the sample axis), e.g. input_shape = list(128, 128, 128, 3) for
a 128x128x128 volume with 3 channels if data_format="channels_last".
References
• A guide to convolution arithmetic for deep learning
• Deconvolutional Networks
See Also
Other convolutional layers: layer_conv_1d, layer_conv_2d_transpose, layer_conv_2d, layer_conv_3d,
layer_conv_lstm_2d Convolutional LSTM.
Description
It is similar to an LSTM layer, but the input transformations and recurrent transformations are both
convolutional.
Usage
layer_conv_lstm_2d(object, filters, kernel_size, strides = c(1L, 1L),
padding = "valid", data_format = NULL, dilation_rate = c(1L, 1L),
activation = "tanh", recurrent_activation = "hard_sigmoid",
use_bias = TRUE, kernel_initializer = "glorot_uniform",
recurrent_initializer = "orthogonal", bias_initializer = "zeros",
unit_forget_bias = TRUE, kernel_regularizer = NULL,
210 layer_conv_lstm_2d
recurrent_regularizer = NULL, bias_regularizer = NULL,

recurrent_constraint = NULL, bias_constraint = NULL,
return_sequences = FALSE, go_backwards = FALSE, stateful = FALSE,
dropout = 0, recurrent_dropout = 0, batch_size = NULL,
name = NULL, trainable = NULL, weights = NULL,
input_shape = NULL)
Arguments
kernel_size An integer or list of n integers, specifying the dimensions of the convolution
window.
strides An integer or list of n integers, specifying the strides of the convolution. Speci-
fying any stride value != 1 is incompatible with specifying any dilation_rate
value != 1.
shape (batch, time, ..., channels) while channels_first corresponds
to inputs with shape (batch, time, channels, ...). It defaults to the
image_data_format value found in your Keras config file at ~/.keras/keras.json.
If you never set it, then it will be "channels_last".
dilation_rate An integer or list of n integers, specifying the dilation rate to use for dilated con-
volution. Currently, specifying any dilation_rate value != 1 is incompatible
with specifying any strides value != 1.
recurrent_activation
Activation function to use for the recurrent step.
kernel_initializer
Initializer for the kernel weights matrix, used for the linear transformation of
the inputs..
recurrent_initializer
Initializer for the recurrent_kernel weights matrix, used for the linear trans-
formation of the recurrent state..
bias_initializer
unit_forget_bias
Boolean. If TRUE, add 1 to the bias of the forget gate at initialization. Use
in combination with bias_initializer="zeros". This is recommended in
Jozefowicz et al.
layer_conv_lstm_2d 211
kernel_regularizer
recurrent_regularizer
Regularizer function applied to the recurrent_kernel weights matrix.
bias_regularizer
kernel_constraint
Constraint function applied to the kernel weights matrix.
recurrent_constraint
Constraint function applied to the recurrent_kernel weights matrix.
bias_constraint
return_sequences
Boolean. Whether to return the last output in the output sequence, or the full
sequence.
go_backwards Boolean (default FALSE). If TRUE, rocess the input sequence backwards.
stateful Boolean (default FALSE). If TRUE, the last state for each sample at index i in a
batch will be used as initial state for the sample of index i in the following batch.
dropout Float between 0 and 1. Fraction of the units to drop for the linear transformation
of the inputs.
recurrent_dropout
Float between 0 and 1. Fraction of the units to drop for the linear transformation
of the recurrent state.
Input shape
• if data_format=’channels_first’ 5D tensor with shape: (samples,time, channels, rows, cols)
– if data_format=’channels_last’ 5D tensor with shape: (samples,time, rows, cols, channels)
References
• Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting
The current implementation does not include the feedback loop on the cells output
212 layer_cropping_1d
See Also
layer_conv_3d, layer_cropping_1d, layer_cropping_2d, layer_cropping_3d, layer_depthwise_conv_2d,
layer_cropping_1d Cropping layer for 1D input (e.g. temporal sequence).
Description
It crops along the time dimension (axis 1).
Usage
layer_cropping_1d(object, cropping = c(1L, 1L), batch_size = NULL,
Arguments
cropping int or list of int (length 2) How many units should be trimmed off at the begin-
ning and end of the cropping dimension (axis 1). If a single int is provided, the
same value will be used for both.
Input shape
3D tensor with shape (batch, axis_to_crop, features)
Output shape
3D tensor with shape (batch, cropped_axis, features)
See Also
layer_conv_3d, layer_conv_lstm_2d, layer_cropping_2d, layer_cropping_3d, layer_depthwise_conv_2d,
layer_cropping_2d 213
layer_cropping_2d Cropping layer for 2D input (e.g. picture).
Description
It crops along spatial dimensions, i.e. width and height.
Usage
layer_cropping_2d(object, cropping = list(c(0L, 0L), c(0L, 0L)),
data_format = NULL, batch_size = NULL, name = NULL,
Arguments
cropping int, or list of 2 ints, or list of 2 lists of 2 ints.
• If int: the same symmetric cropping is applied to width and height.
• If list of 2 ints: interpreted as two different symmetric cropping values for
height and width: (symmetric_height_crop, symmetric_width_crop).
• If list of 2 lists of 2 ints: interpreted as ((top_crop, bottom_crop), (left_crop, right_crop)
Input shape
4D tensor with shape:
• If data_format is "channels_last": (batch, rows, cols, channels)
• If data_format is "channels_first": (batch, channels, rows, cols)
Output shape
• If data_format is "channels_last": (batch, cropped_rows, cropped_cols, channels)
• If data_format is "channels_first": (batch, channels, cropped_rows, cropped_cols)
214 layer_cropping_3d
See Also
layer_cropping_3d Cropping layer for 3D data (e.g. spatial or spatio-temporal).
Description
Cropping layer for 3D data (e.g. spatial or spatio-temporal).
Usage
layer_cropping_3d(object, cropping = list(c(1L, 1L), c(1L, 1L), c(1L,
1L)), data_format = NULL, batch_size = NULL, name = NULL,
Arguments
cropping int, or list of 3 ints, or list of 3 lists of 2 ints.
• If int: the same symmetric cropping is applied to depth, height, and width.
• If list of 3 ints: interpreted as two different symmetric cropping values for
depth, height, and width: (symmetric_dim1_crop, symmetric_dim2_crop, symmetric_dim3_cro
• If list of 3 list of 2 ints: interpreted as ((left_dim1_crop, right_dim1_crop), (left_dim2_crop
Input shape
• If data_format is "channels_last": (batch, first_axis_to_crop, second_axis_to_crop, third_axis_to_cr
• If data_format is "channels_first": (batch, depth, first_axis_to_crop, second_axis_to_crop, third_ax
layer_cudnn_gru 215
Output shape
• If data_format is "channels_last": (batch, first_cropped_axis, second_cropped_axis, third_cropped_ax

• If data_format is "channels_first": (batch, depth, first_cropped_axis, second_cropped_axis, third_cr
See Also
layer_cudnn_gru Fast GRU implementation backed by

Rhrefhttps://developer.nvidia.com/cudnnCuDNN.
Description
Can only be run on GPU, with the TensorFlow backend.
Usage
layer_cudnn_gru(object, units, kernel_initializer = "glorot_uniform",
kernel_regularizer = NULL, recurrent_regularizer = NULL,
bias_regularizer = NULL, activity_regularizer = NULL,
kernel_constraint = NULL, recurrent_constraint = NULL,
bias_constraint = NULL, return_sequences = FALSE,
return_state = FALSE, stateful = FALSE, input_shape = NULL,
Arguments
units Positive integer, dimensionality of the output space.
kernel_initializer
the inputs.
formation of the recurrent state.
bias_initializer
216 layer_cudnn_gru
kernel_regularizer
bias_regularizer
kernel_constraint
bias_constraint
return_sequences
sequence.
return_state Boolean (default FALSE). Whether to return the last state in addition to the
output.
batch_input_shape
References
• On the Properties of Neural Machine Translation: Encoder-Decoder Approaches
• Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling
• A Theoretically Grounded Application of Dropout in Recurrent Neural Networks
See Also
Other recurrent layers: layer_cudnn_lstm, layer_gru, layer_lstm, layer_simple_rnn
layer_cudnn_lstm 217
layer_cudnn_lstm Fast LSTM implementation backed by

Rhrefhttps://developer.nvidia.com/cudnnCuDNN.
Description
Can only be run on GPU, with the TensorFlow backend.
Usage
layer_cudnn_lstm(object, units, kernel_initializer = "glorot_uniform",
recurrent_constraint = NULL, bias_constraint = NULL,
return_sequences = FALSE, return_state = FALSE, stateful = FALSE,
Arguments
kernel_initializer
the inputs.
bias_initializer
unit_forget_bias
Boolean. If TRUE, add 1 to the bias of the forget gate at initialization. Setting
it to true will also force bias_initializer="zeros". This is recommended in
Jozefowicz et al.
kernel_regularizer
bias_regularizer
kernel_constraint
218 layer_dense
bias_constraint
return_sequences
sequence.
output.
batch_input_shape
References
• Long short-term memory (original 1997 paper)
• Supervised sequence labeling with recurrent neural networks
See Also
Other recurrent layers: layer_cudnn_gru, layer_gru, layer_lstm, layer_simple_rnn
layer_dense Add a densely-connected NN layer to an output
Description
Implements the operation: output = activation(dot(input, kernel) + bias) where activation
is the element-wise activation function passed as the activation argument, kernel is a weights
matrix created by the layer, and bias is a bias vector created by the layer (only applicable if
use_bias is TRUE). Note: if the input to the layer has a rank greater than 2, then it is flattened
prior to the initial dot product with kernel.
layer_dense 219
Usage
layer_dense(object, units, activation = NULL, use_bias = TRUE,
Arguments
activation Name of activation function to use. If you don’t specify anything, no activation
is applied (ie. "linear" activation: a(x) = x).
use_bias Whether the layer uses a bias vector.
kernel_initializer
bias_initializer
kernel_regularizer
bias_regularizer
kernel_constraint
bias_constraint
batch_input_shape
220 layer_depthwise_conv_2d
Input and Output Shapes

Input shape: nD tensor with shape: (batch_size, ..., input_dim). The most common situation
would be a 2D input with shape (batch_size, input_dim).
Output shape: nD tensor with shape: (batch_size, ..., units). For instance, for a 2D input
with shape (batch_size, input_dim), the output would have shape (batch_size, unit).
See Also
Other core layers: layer_activation, layer_activity_regularization, layer_dropout, layer_flatten,
layer_depthwise_conv_2d
Depthwise separable 2D convolution.
Description
Depthwise Separable convolutions consists in performing just the first step in a depthwise spatial
convolution (which acts on each input channel separately). The depth_multiplier argument con-
trols how many output channels are generated per input channel in the depthwise step.
Usage
layer_depthwise_conv_2d(object, kernel_size, strides = c(1, 1),
padding = "valid", depth_multiplier = 1, data_format = NULL,
depthwise_initializer = "glorot_uniform", bias_initializer = "zeros",
depthwise_regularizer = NULL, bias_regularizer = NULL,
activity_regularizer = NULL, depthwise_constraint = NULL,
Arguments
dimensions.
layer_depthwise_conv_2d 221
depth_multiplier
The number of depthwise convolution output channels for each input channel.
The total number of depthwise convolution output channels will be equal to
filters_in * depth_multiplier.
depthwise_initializer
Initializer for the depthwise kernel matrix.
bias_initializer
depthwise_regularizer
Regularizer function applied to the depthwise kernel matrix.
bias_regularizer
depthwise_constraint
Constraint function applied to the depthwise kernel matrix.
bias_constraint
batch_input_shape
See Also
222 layer_dot

layer_dot Layer that computes a dot product between samples in two tensors.
Description
Layer that computes a dot product between samples in two tensors.
Usage
layer_dot(inputs, axes, normalize = FALSE, batch_size = NULL,

Arguments

axes Integer or list of integers, axis or axes along which to take the dot product.
normalize Whether to L2-normalize samples along the dot product axis before taking the
dot product. If set to TRUE, then the output of the dot product is the cosine
proximity between the two samples. **kwargs: Standard layer keyword argu-
ments.
Value
A tensor, the dot product of the samples from the inputs.
See Also
Other merge layers: layer_add, layer_average, layer_concatenate, layer_maximum, layer_minimum,

layer_dropout 223
layer_dropout Applies Dropout to the input.
Description
Dropout consists in randomly setting a fraction rate of input units to 0 at each update during
training time, which helps prevent overfitting.
Usage
layer_dropout(object, rate, noise_shape = NULL, seed = NULL,
Arguments
rate float between 0 and 1. Fraction of the input units to drop.
noise_shape 1D integer tensor representing the shape of the binary dropout mask that will be
multiplied with the input. For instance, if your inputs have shape (batch_size, timesteps, features)
and you want the dropout mask to be the same for all timesteps, you can use
noise_shape=c(batch_size, 1, features).
seed integer to use as random seed.
batch_input_shape
See Also
Other core layers: layer_activation, layer_activity_regularization, layer_dense, layer_flatten,
Other dropout layers: layer_spatial_dropout_1d, layer_spatial_dropout_2d, layer_spatial_dropout_3d
224 layer_embedding
layer_embedding Turns positive integers (indexes) into dense vectors of fixed size.
Description
For example, list(4L, 20L) -> list(c(0.25, 0.1), c(0.6, -0.2)) This layer can only be
used as the first layer in a model.
Usage
layer_embedding(object, input_dim, output_dim,
embeddings_initializer = "uniform", embeddings_regularizer = NULL,
activity_regularizer = NULL, embeddings_constraint = NULL,
mask_zero = FALSE, input_length = NULL, batch_size = NULL,
Arguments
input_dim int > 0. Size of the vocabulary, i.e. maximum integer index + 1.
output_dim int >= 0. Dimension of the dense embedding.
embeddings_initializer
Initializer for the embeddings matrix.
embeddings_regularizer
Regularizer function applied to the embeddings matrix.
embeddings_constraint
Constraint function applied to the embeddings matrix.
mask_zero Whether or not the input value 0 is a special "padding" value that should be
masked out. This is useful when using recurrent layers, which may take variable
length inputs. If this is TRUE then all subsequent layers in the model need to
support masking or an exception will be raised. If mask_zero is set to TRUE,
as a consequence, index 0 cannot be used in the vocabulary (input_dim should
equal size of vocabulary + 1).
input_length Length of input sequences, when it is constant. This argument is required if you
are going to connect Flatten then Dense layers upstream (without it, the shape
of the dense outputs cannot be computed).
layer_flatten 225
Input shape
2D tensor with shape: (batch_size, sequence_length).
Output shape
3D tensor with shape: (batch_size, sequence_length, output_dim).
References
layer_flatten Flattens an input
Description
Flatten a given input, does not affect the batch size.
Usage
layer_flatten(object, data_format = NULL, input_shape = NULL,
Arguments
data_format A string. one of channels_last (default) or channels_first. The order-
ing of the dimensions in the inputs. The purpose of this argument is to pre-
serve weight ordering when switching a model from one data format to another.
channels_last corresponds to inputs with shape (batch, ..., channels)
while channels_first corresponds to inputs with shape (batch, channels, ...).
See Also
Other core layers: layer_activation, layer_activity_regularization, layer_dense, layer_dropout,
226 layer_gaussian_dropout
layer_gaussian_dropout
Apply multiplicative 1-centered Gaussian noise.
Description
As it is a regularization layer, it is only active at training time.
Usage
layer_gaussian_dropout(object, rate, input_shape = NULL,
Arguments
rate float, drop probability (as with Dropout). The multiplicative noise will have
standard deviation sqrt(rate / (1 - rate)).
batch_input_shape
Input shape
Output shape
References
• Dropout: A Simple Way to Prevent Neural Networks from Overfitting Srivastava, Hinton, et
al. 2014
layer_gaussian_noise 227
See Also
Other noise layers: layer_alpha_dropout, layer_gaussian_noise
layer_gaussian_noise Apply additive zero-centered Gaussian noise.
Description
This is useful to mitigate overfitting (you could see it as a form of random data augmentation).
Gaussian Noise (GS) is a natural choice as corruption process for real valued inputs. As it is a
regularization layer, it is only active at training time.
Usage
layer_gaussian_noise(object, stddev, input_shape = NULL,
Arguments
stddev float, standard deviation of the noise distribution.
batch_input_shape
Input shape
Output shape
228 layer_global_average_pooling_1d
See Also
Other noise layers: layer_alpha_dropout, layer_gaussian_dropout
layer_global_average_pooling_1d
Global average pooling operation for temporal data.
Description
Global average pooling operation for temporal data.
Usage
layer_global_average_pooling_1d(object, data_format = "channels_last",
Arguments
Input shape
Output shape
2D tensor with shape: (batch_size, channels)
See Also
Other pooling layers: layer_average_pooling_1d, layer_average_pooling_2d, layer_average_pooling_3d,
layer_global_average_pooling_2d 229
Global average pooling operation for spatial data.
Description
Global average pooling operation for spatial data.
Usage
layer_global_average_pooling_2d(object, data_format = NULL,
Arguments
Input shape
Output shape
See Also
230 layer_global_average_pooling_3d
Global Average pooling operation for 3D data.
Description
Global Average pooling operation for 3D data.
Usage
layer_global_average_pooling_3d(object, data_format = NULL,
Arguments
Input shape
Output shape
See Also
layer_global_max_pooling_1d 231
layer_global_max_pooling_1d
Global max pooling operation for temporal data.
Description
Global max pooling operation for temporal data.
Usage
layer_global_max_pooling_1d(object, data_format = "channels_last",

Arguments

Input shape
Output shape
See Also

layer_global_average_pooling_1d, layer_global_average_pooling_2d, layer_global_average_pooling_3d,
232 layer_global_max_pooling_2d
Global max pooling operation for spatial data.
Description
Global max pooling operation for spatial data.
Usage
layer_global_max_pooling_2d(object, data_format = NULL,
Arguments
Input shape
Output shape
See Also
layer_global_max_pooling_3d 233
Global Max pooling operation for 3D data.
Description
Global Max pooling operation for 3D data.
Usage
layer_global_max_pooling_3d(object, data_format = NULL,
Arguments
Input shape
Output shape
See Also
234 layer_gru
layer_gru Gated Recurrent Unit - Cho et al.
Description
There are two variants. The default one is based on 1406.1078v3 and has reset gate applied to
hidden state before matrix multiplication. The other one is based on original 1406.1078v1 and has
the order reversed.
Usage
layer_gru(object, units, activation = "tanh",
recurrent_activation = "hard_sigmoid", use_bias = TRUE,
return_sequences = FALSE, return_state = FALSE,
go_backwards = FALSE, stateful = FALSE, unroll = FALSE,
reset_after = FALSE, kernel_initializer = "glorot_uniform",
bias_constraint = NULL, dropout = 0, recurrent_dropout = 0,
Arguments
activation Activation function to use. Default: hyperbolic tangent (tanh). If you pass
NULL, no activation is applied (ie. "linear" activation: a(x) = x).
return_sequences
sequence.
output.
go_backwards Boolean (default FALSE). If TRUE, process the input sequence backwards and
unroll Boolean (default FALSE). If TRUE, the network will be unrolled, else a sym-
bolic loop will be used. Unrolling can speed-up a RNN, although it tends to be
more memory-intensive. Unrolling is only suitable for short sequences.
layer_gru 235
reset_after GRU convention (whether to apply reset gate after or before matrix multiplica-
tion). FALSE = "before" (default), TRUE = "after" (CuDNN compatible).
kernel_initializer
the inputs.
bias_initializer
kernel_regularizer
bias_regularizer
kernel_constraint
bias_constraint
of the inputs.
recurrent_dropout
batch_input_shape
236 layer_gru
Details
The second variant is compatible with CuDNNGRU (GPU-only) and allows inference on CPU.
Thus it has separate biases for kernel and recurrent_kernel. Use reset_after = TTRUE and
recurrent_activation = "sigmoid".
Input shapes
3D tensor with shape (batch_size, timesteps, input_dim), (Optional) 2D tensors with shape
(batch_size, output_dim).
Output shape
• if return_state: a list of tensors. The first tensor is the output. The remaining tensors are
the last states, each with shape (batch_size, units).
• if return_sequences: 3D tensor with shape (batch_size, timesteps, units).
• else, 2D tensor with shape (batch_size, units).
Masking
This layer supports masking for input data with a variable number of timesteps. To introduce masks
to your data, use an embedding layer with the mask_zero parameter set to TRUE.
Statefulness in RNNs
You can set RNN layers to be ’stateful’, which means that the states computed for the samples in
one batch will be reused as initial states for the samples in the next batch. This assumes a one-to-one
mapping between samples in different successive batches.
To enable statefulness:
• Specify stateful=TRUE in the layer constructor.

• Specify a fixed batch size for your model. For sequential models, pass batch_input_shape = c(...)
to the first layer in your model. For functional models with 1 or more Input layers, pass
batch_shape = c(...) to all the first layers in your model. This is the expected shape of
your inputs including the batch size. It should be a vector of integers, e.g. c(32, 10, 100).
• Specify shuffle = FALSE when calling fit().
To reset the states of your model, call reset_states() on either a specific layer, or on your entire
model.
Initial State of RNNs

You can specify the initial state of RNN layers symbolically by calling them with the keyword
argument initial_state. The value of initial_state should be a tensor or list of tensors repre-
senting the initial state of the RNN layer.
You can specify the initial state of RNN layers numerically by calling reset_states with the
keyword argument states. The value of states should be a numpy array or list of numpy arrays
representing the initial state of the RNN layer.
layer_input 237
References
• Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Trans-
lation
• On the Properties of Neural Machine Translation: Encoder-Decoder Approaches
• Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling
See Also
Other recurrent layers: layer_cudnn_gru, layer_cudnn_lstm, layer_lstm, layer_simple_rnn
layer_input Input layer
Description
Layer to be used as an entry point into a graph.
Usage
layer_input(shape = NULL, batch_shape = NULL, name = NULL,
dtype = NULL, sparse = FALSE, tensor = NULL)
Arguments
shape Shape, not including the batch size. For instance, shape=c(32) indicates that
the expected input will be batches of 32-dimensional vectors.
batch_shape Shape, including the batch size. For instance, shape = c(10,32) indicates that
the expected input will be batches of 10 32-dimensional vectors. batch_shape = list(NULL, 32)
indicates batches of an arbitrary number of 32-dimensional vectors.
sparse Boolean, whether the placeholder created is meant to be sparse.
tensor Existing tensor to wrap into the Input layer. If set, the layer will not create a
placeholder tensor.
Value
A tensor
See Also
layer_flatten, layer_lambda, layer_masking, layer_permute, layer_repeat_vector, layer_reshape
238 layer_lambda
layer_lambda Wraps arbitrary expression as a layer
Description
Wraps arbitrary expression as a layer
Usage
layer_lambda(object, f, output_shape = NULL, mask = NULL,
arguments = NULL, input_shape = NULL, batch_input_shape = NULL,
weights = NULL)
Arguments
f The function to be evaluated. Takes input tensor as first argument.
output_shape Expected output shape from the function (not required when using TensorFlow
back-end).
mask mask
arguments optional named list of keyword arguments to be passed to the function.
batch_input_shape
Input shape
Output shape
Arbitrary (based on tensor returned from the function)
layer_locally_connected_1d 239
See Also
layer_flatten, layer_input, layer_masking, layer_permute, layer_repeat_vector, layer_reshape
layer_locally_connected_1d
Locally-connected layer for 1D inputs.
Description
layer_locally_connected_1d() works similarly to layer_conv_1d() , except that weights are
unshared, that is, a different set of filters is applied at each different patch of the input.
Usage
layer_locally_connected_1d(object, filters, kernel_size, strides = 1L,
padding = "valid", data_format = NULL, activation = NULL,
bias_initializer = "zeros", kernel_regularizer = NULL,
kernel_constraint = NULL, bias_constraint = NULL,
Arguments
filters Integer, the dimensionality of the output space (i.e. the number output of filters
kernel_size An integer or list of a single integer, specifying the length of the 1D convolution
window.
strides An integer or list of a single integer, specifying the stride length of the con-
volution. Specifying any stride value != 1 is incompatible with specifying any
dilation_rate value != 1.
padding Currently only supports "valid" (case-insensitive). "same" may be supported
in the future.
240 layer_locally_connected_2d
kernel_initializer
bias_initializer
kernel_regularizer
bias_regularizer
kernel_constraint
bias_constraint
Input shape
3D tensor with shape: (batch_size, steps, input_dim)
Output shape
3D tensor with shape: (batch_size, new_steps, filters) steps value might have changed
due to padding or strides.
See Also
Other locally connected layers: layer_locally_connected_2d
layer_locally_connected_2d
Locally-connected layer for 2D inputs.
Description
layer_locally_connected_2d works similarly to layer_conv_2d(), except that weights are un-

shared, that is, a different set of filters is applied at each different patch of the input.
layer_locally_connected_2d 241
Usage
layer_locally_connected_2d(object, filters, kernel_size, strides = c(1L,
1L), padding = "valid", data_format = NULL, activation = NULL,
bias_initializer = "zeros", kernel_regularizer = NULL,
kernel_constraint = NULL, bias_constraint = NULL,
Arguments
filters Integer, the dimensionality of the output space (i.e. the number output of filters
dimensions.
padding Currently only supports "valid" (case-insensitive). "same" may be supported
in the future.
shape (batch, width, height, channels) while channels_first cor-
responds to inputs with shape (batch, channels, width, height). It
kernel_initializer
bias_initializer
kernel_regularizer
bias_regularizer
kernel_constraint
242 layer_lstm
bias_constraint
Input shape
Output shape
4D tensor with shape: (samples, filters, new_rows, new_cols) if data_format=’channels_first’
or 4D tensor with shape: (samples, new_rows, new_cols, filters) if data_format=’channels_last’.
See Also
Other locally connected layers: layer_locally_connected_1d
layer_lstm Long Short-Term Memory unit - Hochreiter 1997.
Description
For a step-by-step description of the algorithm, see this tutorial.
Usage
layer_lstm(object, units, activation = "tanh",
recurrent_activation = "hard_sigmoid", use_bias = TRUE,
kernel_initializer = "glorot_uniform",
recurrent_constraint = NULL, bias_constraint = NULL, dropout = 0,
recurrent_dropout = 0, input_shape = NULL,
layer_lstm 243
Arguments
return_sequences
sequence.
output.
kernel_initializer
the inputs.
bias_initializer
unit_forget_bias
Boolean. If TRUE, add 1 to the bias of the forget gate at initialization. Setting
it to true will also force bias_initializer="zeros". This is recommended in
Jozefowicz et al.
kernel_regularizer
bias_regularizer
kernel_constraint
244 layer_lstm
bias_constraint
of the inputs.
recurrent_dropout
batch_input_shape
Input shapes
Output shape
Masking

layer_masking 245
model.

References
• Long short-term memory (original 1997 paper)
• Supervised sequence labeling with recurrent neural networks
See Also
Other recurrent layers: layer_cudnn_gru, layer_cudnn_lstm, layer_gru, layer_simple_rnn
Other recurrent layers: layer_cudnn_gru, layer_cudnn_lstm, layer_gru, layer_simple_rnn
layer_masking Masks a sequence by using a mask value to skip timesteps.
Description
For each timestep in the input tensor (dimension #1 in the tensor), if all values in the input tensor at
that timestep are equal to mask_value, then the timestep will be masked (skipped) in all downstream
layers (as long as they support masking). If any downstream layer does not support masking yet
receives such an input mask, an exception will be raised.
Usage
layer_masking(object, mask_value = 0, input_shape = NULL,
246 layer_maximum
Arguments
mask_value float, mask value
batch_input_shape
See Also
layer_flatten, layer_input, layer_lambda, layer_permute, layer_repeat_vector, layer_reshape
layer_maximum Layer that computes the maximum (element-wise) a list of inputs.
Description
shape).
Usage
layer_maximum(inputs, batch_size = NULL, dtype = NULL, name = NULL,
Arguments
layer_max_pooling_1d 247
Value
A tensor, the element-wise maximum of the inputs.
See Also
Other merge layers: layer_add, layer_average, layer_concatenate, layer_dot, layer_minimum,
layer_max_pooling_1d Max pooling operation for temporal data.
Description
Max pooling operation for temporal data.
Usage
layer_max_pooling_1d(object, pool_size = 2L, strides = NULL,
padding = "valid", batch_size = NULL, name = NULL,
Arguments
pool_size Integer, size of the max pooling windows.
strides Integer, or NULL. Factor by which to downscale. E.g. 2 will halve the input. If
NULL, it will default to pool_size.
Input shape
Output shape
3D tensor with shape: (batch_size, downsampled_steps, features).
248 layer_max_pooling_2d
See Also
layer_global_max_pooling_1d, layer_global_max_pooling_2d, layer_global_max_pooling_3d,
layer_max_pooling_2d, layer_max_pooling_3d
layer_max_pooling_2d Max pooling operation for spatial data.
Description
Max pooling operation for spatial data.
Usage
layer_max_pooling_2d(object, pool_size = c(2L, 2L), strides = NULL,
Arguments
pool_size integer or list of 2 integers, factors by which to downscale (vertical, horizontal).
(2, 2) will halve the input in both spatial dimension. If only one integer is
specified, the same window length will be used for both dimensions.
strides Integer, list of 2 integers, or NULL. Strides values. If NULL, it will default to
pool_size.
Input shape
layer_max_pooling_3d 249
Output shape
• If data_format='channels_last': 4D tensor with shape: (batch_size, pooled_rows, pooled_cols, channels)
• If data_format='channels_first': 4D tensor with shape: (batch_size, channels, pooled_rows, pooled_cols
See Also
layer_max_pooling_3d Max pooling operation for 3D data (spatial or spatio-temporal).
Description
Max pooling operation for 3D data (spatial or spatio-temporal).
Usage
layer_max_pooling_3d(object, pool_size = c(2L, 2L, 2L), strides = NULL,
Arguments
pool_size list of 3 integers, factors by which to downscale (dim1, dim2, dim3). (2, 2, 2)
will halve the size of the 3D input in each dimension.
strides list of 3 integers, or NULL. Strides values.
250 layer_minimum
Input shape
Output shape
• If data_format='channels_last': 5D tensor with shape: (batch_size, pooled_dim1, pooled_dim2, pooled_di
• If data_format='channels_first': 5D tensor with shape: (batch_size, channels, pooled_dim1, pooled_dim2
See Also
layer_minimum Layer that computes the minimum (element-wise) a list of inputs.
Description
shape).
Usage
layer_minimum(inputs, batch_size = NULL, dtype = NULL, name = NULL,
Arguments
Value
A tensor, the element-wise maximum of the inputs.
See Also
Other merge layers: layer_add, layer_average, layer_concatenate, layer_dot, layer_maximum,
layer_multiply 251
layer_multiply Layer that multiplies (element-wise) a list of inputs.
Description
shape).
Usage
layer_multiply(inputs, batch_size = NULL, dtype = NULL, name = NULL,
Arguments
Value
A tensor, the element-wise product of the inputs.
See Also
layer_minimum, layer_subtract
layer_permute Permute the dimensions of an input according to a given pattern
Description
Permute the dimensions of an input according to a given pattern
Usage
layer_permute(object, dims, input_shape = NULL,
252 layer_repeat_vector
Arguments
dims List of integers. Permutation pattern, does not include the samples dimension.
Indexing starts at 1. For instance, (2, 1) permutes the first and second dimen-
sion of the input.
batch_input_shape

Input shape: Arbitrary
Output shape: Same as the input shape, but with the dimensions re-ordered according to the speci-
fied pattern.
Note
Useful for e.g. connecting RNNs and convnets together.
See Also
layer_flatten, layer_input, layer_lambda, layer_masking, layer_repeat_vector, layer_reshape
layer_repeat_vector Repeats the input n times.
Description
Repeats the input n times.
Usage
layer_repeat_vector(object, n, batch_size = NULL, name = NULL,
layer_reshape 253
Arguments
n integer, repetition factor.
Input shape
2D tensor of shape (num_samples, features).
Output shape
3D tensor of shape (num_samples, n, features).
See Also
layer_flatten, layer_input, layer_lambda, layer_masking, layer_permute, layer_reshape
layer_reshape Reshapes an output to a certain shape.
Description
Reshapes an output to a certain shape.
Usage
layer_reshape(object, target_shape, input_shape = NULL,
Arguments
target_shape List of integers, does not include the samples dimension (batch size).
batch_input_shape
254 layer_separable_conv_1d


Input shape: Arbitrary, although all dimensions in the input shaped must be fixed.
Output shape: (batch_size,) + target_shape.
See Also
layer_flatten, layer_input, layer_lambda, layer_masking, layer_permute, layer_repeat_vector
layer_separable_conv_1d
Depthwise separable 1D convolution.
Description
Separable convolutions consist in first performing a depthwise spatial convolution (which acts on
each input channel separately) followed by a pointwise convolution which mixes together the re-
sulting output channels. The depth_multiplier argument controls how many output channels are
generated per input channel in the depthwise step. Intuitively, separable convolutions can be under-
stood as a way to factorize a convolution kernel into two smaller kernels, or as an extreme version
of an Inception block.
Usage
layer_separable_conv_1d(object, filters, kernel_size, strides = 1,
dilation_rate = 1, depth_multiplier = 1, activation = NULL,
use_bias = TRUE, depthwise_initializer = "glorot_uniform",
pointwise_initializer = "glorot_uniform", bias_initializer = "zeros",
depthwise_regularizer = NULL, pointwise_regularizer = NULL,
depthwise_constraint = NULL, pointwise_constraint = NULL,
layer_separable_conv_1d 255
Arguments
dimensions.
depth_multiplier
filterss_in * depth_multiplier.
pointwise_initializer
Initializer for the pointwise kernel matrix.
bias_initializer
pointwise_regularizer
Regularizer function applied to the pointwise kernel matrix.
bias_regularizer
pointwise_constraint
Constraint function applied to the pointwise kernel matrix.
bias_constraint
batch_input_shape
Input shape
3D tensor with shape: (batch, channels, steps) if data_format=’channels_first’ or 3D tensor

with shape: (batch, steps, channels) if data_format=’channels_last’.
Output shape
3D tensor with shape: (batch, filters, new_steps) if data_format=’channels_first’ or 3D

tensor with shape: (batch, new_steps, filters) if data_format=’channels_last’. new_steps
values might have changed due to padding or strides.
See Also

layer_depthwise_conv_2d, layer_separable_conv_2d, layer_upsampling_1d, layer_upsampling_2d,
layer_separable_conv_2d
Separable 2D convolution.
layer_separable_conv_2d 257
Description
Separable convolutions consist in first performing a depthwise spatial convolution (which acts on
each input channel separately) followed by a pointwise convolution which mixes together the re-
sulting output channels. The depth_multiplier argument controls how many output channels are
generated per input channel in the depthwise step. Intuitively, separable convolutions can be under-
stood as a way to factorize a convolution kernel into two smaller kernels, or as an extreme version
of an Inception block.
Usage
layer_separable_conv_2d(object, filters, kernel_size, strides = c(1, 1),

padding = "valid", data_format = NULL, dilation_rate = 1,
depth_multiplier = 1, activation = NULL, use_bias = TRUE,
depthwise_initializer = "glorot_uniform",
pointwise_initializer = "glorot_uniform", bias_initializer = "zeros",
depthwise_regularizer = NULL, pointwise_regularizer = NULL,
depthwise_constraint = NULL, pointwise_constraint = NULL,
Arguments

dimensions.
depth_multiplier
filters_in * depth_multiplier.
pointwise_initializer
Initializer for the pointwise kernel matrix.
bias_initializer
pointwise_regularizer
Regularizer function applied to the pointwise kernel matrix.
bias_regularizer
pointwise_constraint
Constraint function applied to the pointwise kernel matrix.
bias_constraint
batch_input_shape
Input shape
4D tensor with shape: (batch, channels, rows, cols) if data_format=’channels_first’ or 4D
tensor with shape: (batch, rows, cols, channels) if data_format=’channels_last’.
layer_simple_rnn 259
Output shape
4D tensor with shape: (batch, filters, new_rows, new_cols) if data_format=’channels_first’
or 4D tensor with shape: (batch, new_rows, new_cols, filters) if data_format=’channels_last’.
See Also
layer_depthwise_conv_2d, layer_separable_conv_1d, layer_upsampling_1d, layer_upsampling_2d,
layer_simple_rnn Fully-connected RNN where the output is to be fed back to input.
Description
Fully-connected RNN where the output is to be fed back to input.
Usage
layer_simple_rnn(object, units, activation = "tanh", use_bias = TRUE,
kernel_initializer = "glorot_uniform",
bias_constraint = NULL, dropout = 0, recurrent_dropout = 0,
Arguments
return_sequences
sequence.
output.
260 layer_simple_rnn
kernel_initializer
the inputs.
bias_initializer
kernel_regularizer
bias_regularizer
kernel_constraint
bias_constraint
of the inputs.
recurrent_dropout
batch_input_shape
layer_simple_rnn 261

Input shapes
Output shape
Masking
model.

References
262 layer_spatial_dropout_1d
See Also
Other recurrent layers: layer_cudnn_gru, layer_cudnn_lstm, layer_gru, layer_lstm
layer_spatial_dropout_1d
Spatial 1D version of Dropout.
Description
This version performs the same function as Dropout, however it drops entire 1D feature maps in-
stead of individual elements. If adjacent frames within feature maps are strongly correlated (as is
normally the case in early convolution layers) then regular dropout will not regularize the activations
and will otherwise just result in an effective learning rate decrease. In this case, layer_spatial_dropout_1d
will help promote independence between feature maps and should be used instead.
Usage
layer_spatial_dropout_1d(object, rate, batch_size = NULL, name = NULL,
Arguments
Input shape
3D tensor with shape: (samples, timesteps, channels)
Output shape
Same as input
References
- Efficient Object Localization Using Convolutional Networks
See Also
Other dropout layers: layer_dropout, layer_spatial_dropout_2d, layer_spatial_dropout_3d
layer_spatial_dropout_2d 263
Description
stead of individual elements. If adjacent pixels within feature maps are strongly correlated (as is
Usage
layer_spatial_dropout_2d(object, rate, data_format = NULL,
Arguments
mension (the depth) is at index 1, in ’channels_last’ mode is it at index 3. It
Input shape
Output shape
Same as input
References
See Also
264 layer_spatial_dropout_3d
Description
stead of individual elements. If adjacent voxels within feature maps are strongly correlated (as is
Usage
layer_spatial_dropout_3d(object, rate, data_format = NULL,
Arguments
mension (the depth) is at index 1, in ’channels_last’ mode is it at index 4. It
Input shape
5D tensor with shape: (samples, channels, dim1, dim2, dim3) if data_format=’channels_first’
or 5D tensor with shape: (samples, dim1, dim2, dim3, channels) if data_format=’channels_last’.
Output shape
Same as input
References
See Also
layer_subtract 265
layer_subtract Layer that subtracts two inputs.
Description
It takes as input a list of tensors of size 2, both of the same shape, and returns a single tensor,
(inputs[[1]] - inputs[[2]]), also of the same shape.
Usage
layer_subtract(inputs, batch_size = NULL, dtype = NULL, name = NULL,
Arguments
inputs A list of input tensors (exactly 2).
Value
A tensor, the difference of the inputs.
See Also
layer_minimum, layer_multiply
layer_upsampling_1d Upsampling layer for 1D inputs.
Description
Repeats each temporal step size times along the time axis.
Usage
layer_upsampling_1d(object, size = 2L, batch_size = NULL,
266 layer_upsampling_2d
Arguments
size integer. Upsampling factor.
Input shape
3D tensor with shape: (batch, steps, features).
Output shape
3D tensor with shape: (batch, upsampled_steps, features).
See Also
Description
Repeats the rows and columns of the data by size[[0]] and size[[1]] respectively.
Usage
layer_upsampling_2d(object, size = c(2L, 2L), data_format = NULL,
interpolation = "nearest", batch_size = NULL, name = NULL,
Arguments
size int, or list of 2 integers. The upsampling factors for rows and columns.
layer_upsampling_3d 267
interpolation A string, one of nearest or bilinear. Note that CNTK does not support yet
the bilinear upscaling and that with Theano, only size=(2, 2) is possible.
Input shape

Output shape
• If data_format is "channels_last": (batch, upsampled_rows, upsampled_cols, channels)

• If data_format is "channels_first": (batch, channels, upsampled_rows, upsampled_cols)
See Also

Description
Repeats the 1st, 2nd and 3rd dimensions of the data by size[[0]], size[[1]] and size[[2]]
respectively.
Usage
layer_upsampling_3d(object, size = c(2L, 2L, 2L), data_format = NULL,

268 layer_zero_padding_1d
Arguments
size int, or list of 3 integers. The upsampling factors for dim1, dim2 and dim3.
Input shape
• If data_format is "channels_last": (batch, dim1, dim2, dim3, channels)

• If data_format is "channels_first": (batch, channels, dim1, dim2, dim3)
Output shape
• If data_format is "channels_last": (batch, upsampled_dim1, upsampled_dim2, upsampled_dim3, channels)

• If data_format is "channels_first": (batch, channels, upsampled_dim1, upsampled_dim2, upsampled_dim3
See Also
layer_zero_padding_1d Zero-padding layer for 1D input (e.g. temporal sequence).
Description
Zero-padding layer for 1D input (e.g. temporal sequence).
layer_zero_padding_2d 269
Usage
layer_zero_padding_1d(object, padding = 1L, batch_size = NULL,
Arguments
padding int, or list of int (length 2)
• If int: How many zeros to add at the beginning and end of the padding
dimension (axis 1).
• If list of int (length 2): How many zeros to add at the beginning and at the
end of the padding dimension ((left_pad, right_pad)).
Input shape
3D tensor with shape (batch, axis_to_pad, features)
Output shape
3D tensor with shape (batch, padded_axis, features)
See Also
layer_upsampling_2d, layer_upsampling_3d, layer_zero_padding_2d, layer_zero_padding_3d
layer_zero_padding_2d Zero-padding layer for 2D input (e.g. picture).
Description
This layer can add rows and columns of zeros at the top, bottom, left and right side of an image
tensor.
Usage
layer_zero_padding_2d(object, padding = c(1L, 1L), data_format = NULL,
270 layer_zero_padding_2d
Arguments

padding int, or list of 2 ints, or list of 2 lists of 2 ints.
• If int: the same symmetric padding is applied to width and height.
• If list of 2 ints: interpreted as two different symmetric padding values for
height and width: (symmetric_height_pad, symmetric_width_pad).
• If list of 2 lists of 2 ints: interpreted as ((top_pad, bottom_pad), (left_pad, right_pad))
Input shape

Output shape
• If data_format is "channels_last": (batch, padded_rows, padded_cols, channels)

• If data_format is "channels_first": (batch, channels, padded_rows, padded_cols)
See Also

layer_zero_padding_3d 271
layer_zero_padding_3d Zero-padding layer for 3D data (spatial or spatio-temporal).
Description
Zero-padding layer for 3D data (spatial or spatio-temporal).
Usage
layer_zero_padding_3d(object, padding = c(1L, 1L, 1L),
data_format = NULL, batch_size = NULL, name = NULL,
Arguments
padding int, or list of 3 ints, or list of 3 lists of 2 ints.
• If int: the same symmetric padding is applied to width and height.
• If list of 3 ints: interpreted as three different symmetric padding values:
(symmetric_dim1_pad, symmetric_dim2_pad, symmetric_dim3_pad).
• If list of 3 lists of 2 ints: interpreted as ((left_dim1_pad, right_dim1_pad), (left_dim2_pad, r
Input shape
• If data_format is "channels_last": (batch, first_axis_to_pad, second_axis_to_pad, third_axis_to_pad,
• If data_format is "channels_first": (batch, depth, first_axis_to_pad, second_axis_to_pad, third_axis
Output shape
• If data_format is "channels_last": (batch, first_padded_axis, second_padded_axis, third_axis_to_pad,
• If data_format is "channels_first": (batch, depth, first_padded_axis, second_padded_axis, third_axis
272 loss_mean_squared_error
See Also

loss_mean_squared_error
Model loss functions
Description
Model loss functions
Usage
loss_mean_squared_error(y_true, y_pred)
loss_mean_absolute_error(y_true, y_pred)
loss_mean_absolute_percentage_error(y_true, y_pred)
loss_mean_squared_logarithmic_error(y_true, y_pred)
loss_squared_hinge(y_true, y_pred)
loss_hinge(y_true, y_pred)
loss_categorical_hinge(y_true, y_pred)
loss_logcosh(y_true, y_pred)
loss_categorical_crossentropy(y_true, y_pred)
loss_sparse_categorical_crossentropy(y_true, y_pred)
loss_binary_crossentropy(y_true, y_pred)
loss_kullback_leibler_divergence(y_true, y_pred)
loss_poisson(y_true, y_pred)
loss_cosine_proximity(y_true, y_pred)
make_sampling_table 273
Arguments
y_true True labels (Tensor)
y_pred Predictions (Tensor of the same shape as y_true)
Details
Loss functions are to be supplied in the loss parameter of the compile.keras.engine.training.Model()
function.
Loss functions can be specified either using the name of a built in loss function (e.g. ’loss = bi-
nary_crossentropy’), a reference to a built in loss function (e.g. ’loss = loss_binary_crossentropy()’)
or by passing an artitrary function that returns a scalar for each data-point and takes the following
two arguments:
• y_true True labels (Tensor)
• y_pred Predictions (Tensor of the same shape as y_true)
The actual optimized objective is the mean of the output array across all datapoints.
Categorical Crossentropy
When using the categorical_crossentropy loss, your targets should be in categorical format (e.g. if
you have 10 classes, the target for each sample should be a 10-dimensional vector that is all-zeros
except for a 1 at the index corresponding to the class of the sample). In order to convert integer
targets into categorical targets, you can use the Keras utility function to_categorical():
categorical_labels <- to_categorical(int_labels, num_classes = NULL)
loss_logcosh
log(cosh(x)) is approximately equal to (x ** 2) / 2 for small x and to abs(x) - log(2)
for large x. This means that ’logcosh’ works mostly like the mean squared error, but will not be so
strongly affected by the occasional wildly incorrect prediction. However, it may return NaNs if the
intermediate value cosh(y_pred - y_true) is too large to be represented in the chosen precision.
See Also
compile.keras.engine.training.Model()
make_sampling_table Generates a word rank-based probabilistic sampling table.
Description
Generates a word rank-based probabilistic sampling table.
Usage
make_sampling_table(size, sampling_factor = 1e-05)
274 metric_binary_accuracy
Arguments
size Int, number of possible words to sample.

sampling_factor
The sampling factor in the word2vec formula.
Details
Used for generating the sampling_table argument for skipgrams(). sampling_table[[i]] is

the probability of sampling the word i-th most common word in a dataset (more common words
should be sampled less frequently, for balance).
The sampling probabilities are generated according to the sampling distribution used in word2vec:
p(word) = min(1, sqrt(word_frequency / sampling_factor) / (word_frequency / sampling_factor))
We assume that the word frequencies follow Zipf’s law (s=1) to derive a numerical approximation
of frequency(rank):
frequency(rank) ~ 1/(rank * (log(rank) + gamma) + 1/2 - 1/(12*rank))
where gamma is the Euler-Mascheroni constant.
Value
An array of length size where the ith entry is the probability that a word of rank i should be
sampled.
Note
The word2vec formula is: p(word) = min(1, sqrt(word.frequency/sampling_factor) / (word.frequency/sampling_factor))
See Also
Other text preprocessing: pad_sequences, skipgrams, text_hashing_trick, text_one_hot,

text_to_word_sequence
metric_binary_accuracy
Model performance metrics
Description
Model performance metrics

metric_binary_accuracy 275
Usage
metric_binary_accuracy(y_true, y_pred)
metric_binary_crossentropy(y_true, y_pred)
metric_categorical_accuracy(y_true, y_pred)
metric_categorical_crossentropy(y_true, y_pred)
metric_cosine_proximity(y_true, y_pred)
metric_hinge(y_true, y_pred)
metric_kullback_leibler_divergence(y_true, y_pred)
metric_mean_absolute_error(y_true, y_pred)
metric_mean_absolute_percentage_error(y_true, y_pred)
metric_mean_squared_error(y_true, y_pred)
metric_mean_squared_logarithmic_error(y_true, y_pred)
metric_poisson(y_true, y_pred)
metric_sparse_categorical_crossentropy(y_true, y_pred)
metric_squared_hinge(y_true, y_pred)
metric_top_k_categorical_accuracy(y_true, y_pred, k = 5)
metric_sparse_top_k_categorical_accuracy(y_true, y_pred, k = 5)
custom_metric(name, metric_fn)
Arguments
y_true True labels (tensor)
y_pred Predictions (tensor of the same shape as y_true).
k An integer, number of top elements to consider.
name Name of custom metric
metric_fn Custom metric function
Custom Metrics
You can provide an arbitrary R function as a custom metric. Note that the y_true and y_pred
parameters are tensors, so computations on them should use backend tensor functions.
276 metric_binary_accuracy
Use the custom_metric() function to define a custom metric. Note that a name (’mean_pred’) is
provided for the custom metric function: this name is used within training progress output. See
below for an example.
If you want to save and load a model with custom metrics, you should also specify the metric in the
call the load_model_hdf5(). For example: load_model_hdf5("my_model.h5", c('mean_pred' = metric_mean_pred))
Alternatively, you can wrap all of your code in a call to with_custom_object_scope() which will
allow you to refer to the metric by name just like you do with built in keras metrics.
Documentation on the available backend tensor functions can be found at https://keras.rstudio.
com/articles/backend.html#backend-functions.
Metrics with Parameters

To use metrics with parameters (e.g. metric_top_k_categorical_accurary()) you should create
a custom metric that wraps the call with the parameter. See below for an example.
Note
Metric functions are to be supplied in the metrics parameter of the compile.keras.engine.training.Model()
function.
Examples
## Not run:
# create metric using backend tensor functions

metric_mean_pred <- custom_metric("mean_pred", function(y_true, y_pred) {
k_mean(y_pred)
})
model %>% compile(

optimizer = optimizer_rmsprop(),
loss = loss_binary_crossentropy,
metrics = c('accuracy', metric_mean_pred)
)
# create custom metric to wrap metric with parameter

metric_top_3_categorical_accuracy <-
custom_metric("top_3_categorical_accuracy", function(y_true, y_pred) {
})
model %>% compile(

optimizer = optimizer_rmsprop(),
metrics = metric_top_3_categorical_accuracy
)
## End(Not run)
model_to_json 277
model_to_json Model configuration as JSON
Description
Save and re-load models configurations as JSON. Note that the representation does not include the
weights, only the architecture.
Usage
model_to_json(object)
model_from_json(json, custom_objects = NULL)
Arguments
object Model object to save
json JSON with model configuration
custom_objects Optional named list mapping names to custom classes or functions to be consid-
ered during deserialization.
See Also
Other model persistence: get_weights, model_to_yaml, save_model_hdf5, save_model_weights_hdf5,
serialize_model
model_to_yaml Model configuration as YAML
Description
Save and re-load models configurations as YAML Note that the representation does not include the
weights, only the architecture.
Usage
model_to_yaml(object)
model_from_yaml(yaml, custom_objects = NULL)
Arguments
yaml YAML with model configuration
custom_objects Optional named list mapping names to custom classes or functions to be consid-
ered during deserialization.
278 multi_gpu_model
See Also
Other model persistence: get_weights, model_to_json, save_model_hdf5, save_model_weights_hdf5,
serialize_model
multi_gpu_model Replicates a model on different GPUs.
Description
Replicates a model on different GPUs.
Usage
multi_gpu_model(model, gpus = NULL, cpu_merge = TRUE,
cpu_relocation = FALSE)
Arguments
model A Keras model instance. To avoid OOM errors, this model could have been built
on CPU, for instance (see usage example below).
gpus NULL to use all available GPUs (default). Integer >= 2 or list of integers, number
of GPUs or list of GPU IDs on which to create model replicas.
cpu_merge A boolean value to identify whether to force merging model weights under the
scope of the CPU or not.
cpu_relocation A boolean value to identify whether to create the model’s weights under the
scope of the CPU. If the model is not defined under any preceding device scope,
you can still rescue it by activating this option.
Details
Specifically, this function implements single-machine multi-GPU data parallelism. It works in the
following way:
• Divide the model’s input(s) into multiple sub-batches.

• Apply a model copy on each sub-batch. Every model copy is executed on a dedicated GPU.
• Concatenate the results (on CPU) into one big batch.
E.g. if your batch_size is 64 and you use gpus=2, then we will divide the input into 2 sub-batches
of 32 samples, process each sub-batch on one GPU, then return the full batch of 64 processed
samples.
This induces quasi-linear speedup on up to 8 GPUs.
This function is only available with the TensorFlow backend for the time being.
multi_gpu_model 279
Value
A Keras model object which can be used just like the initial model argument, but which distributes
its workload on multiple GPUs.
Model Saving
To save the multi-gpu model, use save_model_hdf5() or save_model_weights_hdf5() with the
template model (the argument you passed to multi_gpu_model), rather than the model returned by
multi_gpu_model.
See Also
keras_model_sequential, keras_model, pop_layer, predict.keras.engine.training.Model,
train_on_batch
Examples
## Not run:
library(keras)
library(tensorflow)
num_samples <- 1000

height <- 224
width <- 224
num_classes <- 1000
# Instantiate the base model (or "template" model).

# We recommend doing this with under a CPU device scope,
# so that the model's weights are hosted on CPU memory.
# Otherwise they may end up hosted on a GPU, which would
# complicate weight sharing.
with(tf$device("/cpu:0"), {
model <- application_xception(
weights = NULL,
input_shape = c(height, width, 3),
classes = num_classes
)
})
# Replicates the model on 8 GPUs.

# This assumes that your machine has 8 available GPUs.
parallel_model <- multi_gpu_model(model, gpus = 8)
parallel_model %>% compile(
loss = "categorical_crossentropy",
optimizer = "rmsprop"
)
280 normalize
# Generate dummy data.

x <- array(runif(num_samples * height * width*3),
dim = c(num_samples, height, width, 3))
y <- array(runif(num_samples * num_classes),
dim = c(num_samples, num_classes))
# This `fit` call will be distributed on 8 GPUs.

# Since the batch size is 256, each GPU will process 32 samples.
parallel_model %>% fit(x, y, epochs = 20, batch_size = 256)
# Save model via the template model (which shares the same weights):
model %>% save_model_hdf5("my_model.h5")
## End(Not run)
normalize Normalize a matrix or nd-array
Description
Normalize a matrix or nd-array
Usage
normalize(x, axis = -1, order = 2)
Arguments
x Matrix or array to normalize

axis Axis along which to normalize. Axis indexes are 1-based (pass -1 to select the
last axis).
order Normalization order (e.g. 2 for L2 norm)
Value
A normalized copy of the array.

optimizer_adadelta 281
optimizer_adadelta Adadelta optimizer.
Description
Adadelta optimizer as described in ADADELTA: An Adaptive Learning Rate Method.
Usage
optimizer_adadelta(lr = 1, rho = 0.95, epsilon = NULL, decay = 0,
clipnorm = NULL, clipvalue = NULL)
Arguments
lr float >= 0. Learning rate.
rho float >= 0. Decay factor.
epsilon float >= 0. Fuzz factor. If NULL, defaults to k_epsilon().
decay float >= 0. Learning rate decay over each update.
clipnorm Gradients will be clipped when their L2 norm exceeds this value.
clipvalue Gradients will be clipped when their absolute value exceeds this value.
Note
It is recommended to leave the parameters of this optimizer at their default values.
See Also
Other optimizers: optimizer_adagrad, optimizer_adamax, optimizer_adam, optimizer_nadam,
optimizer_rmsprop, optimizer_sgd
optimizer_adagrad Adagrad optimizer.
Description
Adagrad optimizer as described in Adaptive Subgradient Methods for Online Learning and Stochas-
tic Optimization.
Usage
optimizer_adagrad(lr = 0.01, epsilon = NULL, decay = 0,
282 optimizer_adam
Arguments
Note
It is recommended to leave the parameters of this optimizer at their default values.
See Also
Other optimizers: optimizer_adadelta, optimizer_adamax, optimizer_adam, optimizer_nadam,
optimizer_adam Adam optimizer
Description
Adam optimizer as described in Adam - A Method for Stochastic Optimization.
Usage
optimizer_adam(lr = 0.001, beta_1 = 0.9, beta_2 = 0.999,
epsilon = NULL, decay = 0, amsgrad = FALSE, clipnorm = NULL,
clipvalue = NULL)
Arguments
beta_1 The exponential decay rate for the 1st moment estimates. float, 0 < beta < 1.
Generally close to 1.
beta_2 The exponential decay rate for the 2nd moment estimates. float, 0 < beta < 1.
amsgrad Whether to apply the AMSGrad variant of this algorithm from the paper "On the
Convergence of Adam and Beyond".
optimizer_adamax 283
References
• Adam - A Method for Stochastic Optimization
• On the Convergence of Adam and Beyond
Note
Default parameters follow those provided in the original paper.
See Also
Other optimizers: optimizer_adadelta, optimizer_adagrad, optimizer_adamax, optimizer_nadam,
optimizer_adamax Adamax optimizer
Description
Adamax optimizer from Section 7 of the Adam paper. It is a variant of Adam based on the infinity
norm.
Usage
optimizer_adamax(lr = 0.002, beta_1 = 0.9, beta_2 = 0.999,
epsilon = NULL, decay = 0, clipnorm = NULL, clipvalue = NULL)
Arguments
See Also
Other optimizers: optimizer_adadelta, optimizer_adagrad, optimizer_adam, optimizer_nadam,
284 optimizer_nadam
optimizer_nadam Nesterov Adam optimizer
Description
Much like Adam is essentially RMSprop with momentum, Nadam is Adam RMSprop with Nesterov
momentum.
Usage
optimizer_nadam(lr = 0.002, beta_1 = 0.9, beta_2 = 0.999,

epsilon = NULL, schedule_decay = 0.004, clipnorm = NULL,
clipvalue = NULL)
Arguments

schedule_decay Schedule deacy.
Details
Default parameters follow those provided in the paper. It is recommended to leave the parameters
of this optimizer at their default values.
See Also
On the importance of initialization and momentum in deep learning.

Other optimizers: optimizer_adadelta, optimizer_adagrad, optimizer_adamax, optimizer_adam,
optimizer_rmsprop 285
optimizer_rmsprop RMSProp optimizer
Description
RMSProp optimizer
Usage
optimizer_rmsprop(lr = 0.001, rho = 0.9, epsilon = NULL, decay = 0,
Arguments
rho float >= 0. Decay factor.
Note
It is recommended to leave the parameters of this optimizer at their default values (except the
learning rate, which can be freely tuned).
This optimizer is usually a good choice for recurrent neural networks.
See Also
optimizer_nadam, optimizer_sgd
optimizer_sgd Stochastic gradient descent optimizer
Description
Stochastic gradient descent optimizer with support for momentum, learning rate decay, and Nes-
terov momentum.
Usage
optimizer_sgd(lr = 0.01, momentum = 0, decay = 0, nesterov = FALSE,
286 pad_sequences
Arguments

momentum float >= 0. Parameter that accelerates SGD in the relevant direction and dampens
oscillations.
nesterov boolean. Whether to apply Nesterov momentum.
Value
Optimizer for use with compile.keras.engine.training.Model.
See Also

optimizer_nadam, optimizer_rmsprop
pad_sequences Pads sequences to the same length
Description
Pads sequences to the same length
Usage
pad_sequences(sequences, maxlen = NULL, dtype = "int32",

padding = "pre", truncating = "pre", value = 0)
Arguments
sequences List of lists where each element is a sequence

maxlen int, maximum length of all sequences
dtype type of the output sequences
padding ’pre’ or ’post’, pad either before or after each sequence.
truncating ’pre’ or ’post’, remove values from sequences larger than maxlen either in the
beginning or in the end of the sequence
value float, padding value
plot.keras_training_history 287
Details
This function transforms a list of num_samples sequences (lists of integers) into a matrix of shape
(num_samples, num_timesteps). num_timesteps is either the maxlen argument if provided, or
the length of the longest sequence otherwise.
Sequences that are shorter than num_timesteps are padded with value at the end.
Sequences longer than num_timesteps are truncated so that they fit the desired length. The position
where padding or truncation happens is determined by the arguments padding and truncating,
respectively.
Pre-padding is the default.
Value
Matrix with dimensions (number_of_sequences, maxlen)
See Also
Other text preprocessing: make_sampling_table, skipgrams, text_hashing_trick, text_one_hot,
plot.keras_training_history
Plot training history
Description
Plots metrics recorded during training.
Usage
## S3 method for class 'keras_training_history'
plot(x, y, metrics = NULL,
method = c("auto", "ggplot2", "base"),
smooth = getOption("keras.plot.history.smooth", TRUE),
theme_bw = getOption("keras.plot.history.theme_bw", FALSE), ...)
Arguments
x Training history object returned from fit.keras.engine.training.Model().
y Unused.
metrics One or more metrics to plot (e.g. c('loss', 'accuracy')). Defaults to plot-
ting all captured metrics.
method Method to use for plotting. The default "auto" will use ggplot2 if available, and
otherwise will use base graphics.
288 predict.keras.engine.training.Model
smooth Whether a loess smooth should be added to the plot, only available for the
ggplot2 method. If the number of epochs is smaller than ten, it is forced to
false.
theme_bw Use ggplot2::theme_bw() to plot the history in black and white.
... Additional parameters to pass to the plot() method.
pop_layer Remove the last layer in a model
Description
Remove the last layer in a model
Usage
pop_layer(object)
Arguments
See Also
keras_model_sequential, keras_model, multi_gpu_model, predict.keras.engine.training.Model,
train_on_batch
predict.keras.engine.training.Model
Generate predictions from a Keras model
Description
Generates output predictions for the input samples, processing the samples in a batched way.
Usage
predict(object, x,
batch_size = NULL, verbose = 0, steps = NULL, ...)
predict_generator 289
Arguments
object Keras model
x Input data (vector, matrix, or array)
batch_size Integer. If unspecified, it will default to 32.
verbose Verbosity mode, 0 or 1.
finished. Ignored with the default value of NULL.
... Unused
Value
vector, matrix, or array of predictions
See Also
keras_model_sequential, keras_model, multi_gpu_model, pop_layer, predict_generator,
predict_generator Generates predictions for the input samples from a data generator.
Description
The generator should return the same kind of data as accepted by predict_on_batch().
Usage
predict_generator(object, generator, steps, max_queue_size = 10,
workers = 1, verbose = 0)
Arguments
generator Generator yielding batches of input samples.
steps Total number of steps (batches of samples) to yield from generator before
stopping.
default to 10.
290 predict_on_batch
Value
Numpy array(s) of predictions.
Raises
ValueError: In case the generator yields data in an invalid format.
See Also

keras_model_sequential, keras_model, multi_gpu_model, pop_layer, predict.keras.engine.training.Model,
predict_on_batch Returns predictions for a single batch of samples.
Description
Returns predictions for a single batch of samples.
Usage
predict_on_batch(object, x)
Arguments

Value
array of predictions.
See Also

predict_generator, predict_proba, summary.keras.engine.training.Model, train_on_batch
predict_proba 291
predict_proba Generates probability or class probability predictions for the input

samples.
Description
Generates probability or class probability predictions for the input samples.
Usage
predict_proba(object, x, batch_size = NULL, verbose = 0,

steps = NULL)
predict_classes(object, x, batch_size = NULL, verbose = 0,

steps = NULL)
Arguments

batch_size Integer. If unspecified, it will default to 32.
verbose Verbosity mode, 0 or 1.
finished. The default NULL is equal to the number of samples in your dataset
divided by the batch size.
Details
The input samples are processed batch by batch.
See Also

predict_generator, predict_on_batch, summary.keras.engine.training.Model, train_on_batch
292 reset_states
regularizer_l1 L1 and L2 regularization
Description
L1 and L2 regularization
Usage
regularizer_l1(l = 0.01)
regularizer_l2(l = 0.01)
regularizer_l1_l2(l1 = 0.01, l2 = 0.01)
Arguments
l Regularization factor.
l1 L1 regularization factor.
l2 L2 regularization factor.
reset_states Reset the states for a layer
Description
Reset the states for a layer
Usage
reset_states(object)
Arguments
See Also
Other layer methods: count_params, get_config, get_input_at, get_weights
save_model_hdf5 293
save_model_hdf5 Save/Load models using HDF5 files
Description
Save/Load models using HDF5 files
Usage
save_model_hdf5(object, filepath, overwrite = TRUE,
include_optimizer = TRUE)
load_model_hdf5(filepath, custom_objects = NULL, compile = TRUE)
Arguments
filepath File path
overwrite Overwrite existing file if necessary
include_optimizer
If TRUE, save optimizer’s state.
custom_objects Mapping class names (or function names) of custom (non-Keras) objects to
class/functions (for example, custom metrics or custom loss functions).
compile Whether to compile the model after loading.
Details
The following components of the model are saved:
• The model architecture, allowing to re-instantiate the model.

• The model weights.
• The state of the optimizer, allowing to resume training exactly where you left off. This allows
you to save the entirety of the state of a model in a single file.
Saved models can be reinstantiated via load_model_hdf5(). The model returned by load_model_hdf5()
is a compiled model ready to be used (unless the saved model was never compiled in the first place
or compile = FALSE is specified).
As an alternative to providing the custom_objects argument, you can execute the definition and
persistence of your model using the with_custom_object_scope() function.
Note
The serialize_model() function enables saving Keras models to R objects that can be persisted
across R sessions.
294 save_model_weights_hdf5
See Also
Other model persistence: get_weights, model_to_json, model_to_yaml, save_model_weights_hdf5,
serialize_model
save_model_weights_hdf5
Save/Load model weights using HDF5 files
Description
Save/Load model weights using HDF5 files
Usage
save_model_weights_hdf5(object, filepath, overwrite = TRUE)
load_model_weights_hdf5(object, filepath, by_name = FALSE,

skip_mismatch = FALSE, reshape = FALSE)
Arguments
object Model object to save/load
filepath Path to the file
overwrite Whether to silently overwrite any existing file at the target location
by_name Whether to load weights by name or by topological order.
skip_mismatch Logical, whether to skip loading of layers where there is a mismatch in the
number of weights, or a mismatch in the shape of the weight (only valid when
by_name = FALSE).
reshape Reshape weights to fit the layer when the correct number of values are present
but the shape does not match.
Details
The weight file has:
• layer_names (attribute), a list of strings (ordered names of model layers).
• For every layer, a group named layer.name
• For every such layer group, a group attribute weight_names, a list of strings (ordered names
of weights tensor of the layer).
• For every weight in the layer, a dataset storing the weight value, named after the weight tensor.
For load_model_weights(), if by_name is FALSE (default) weights are loaded based on the net-
work’s topology, meaning the architecture should be the same as when the weights were saved.
Note that layers that don’t have weights are not taken into account in the topological ordering, so
adding or removing layers is fine as long as they don’t have weights.
If by_name is TRUE, weights are loaded into layers only if they share the same name. This is useful
for fine-tuning or transfer-learning models where some of the layers have changed.
save_text_tokenizer 295
See Also
Other model persistence: get_weights, model_to_json, model_to_yaml, save_model_hdf5,
serialize_model
save_text_tokenizer Save a text tokenizer to an external file
Description
Enables persistence of text tokenizers alongside saved models.
Usage
save_text_tokenizer(object, filename)
load_text_tokenizer(filename)
Arguments
object Text tokenizer fit with fit_text_tokenizer()
filename File to save/load
Details
You should always use the same text tokenizer for training and prediction. In many cases however
prediction will occur in another session with a version of the model loaded via load_model_hdf5().
In this case you need to save the text tokenizer object after training and then reload it prior to
prediction.
See Also
Other text tokenization: fit_text_tokenizer, sequences_to_matrix, text_tokenizer, texts_to_matrix,
Examples
## Not run:
# vectorize texts then save for use in prediction

tokenizer <- text_tokenizer(num_words = 10000) %>%
fit_text_tokenizer(tokenizer, texts)
save_text_tokenizer(tokenizer, "tokenizer")
# (train model, etc.)
# ...later in another session

tokenizer <- load_text_tokenizer("tokenizer")
296 serialize_model
# (use tokenizer to preprocess data for prediction)
## End(Not run)
sequences_to_matrix Convert a list of sequences into a matrix.
Description
Convert a list of sequences into a matrix.
Usage
sequences_to_matrix(tokenizer, sequences, mode = c("binary", "count",
"tfidf", "freq"))
Arguments
tokenizer Tokenizer
sequences List of sequences (a sequence is a list of integer word indices).
mode one of "binary", "count", "tfidf", "freq".
Value
A matrix
See Also
Other text tokenization: fit_text_tokenizer, save_text_tokenizer, text_tokenizer, texts_to_matrix,
serialize_model Serialize a model to an R object
Description
Model objects are external references to Keras objects which cannot be saved and restored across
R sessions. The serialize_model() and unserialize_model() functions provide facilities to
convert Keras models to R objects for persistence within R data files.
Usage
serialize_model(model, include_optimizer = TRUE)
unserialize_model(model, custom_objects = NULL, compile = TRUE)

skipgrams 297
Arguments
model Keras model or R "raw" object containing serialized Keras model.
include_optimizer
If TRUE, save optimizer’s state.
custom_objects Mapping class names (or function names) of custom (non-Keras) objects to
class/functions (for example, custom metrics or custom loss functions).
compile Whether to compile the model after loading.
Value
serialize_model() returns an R "raw" object containing an hdf5 version of the Keras model.
unserialize_model() returns a Keras model.
Note
The save_model_hdf5() function enables saving Keras models to external hdf5 files.
See Also
Other model persistence: get_weights, model_to_json, model_to_yaml, save_model_hdf5,
save_model_weights_hdf5
skipgrams Generates skipgram word pairs.
Description
Generates skipgram word pairs.
Usage
skipgrams(sequence, vocabulary_size, window_size = 4,
negative_samples = 1, shuffle = TRUE, categorical = FALSE,
sampling_table = NULL, seed = NULL)
Arguments
sequence A word sequence (sentence), encoded as a list of word indices (integers). If
using a sampling_table, word indices are expected to match the rank of the
words in a reference dataset (e.g. 10 would encode the 10-th most frequently
occuring token). Note that index 0 is expected to be a non-word and will be
skipped.
vocabulary_size
Int, maximum possible word index + 1
window_size Int, size of sampling windows (technically half-window). The window of a word
w_i will be [i-window_size, i+window_size+1]
298 summary.keras.engine.training.Model
negative_samples
float >= 0. 0 for no negative (i.e. random) samples. 1 for same number as
positive samples.
shuffle whether to shuffle the word couples before returning them.
categorical bool. if FALSE, labels will be integers (eg. [0, 1, 1 .. ]), if TRUE labels will
be categorical eg. [[1,0],[0,1],[0,1] .. ]
sampling_table 1D array of size vocabulary_size where the entry i encodes the probabibily to
sample a word of rank i.
seed Random seed
Details
This function transforms a list of word indexes (lists of integers) into lists of words of the form:
• (word, word in the same window), with label 1 (positive samples).
• (word, random word from the vocabulary), with label 0 (negative samples).
Read more about Skipgram in this gnomic paper by Mikolov et al.: Efficient Estimation of Word
Representations in Vector Space
Value
List of couples, labels where:
• couples is a list of 2-element integer vectors: [word_index, other_word_index].
• labels is an integer vector of 0 and 1, where 1 indicates that other_word_index was found
in the same window as word_index, and 0 indicates that other_word_index was random.
• if categorical is set to TRUE, the labels are categorical, ie. 1 becomes [0,1], and 0 becomes
[1, 0].
See Also
Other text preprocessing: make_sampling_table, pad_sequences, text_hashing_trick, text_one_hot,
summary.keras.engine.training.Model
Print a summary of a Keras model
Description
Print a summary of a Keras model
Usage
summary(object,
line_length = getOption("width"), positions = NULL, ...)
texts_to_matrix 299
Arguments
object Keras model instance
line_length Total length of printed lines
positions Relative or absolute positions of log elements in each line. If not provided,
defaults to c(0.33, 0.55, 0.67, 1.0).
... Unused
See Also
predict_generator, predict_on_batch, predict_proba, train_on_batch
texts_to_matrix Convert a list of texts to a matrix.
Description
Convert a list of texts to a matrix.
Usage
texts_to_matrix(tokenizer, texts, mode = c("binary", "count", "tfidf",
"freq"))
Arguments
tokenizer Tokenizer
texts Vector/list of texts (strings).
mode one of "binary", "count", "tfidf", "freq".
Value
A matrix
See Also
Other text tokenization: fit_text_tokenizer, save_text_tokenizer, sequences_to_matrix,
text_tokenizer, texts_to_sequences_generator, texts_to_sequences
300 texts_to_sequences_generator
texts_to_sequences Transform each text in texts in a sequence of integers.
Description
Only top "num_words" most frequent words will be taken into account. Only words known by the
tokenizer will be taken into account.
Usage
texts_to_sequences(tokenizer, texts)
Arguments
tokenizer Tokenizer
See Also
text_tokenizer, texts_to_matrix, texts_to_sequences_generator
texts_to_sequences_generator
Transforms each text in texts in a sequence of integers.
Description
Only top "num_words" most frequent words will be taken into account. Only words known by the
tokenizer will be taken into account.
Usage
texts_to_sequences_generator(tokenizer, texts)
Arguments
tokenizer Tokenizer
Value
Generator which yields individual sequences
See Also
text_tokenizer, texts_to_matrix, texts_to_sequences
text_hashing_trick 301
text_hashing_trick Converts a text to a sequence of indexes in a fixed-size hashing space.
Description
Converts a text to a sequence of indexes in a fixed-size hashing space.
Usage
text_hashing_trick(text, n, hash_function = NULL,

filters = "!\"#$%&()*+,-./:;<=>?@[\\]^_`{|}~\t\n",
lower = TRUE, split = " ")
Arguments
text Input text (string).

n Dimension of the hashing space.
hash_function if NULL uses python hash function, can be ’md5’ or any function that takes in
input a string and returns a int. Note that hash is not a stable hashing function, so
it is not consistent across different runs, while ’md5’ is a stable hashing function.
filters Sequence of characters to filter out such as punctuation. Default includes basic
punctuation, tabs, and newlines.
lower Whether to convert the input to lowercase.
split Sentence split marker (string).
Details
Two or more words may be assigned to the same index, due to possible collisions by the hashing
function.
Value
A list of integer word indices (unicity non-guaranteed).
See Also
Other text preprocessing: make_sampling_table, pad_sequences, skipgrams, text_one_hot,

302 text_tokenizer
text_one_hot One-hot encode a text into a list of word indexes in a vocabulary of

size n.
Description
One-hot encode a text into a list of word indexes in a vocabulary of size n.
Usage
text_one_hot(text, n,
filters = "!\"#$%&()*+,-./:;<=>?@[\\]^_`{|}~\t\n",
Arguments
n Size of vocabulary (integer)
Value
List of integers in [1, n]. Each integer encodes a word (unicity non-guaranteed).
See Also
Other text preprocessing: make_sampling_table, pad_sequences, skipgrams, text_hashing_trick,
text_tokenizer Text tokenization utility
Description
Vectorize a text corpus, by turning each text into either a sequence of integers (each integer being
the index of a token in a dictionary) or into a vector where the coefficient for each token could be
binary, based on word count, based on tf-idf...
Usage
text_tokenizer(num_words = NULL,
filters = "!\"#$%&()*+,-./:;<=>?@[\\]^_`{|}~\t\n",
lower = TRUE, split = " ", char_level = FALSE, oov_token = NULL)
text_to_word_sequence 303
Arguments
num_words the maximum number of words to keep, based on word frequency. Only the
most common num_words words will be kept.
filters a string where each element is a character that will be filtered from the texts.
The default is all punctuation, plus tabs and line breaks, minus the ’ character.
lower boolean. Whether to convert the texts to lowercase.
split character or string to use for token splitting.
char_level if TRUE, every character will be treated as a token
oov_token NULL or string If given, it will be added to ‘word_index“ and used to replace
out-of-vocabulary words during text_to_sequence calls.
Details
By default, all punctuation is removed, turning the texts into space-separated sequences of words
(words maybe include the ’ character). These sequences are then split into lists of tokens. They will
then be indexed or vectorized. 0 is a reserved index that won’t be assigned to any word.
Attributes
The tokenizer object has the following attributes:
• word_counts — named list mapping words to the number of times they appeared on during
fit. Only set after fit_text_tokenizer() is called on the tokenizer.
• word_docs — named list mapping words to the number of documents/texts they appeared on
during fit. Only set after fit_text_tokenizer() is called on the tokenizer.
• word_index — named list mapping words to their rank/index (int). Only set after fit_text_tokenizer()
is called on the tokenizer.
• document_count — int. Number of documents (texts/sequences) the tokenizer was trained
on. Only set after fit_text_tokenizer() is called on the tokenizer.
See Also
texts_to_matrix, texts_to_sequences_generator, texts_to_sequences
text_to_word_sequence Convert text to a sequence of words (or tokens).
Description
Convert text to a sequence of words (or tokens).
304 timeseries_generator
Usage
text_to_word_sequence(text,
filters = "!\"#$%&()*+,-./:;<=>?@[\\]^_`{|}~\t\n",
Arguments
Value
Words (or tokens)
See Also
Other text preprocessing: make_sampling_table, pad_sequences, skipgrams, text_hashing_trick,
text_one_hot
timeseries_generator Utility function for generating batches of temporal data.
Description
Utility function for generating batches of temporal data.
Usage
timeseries_generator(data, targets, length, sampling_rate = 1,
stride = 1, start_index = 0, end_index = NULL, shuffle = FALSE,
reverse = FALSE, batch_size = 128)
Arguments
data Object containing consecutive data points (timesteps). The data should be 2D,
and axis 1 is expected to be the time dimension.
targets Targets corresponding to timesteps in data. It should have same length as data.
length Length of the output sequences (in number of timesteps).
sampling_rate Period between successive individual timesteps within sequences. For rate r,
timesteps data[i], data[i-r], ... data[i - length] are used for create a
sample sequence.
time_distributed 305
stride Period between successive output sequences. For stride s, consecutive output
samples would be centered around data[i], data[i+s], data[i+2*s], etc.
start_index, end_index
Data points earlier than start_index or later than end_index will not be used
in the output sequences. This is useful to reserve part of the data for test or
validation.
shuffle Whether to shuffle output samples, or instead draw them in chronological order.
reverse Boolean: if true, timesteps in each output sample will be in reverse chronolog-
ical order.
batch_size Number of timeseries samples in each batch (except maybe the last one).
Value
An object that can be passed to generator based training functions (e.g. fit_generator()).ma
time_distributed Apply a layer to every temporal slice of an input.
Description
The input should be at least 3D, and the dimension of index one will be considered to be the temporal
dimension.
Usage
time_distributed(object, layer, input_shape = NULL,
Arguments
layer A layer instance.
batch_input_shape
306 to_categorical
Details
Consider a batch of 32 samples, where each sample is a sequence of 10 vectors of 16 dimensions.

The batch input shape of the layer is then (32, 10, 16), and the input_shape, not including the
samples dimension, is (10, 16). You can then use time_distributed to apply a layer_dense to
each of the 10 timesteps, independently.
See Also
Other layer wrappers: bidirectional
to_categorical Converts a class vector (integers) to binary class matrix.
Description
Converts a class vector (integers) to binary class matrix.
Usage
to_categorical(y, num_classes = NULL, dtype = "float32")
Arguments
y Class vector to be converted into a matrix (integers from 0 to num_classes).

num_classes Total number of classes.
dtype The data type expected by the input, as a string
Details
E.g. for use with loss_categorical_crossentropy().
Value
A binary matrix representation of the input.

train_on_batch 307
train_on_batch Single gradient update or model evaluation over one batch of samples.
Description
Single gradient update or model evaluation over one batch of samples.
Usage
train_on_batch(object, x, y, class_weight = NULL, sample_weight = NULL)
test_on_batch(object, x, y, sample_weight = NULL)
Arguments
x input data, as an array or list of arrays (if the model has multiple inputs).
y labels, as an array.
class_weight named list mapping classes to a weight value, used for scaling the loss function
(during training only).
sample_weight sample weights, as an array.
Value
Scalar training or test loss (if the model has no metrics) or list of scalars (if the model computes
other metrics). The property model$metrics_names will give you the display labels for the scalar
outputs.
See Also
predict_generator, predict_on_batch, predict_proba, summary.keras.engine.training.Model
use_implementation Select a Keras implementation and backend
Description
Select a Keras implementation and backend
308 with_custom_object_scope
Usage
use_implementation(implementation = c("keras", "tensorflow"))
use_backend(backend = c("tensorflow", "cntk", "theano", "plaidml"))
Arguments
implementation One of "keras" or "tensorflow" (defaults to "keras").
backend One of "tensorflow", "cntk", or "theano" (defaults to "tensorflow")
Details
Keras has multiple implementations (the original keras implementation and the implementation na-
tive to TensorFlow) and supports multiple backends ("tensorflow", "cntk", "theano", and "plaidml").
These functions allow switching between the various implementations and backends.
The functions should be called after library(keras) and before calling other functions within the
package (see below for an example).
The default implementation and backend should be suitable for most use cases. The "tensorflow"
implementation is useful when using Keras in conjunction with TensorFlow Estimators (the tfesti-
mators R package).
Examples
## Not run:
# use the tensorflow implementation
library(keras)
use_implementation("tensorflow")
# use the cntk backend

library(keras)
use_backend("theano")
## End(Not run)
with_custom_object_scope
Provide a scope with mappings of names to custom objects
Description
Provide a scope with mappings of names to custom objects
Usage
with_custom_object_scope(objects, expr)
with_custom_object_scope 309
Arguments
objects Named list of objects
expr Expression to evaluate
Details
There are many elements of Keras models that can be customized with user objects (e.g. losses,
metrics, regularizers, etc.). When loading saved models that use these functions you typically need
to explicitily map names to user objects via the custom_objects parmaeter.
The with_custom_object_scope() function provides an alternative that lets you create a named
alias for a user object that applies to an entire block of code, and is automatically recognized when
loading saved models.
Examples
## Not run:
# define custom metric
metric_top_3_categorical_accuracy <-
custom_metric("top_3_categorical_accuracy", function(y_true, y_pred) {
})
with_custom_object_scope(c(top_k_acc = sparse_top_k_cat_acc), {
# ...define model...
# compile model (refer to "top_k_acc" by name)

model %>% compile(
optimizer = optimizer_nadam(),
metrics = c("top_k_acc")
)
# save the model

save_model_hdf5("my_model.h5")
# loading the model within the custom object scope doesn't

# require explicitly providing the custom_object
load_model_hdf5("my_model.h5")
})
## End(Not run)
Index
∗Topic datasets application_vgg19 (application_vgg), 21

KerasCallback, 78 application_xception, 22
KerasConstraint, 80
KerasLayer, 81 backend, 23
KerasWrapper, 82 backend(), 35
bidirectional, 24, 306
activation_elu (activation_relu), 10
activation_exponential callback_csv_logger, 25, 26–32
(activation_relu), 10 callback_early_stopping, 25, 25, 27–32
activation_hard_sigmoid callback_lambda, 25, 26, 26, 27–32
(activation_relu), 10 callback_learning_rate_scheduler,
activation_linear (activation_relu), 10 25–27, 27, 28–32
activation_relu, 10 callback_model_checkpoint, 25–27, 28,
activation_selu (activation_relu), 10 29–32
activation_sigmoid (activation_relu), 10 callback_progbar_logger, 25–28, 29,
activation_softmax (activation_relu), 10 30–32
activation_softplus (activation_relu), callback_reduce_lr_on_plateau, 25–29,
10 29, 31, 32
activation_softsign (activation_relu), callback_remote_monitor, 25–30, 30, 32
10 callback_tensorboard, 25–31, 31, 32
activation_tanh (activation_relu), 10 callback_terminate_on_naan, 25–32, 32
application_densenet, 11 clone_model, 33
application_densenet121 compile(), 43, 46–48
(application_densenet), 11 compile.keras.engine.training.Model,
application_densenet169 33, 44, 47, 49, 58, 60, 83, 85, 279,
(application_densenet), 11 286, 288–291, 299, 307
application_densenet201 compile.keras.engine.training.Model(),
(application_densenet), 11 273, 276
application_inception_resnet_v2, 12 constraint_maxnorm (constraints), 34
application_inception_v3, 13 constraint_minmaxnorm (constraints), 34
application_mobilenet, 15 constraint_nonneg (constraints), 34
application_mobilenet_v2, 16 constraint_unitnorm (constraints), 34
application_nasnet, 18 constraints, 34, 80
application_nasnetlarge count_params, 36, 58, 60, 61, 292
(application_nasnet), 18 create_layer, 36
application_nasnetmobile create_wrapper, 37
(application_nasnet), 18 custom_metric (metric_binary_accuracy),
application_resnet50, 19 274
application_vgg, 21
application_vgg16 (application_vgg), 21 dataset_boston_housing, 37, 38–43
310
INDEX 311
dataset_cifar10, 38, 38, 39–43 get_input_mask_at (get_input_at), 59

dataset_cifar100, 38, 39, 40–43 get_input_shape_at (get_input_at), 59
dataset_fashion_mnist, 38, 39, 39, 41–43 get_layer, 34, 44, 47, 49, 58, 60, 83, 85, 279,
dataset_imdb, 38–40, 40, 42, 43 288–291, 299, 307
dataset_imdb(), 42, 43 get_output_at (get_input_at), 59
dataset_imdb_word_index (dataset_imdb), get_output_mask_at (get_input_at), 59
40 get_output_shape_at (get_input_at), 59
dataset_mnist, 38–41, 41, 43 get_weights, 36, 58, 60, 61, 277, 278, 292,
dataset_reuters, 38–42, 42 294, 295, 297
dataset_reuters_word_index
(dataset_reuters), 42 hdf5_matrix, 61
densenet_preprocess_input
(application_densenet), 11 image_array_resize (image_to_array), 65
image_array_save (image_to_array), 65
evaluate.keras.engine.training.Model, image_data_generator, 63
34, 43, 44, 47, 49, 58, 60, 83, 85, image_data_generator(), 49, 51, 53, 55, 57
279, 288–291, 299, 307 image_load, 50, 52, 54, 55, 65, 66
evaluate_generator, 34, 44, 44, 47, 49, 58, image_to_array, 50, 52, 54, 55, 65, 65
60, 83, 85, 279, 288–291, 299, 307 imagenet_decode_predictions, 62
evaluate_generator(), 55 imagenet_preprocess_input, 62
export_savedmodel.keras.engine.training.Model,implementation, 66
45 inception_resnet_v2_preprocess_input
(application_inception_resnet_v2),
fit.keras.engine.training.Model, 34, 44, 12
45, 49, 58, 60, 83, 85, 279, 288–291, inception_v3_preprocess_input
299, 307 (application_inception_v3), 13
fit_generator, 34, 44, 47, 47, 58, 60, 83, 85, initializer_constant, 67, 68–75
279, 288–291, 299, 307 initializer_glorot_normal, 67, 67, 68–75
fit_generator(), 305 initializer_glorot_uniform, 67, 68, 68,
fit_image_data_generator, 49, 52, 54, 55, 69–75
65, 66 initializer_he_normal, 67, 68, 69, 70–75
fit_text_tokenizer, 50, 295, 296, 299, 300, initializer_he_uniform, 67–69, 69, 70–75
303 initializer_identity, 67–70, 70, 71–75
fit_text_tokenizer(), 295 initializer_lecun_normal, 67–70, 70,
flow_images_from_data, 50, 51, 54, 55, 65, 71–75, 186
66 initializer_lecun_uniform, 67–71, 71,
flow_images_from_dataframe, 50, 52, 52, 72–75
55, 65, 66 initializer_ones, 67–72, 72, 73–75
flow_images_from_directory, 50, 52, 54, initializer_orthogonal, 67–72, 72, 73–75
54, 65, 66 initializer_random_normal, 67–72, 73, 74,
flow_images_from_directory(), 48 75
freeze_weights, 56 initializer_random_normal(), 74
from_config (get_config), 57 initializer_random_uniform, 67–73, 73,
74, 75
generator_next, 57 initializer_truncated_normal, 67–74, 74,
get_config, 34, 36, 44, 47, 49, 57, 60, 61, 83, 75
85, 279, 288–292, 299, 307 initializer_variance_scaling, 67–74, 74,
get_file, 58 75
get_input_at, 36, 58, 59, 61, 292 initializer_zeros, 67–75, 75
312 INDEX
install_keras, 76 k_flatten, 115

is_keras_available, 78 k_floatx, 115
k_foldl, 116
k_abs, 85 k_foldr, 117
k_all, 86 k_function, 117
k_any, 87 k_gather, 118
k_arange, 87 k_get_session, 119
k_argmax, 88 k_get_uid, 119
k_argmin, 89 k_get_value, 120
k_backend, 89 k_get_variable_shape, 121
k_batch_dot, 90 k_gradients, 121
k_batch_flatten, 91 k_greater, 122
k_batch_get_value, 91 k_greater_equal, 122
k_batch_get_value(), 93 k_greater_equal(), 35
k_batch_normalization, 92 k_hard_sigmoid, 123
k_batch_set_value, 93
k_identity, 124
k_batch_set_value(), 92
k_image_data_format, 124
k_bias_add, 93
k_in_test_phase, 125
k_binary_crossentropy, 94
k_in_top_k, 126
k_cast, 95
k_in_train_phase, 127
k_cast_to_floatx, 95
k_int_shape, 125
k_categorical_crossentropy, 96
k_is_keras_tensor, 127
k_clear_session, 97
k_is_placeholder, 128
k_clip, 97
k_is_sparse, 128
k_concatenate, 98
k_constant, 98 k_is_tensor, 129
k_conv1d, 99 k_l2_normalize, 130
k_conv2d, 100 k_learning_phase, 130
k_conv2d_transpose, 100 k_less, 131
k_conv3d, 101 k_less_equal, 131
k_conv3d_transpose, 102 k_local_conv1d, 132
k_cos, 103 k_local_conv2d, 133
k_count_params, 103 k_log, 134
k_ctc_batch_cost, 104 k_logsumexp, 134
k_ctc_decode, 105 k_manual_variable_initialization, 135
k_ctc_label_dense_to_sparse, 106 k_map_fn, 136
k_cumprod, 106 k_max, 136
k_cumsum, 107 k_maximum, 137
k_depthwise_conv2d, 108 k_mean, 138
k_dot, 108 k_min, 138
k_dropout, 109 k_minimum, 139
k_dtype, 110 k_moving_average_update, 140
k_elu, 110 k_ndim, 140
k_epsilon, 111 k_normalize_batch_in_training, 141
k_equal, 111 k_not_equal, 142
k_eval, 112 k_one_hot, 144
k_exp, 113 k_ones, 142
k_expand_dims, 113 k_ones_like, 143
k_eye, 114 k_permute_dimensions, 144
INDEX 313
k_placeholder, 145 k_temporal_padding, 173

k_pool2d, 146 k_tile, 174
k_pool3d, 146 k_to_dense, 175
k_pow, 147 k_transpose, 175
k_print_tensor, 148 k_truncated_normal, 176
k_prod, 148 k_update, 177
k_random_binomial, 149 k_update_add, 177
k_random_normal, 150 k_update_sub, 178
k_random_normal_variable, 150 k_var, 178
k_random_uniform, 151 k_variable, 179
k_random_uniform_variable, 152 k_zeros, 180
k_relu, 153 k_zeros_like, 180
k_repeat, 153 keras (keras-package), 9
k_repeat_elements, 154 keras-package, 9
k_reset_uids, 155 keras_array, 82
k_reshape, 155 keras_model, 34, 44, 47, 49, 58, 60, 83, 85,
k_resize_images, 156 279, 288–291, 299, 307
k_resize_volumes, 156 keras_model_custom, 84
k_reverse, 157 keras_model_sequential, 34, 44, 47, 49, 58,
k_rnn, 158 60, 83, 84, 279, 288–291, 299, 307
k_round, 159 keras_model_sequential(), 36, 37
k_separable_conv2d, 159 KerasCallback, 78, 79
k_set_epsilon (k_epsilon), 111 KerasConstraint, 35, 80
KerasLayer, 81, 81
k_set_floatx (k_floatx), 115
KerasWrapper, 82, 82
k_set_image_data_format
(k_image_data_format), 124 layer_activation, 181, 182, 183, 185–189,
k_set_learning_phase, 160 220, 223, 225, 237, 239, 246,
k_set_session (k_get_session), 119 252–254
k_set_value, 161 layer_activation(), 10
k_shape, 161 layer_activation_elu, 182, 182, 183,
k_sigmoid, 162 185–188
k_sign, 162 layer_activation_leaky_relu, 182, 183,
k_sin, 163 185–188
k_softmax, 164 layer_activation_parametric_relu, 182,
k_softplus, 164 183, 184, 185–188
k_softsign, 165 layer_activation_relu, 182, 183, 185, 185,
k_sparse_categorical_crossentropy, 165 186–188
k_spatial_2d_padding, 166 layer_activation_selu, 182, 183, 185, 186,
k_spatial_3d_padding, 167 187, 188
k_sqrt, 167 layer_activation_softmax, 182, 183, 185,
k_square, 168 186, 187, 188
k_squeeze, 169 layer_activation_thresholded_relu, 182,
k_stack, 169 183, 185–187, 188
k_std, 170 layer_activity_regularization, 182, 189,
k_stop_gradient, 171 220, 223, 225, 237, 239, 246,
k_sum, 171 252–254
k_switch, 172 layer_add, 190, 192, 198, 222, 247, 250, 251,
k_tanh, 173 265
314 INDEX
layer_alpha_dropout, 186, 190, 227, 228 layer_dot, 190, 192, 198, 222, 247, 250, 251,
layer_average, 190, 192, 198, 222, 247, 250, 265
251, 265 layer_dropout, 182, 189, 220, 223, 225, 237,
layer_average_pooling_1d, 192, 194, 196, 239, 246, 252–254, 262–264
228–233, 248–250 layer_embedding, 224
layer_average_pooling_2d, 193, 193, 196, layer_flatten, 182, 189, 220, 223, 225, 237,
228–233, 248–250 239, 246, 252–254
layer_average_pooling_3d, 193, 194, 195, layer_gaussian_dropout, 191, 226, 228
228–233, 248–250 layer_gaussian_noise, 191, 227, 227
layer_batch_normalization, 196 layer_global_average_pooling_1d, 193,
layer_concatenate, 190, 192, 198, 222, 247, 194, 196, 228, 229–233, 248–250
250, 251, 265 layer_global_average_pooling_2d, 193,
layer_conv_1d, 198, 202, 205, 207, 209, 212, 194, 196, 228, 229, 230–233,
214, 215, 221, 256, 259, 266–270, 248–250
272 layer_global_average_pooling_3d, 193,
layer_conv_1d(), 239 194, 196, 228, 229, 230, 231–233,
layer_conv_2d, 200, 200, 205, 207, 209, 212, 248–250
214, 215, 221, 256, 259, 266–270, layer_global_max_pooling_1d, 193, 194,
272 196, 228–230, 231, 232, 233,
layer_conv_2d(), 240 248–250
layer_conv_2d_transpose, 200, 202, 203, layer_global_max_pooling_2d, 193, 194,
207, 209, 212, 214, 215, 221, 256, 196, 228–231, 232, 233, 248–250
259, 266–270, 272 layer_global_max_pooling_3d, 193, 194,
layer_conv_3d, 200, 202, 205, 205, 209, 212, 196, 228–232, 233, 248–250
214, 215, 221, 256, 259, 266–270, layer_gru, 216, 218, 234, 245, 262
272 layer_input, 182, 189, 220, 223, 225, 237,
layer_conv_3d_transpose, 200, 202, 205, 239, 246, 252–254
207, 207, 212, 214, 215, 221, 256, layer_lambda, 182, 189, 220, 223, 225, 237,
259, 266–270, 272 238, 246, 252–254
layer_conv_lstm_2d, 200, 202, 205, 207, layer_locally_connected_1d, 239, 242
209, 209, 212, 214, 215, 221, 256, layer_locally_connected_2d, 240, 240
259, 266–270, 272 layer_lstm, 216, 218, 237, 242, 262
layer_cropping_1d, 200, 202, 205, 207, 209, layer_masking, 182, 189, 220, 223, 225, 237,
212, 212, 214, 215, 221, 256, 259, 239, 245, 252–254
266–270, 272 layer_max_pooling_1d, 193, 194, 196,
layer_cropping_2d, 200, 202, 205, 207, 209, 228–233, 247, 249, 250
212, 213, 215, 221, 256, 259, layer_max_pooling_2d, 193, 194, 196,
266–270, 272 228–233, 248, 248, 250
layer_cropping_3d, 200, 202, 205, 207, 209, layer_max_pooling_3d, 193, 194, 196,
212, 214, 214, 221, 256, 259, 228–233, 248, 249, 249
266–270, 272 layer_maximum, 190, 192, 198, 222, 246, 250,
layer_cudnn_gru, 215, 218, 237, 245, 262 251, 265
layer_cudnn_lstm, 216, 217, 237, 245, 262 layer_minimum, 190, 192, 198, 222, 247, 250,
layer_dense, 182, 189, 218, 223, 225, 237, 251, 265
239, 246, 252–254 layer_multiply, 190, 192, 198, 222, 247,
layer_depthwise_conv_2d, 200, 202, 205, 250, 251, 265
207, 209, 212, 214, 215, 220, 256, layer_permute, 182, 189, 220, 223, 225, 237,
259, 266–270, 272 239, 246, 251, 253, 254
INDEX 315
layer_repeat_vector, 182, 189, 220, 223, loss_categorical_hinge

225, 237, 239, 246, 252, 252, 254 (loss_mean_squared_error), 272
layer_reshape, 182, 189, 220, 223, 225, 237, loss_cosine_proximity
239, 246, 252, 253, 253 (loss_mean_squared_error), 272
layer_separable_conv_1d, 200, 202, 205, loss_hinge (loss_mean_squared_error),
207, 209, 212, 214, 215, 222, 254, 272
259, 266–270, 272 loss_kullback_leibler_divergence
layer_separable_conv_2d, 200, 202, 205, (loss_mean_squared_error), 272
207, 209, 212, 214, 215, 222, 256, loss_logcosh (loss_mean_squared_error),
256, 266–270, 272 272
layer_simple_rnn, 216, 218, 237, 245, 259 loss_mean_absolute_error
layer_spatial_dropout_1d, 223, 262, 263, (loss_mean_squared_error), 272
264 loss_mean_absolute_percentage_error
layer_spatial_dropout_2d, 223, 262, 263, (loss_mean_squared_error), 272
264 loss_mean_squared_error, 272
layer_spatial_dropout_3d, 223, 262, 263, loss_mean_squared_logarithmic_error
264 (loss_mean_squared_error), 272
layer_subtract, 190, 192, 198, 222, 247, loss_poisson (loss_mean_squared_error),
250, 251, 265 272
layer_upsampling_1d, 200, 202, 205, 207, loss_sparse_categorical_crossentropy
209, 212, 214, 215, 222, 256, 259, (loss_mean_squared_error), 272
265, 267–270, 272 loss_squared_hinge
layer_upsampling_2d, 200, 202, 205, 207, (loss_mean_squared_error), 272
209, 212, 214, 215, 222, 256, 259,
make_sampling_table, 273, 287, 298, 301,
266, 266, 268–270, 272
302, 304
layer_upsampling_3d, 200, 202, 205, 207, metric_binary_accuracy, 274
209, 212, 214, 215, 222, 256, 259, metric_binary_crossentropy
266, 267, 267, 269, 270, 272 (metric_binary_accuracy), 274
layer_zero_padding_1d, 200, 202, 205, 207, metric_categorical_accuracy
209, 212, 214, 215, 222, 256, 259, (metric_binary_accuracy), 274
266–268, 268, 270, 272 metric_categorical_crossentropy
layer_zero_padding_2d, 200, 202, 205, 207, (metric_binary_accuracy), 274
209, 212, 214, 215, 222, 256, 259, metric_cosine_proximity
266–269, 269, 272 (metric_binary_accuracy), 274
layer_zero_padding_3d, 200, 202, 205, 207, metric_hinge (metric_binary_accuracy),
209, 212, 214, 215, 222, 256, 259, 274
266–270, 271 metric_kullback_leibler_divergence
load_model_hdf5 (save_model_hdf5), 293 (metric_binary_accuracy), 274
load_model_hdf5(), 276, 295 metric_mean_absolute_error
load_model_weights_hdf5 (metric_binary_accuracy), 274
(save_model_weights_hdf5), 294 metric_mean_absolute_percentage_error
load_text_tokenizer (metric_binary_accuracy), 274
(save_text_tokenizer), 295 metric_mean_squared_error
loss_binary_crossentropy (metric_binary_accuracy), 274
(loss_mean_squared_error), 272 metric_mean_squared_logarithmic_error
loss_categorical_crossentropy (metric_binary_accuracy), 274
(loss_mean_squared_error), 272 metric_poisson
loss_categorical_crossentropy(), 306 (metric_binary_accuracy), 274
316 INDEX
metric_sparse_categorical_crossentropy predict_generator, 34, 44, 47, 49, 58, 60,

(metric_binary_accuracy), 274 83, 85, 279, 288, 289, 289, 290, 291,
metric_sparse_top_k_categorical_accuracy 299, 307
(metric_binary_accuracy), 274 predict_generator(), 55
metric_squared_hinge predict_on_batch, 34, 44, 47, 49, 58, 60, 83,
(metric_binary_accuracy), 274 85, 279, 288–290, 290, 291, 299, 307
metric_top_k_categorical_accuracy predict_proba, 34, 44, 47, 49, 58, 60, 83, 85,
(metric_binary_accuracy), 274 279, 288–290, 291, 299, 307
mobilenet_decode_predictions py_to_r(), 23
(application_mobilenet), 15
mobilenet_load_model_hdf5 R6Class, 79–82
(application_mobilenet), 15 regularizer_l1, 292
mobilenet_preprocess_input regularizer_l1_l2 (regularizer_l1), 292
(application_mobilenet), 15 regularizer_l2 (regularizer_l1), 292
mobilenet_v2_decode_predictions reset_states, 36, 58, 60, 61, 292
(application_mobilenet_v2), 16 reticulate::py_install(), 77
mobilenet_v2_load_model_hdf5
(application_mobilenet_v2), 16 save_model_hdf5, 61, 277, 278, 293, 295, 297
mobilenet_v2_preprocess_input save_model_hdf5(), 35, 80, 279, 297
(application_mobilenet_v2), 16 save_model_weights_hdf5, 61, 277, 278,
model_from_json (model_to_json), 277 294, 294, 297
model_from_yaml (model_to_yaml), 277 save_model_weights_hdf5(), 35, 80, 279
model_to_json, 61, 277, 278, 294, 295, 297 save_text_tokenizer, 50, 295, 296, 299,
model_to_yaml, 61, 277, 277, 294, 295, 297 300, 303
multi_gpu_model, 34, 44, 47, 49, 58, 60, 83, sequences_to_matrix, 50, 295, 296, 299,
85, 278, 288–291, 299, 307 300, 303
sequences_to_matrix(), 50
serialize_model, 61, 277, 278, 294, 295, 296
nasnet_preprocess_input
serialize_model(), 293
(application_nasnet), 18
set_weights (get_weights), 61
normalize, 280
skipgrams, 274, 287, 297, 301, 302, 304
skipgrams(), 274
optimizer_adadelta, 281, 282–286 summary.keras.engine.training.Model,
optimizer_adagrad, 281, 281, 283–286 34, 44, 47, 49, 58, 60, 83, 85, 279,
optimizer_adam, 281, 282, 282, 283–286 288–291, 298, 307
optimizer_adamax, 281–283, 283, 284–286
optimizer_nadam, 281–283, 284, 285, 286 test_on_batch (train_on_batch), 307
optimizer_rmsprop, 281–284, 285, 286 text_hashing_trick, 274, 287, 298, 301,
optimizer_sgd, 281–285, 285 302, 304
text_one_hot, 274, 287, 298, 301, 302, 304
pad_sequences, 274, 286, 298, 301, 302, 304 text_to_word_sequence, 274, 287, 298, 301,
plot(), 288 302, 303
plot.keras_training_history, 287 text_tokenizer, 50, 295, 296, 299, 300, 302
pop_layer, 34, 44, 47, 49, 58, 60, 83, 85, 279, text_tokenizer(), 50
288, 289–291, 299, 307 texts_to_matrix, 50, 295, 296, 299, 300, 303
predict.keras.engine.training.Model, texts_to_matrix(), 50
34, 44, 47, 49, 58, 60, 83, 85, 279, texts_to_sequences, 50, 295, 296, 299, 300,
288, 288, 290, 291, 299, 307 300, 303
predict_classes (predict_proba), 291 texts_to_sequences(), 50
INDEX 317
texts_to_sequences_generator, 50, 295,

296, 299, 300, 300, 303
time_distributed, 24, 305
timeseries_generator, 304
to_categorical, 306
to_categorical(), 273
train_on_batch, 34, 44, 47, 49, 58, 60, 83,
85, 279, 288–291, 299, 307
unfreeze_weights (freeze_weights), 56
unserialize_model (serialize_model), 296
use_backend (use_implementation), 307
use_implementation, 307
with_custom_object_scope, 308
with_custom_object_scope(), 276, 293
xception_preprocess_input
(application_xception), 22

Package Keras

Uploaded by

Copyright:

Available Formats

Package Keras

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Package Keras

Uploaded by

Copyright:

Available Formats

Package ‘keras’

keras-package R interface to Keras

• JJ Allaire [copyright holder]

• RStudio [contributor, copyright holder, funder]

• Yuan Tang <[email protected]> (0000-0001-5243-233X) [contributor, copyright

activation_relu Activation functions

activation_softmax(x, axis = -1)

application_densenet Instantiates the DenseNet architecture.

application_densenet121(include_top = TRUE, weights = "imagenet",

application_densenet169(include_top = TRUE, weights = "imagenet",

application_densenet201(include_top = TRUE, weights = "imagenet",

densenet_preprocess_input(x, data_format = NULL)

application_inception_v3(include_top = TRUE, weights = "imagenet",

A Keras model instance.

• Rethinking the Inception Architecture for Computer Vision

application_mobilenet MobileNet model architecture.

application_nasnet Instantiates a NASNet model.

application_nasnetlarge(input_shape = NULL, include_top = TRUE,

application_nasnetmobile(input_shape = NULL, include_top = TRUE,

• If filter_multiplier < 1.0, proportionally decreases the number of filters

application_resnet50 ResNet50 model for Keras.

Optionally loads weights pre-trained on ImageNet.

A Keras model instance.

- Deep Residual Learning for Image Recognition

# instantiate the model

# load the image

# ensure we have a 4d tensor with single element in the batch dimension,

# make predictions then decode and print them

application_vgg VGG16 and VGG19 models for Keras.

application_vgg19(include_top = TRUE, weights = "imagenet",

model <- application_vgg16(weights = 'imagenet', include_top = FALSE)

img_path <- "elephant.jpg"

features <- model %>% predict(x)

application_xception Xception V1 model for Keras.

backend Keras backend tensor engine

bidirectional Bidirectional wrapper for RNNs.

callback_csv_logger Callback that streams epoch results to a csv file

Supports all values that can be represented as a string

callback_csv_logger(filename, separator = ",", append = FALSE)

filename filename of the csv file, e.g. ’run/log.csv’.

Other callbacks: callback_early_stopping, callback_lambda, callback_learning_rate_scheduler,

Stop training when a monitored quantity has stopped improving.

callback_early_stopping(monitor = "val_loss", min_delta = 0,

monitor quantity to be monitored.

Other callbacks: callback_csv_logger, callback_lambda, callback_learning_rate_scheduler,

callback_lambda Create a custom callback

• on_epoch_begin and on_epoch_end expect two positional arguments: epoch, logs

callback_lambda(on_epoch_begin = NULL, on_epoch_end = NULL,

on_epoch_begin called at the beginning of every epoch.

Other callbacks: callback_csv_logger, callback_early_stopping, callback_learning_rate_scheduler,

Learning rate scheduler.

Other callbacks: callback_csv_logger, callback_early_stopping, callback_lambda, callback_model_checkpoint,

callback_tensorboard TensorBoard basic visualizations

clone_model Clone a model instance.

constraints Weight constraints

constraint_minmaxnorm(min_value = 0, max_value = 1, rate = 1,

nonneg_constraint <- function(w) {

layer_dense(units = 32, input_shape = c(784),

count_params Count the total number of scalars composing the weights.

create_layer Create a Keras Layer