This operation creates a new tensor by adding sparse updates to the passed
in tensor.
This operation is very similar to tf.compat.v1.scatter_nd_add, except that the
updates are added onto an existing tensor (as opposed to a variable). If the
memory for the existing tensor cannot be re-used, a copy is made and updated.
indices is an integer tensor containing indices into a new tensor of shape
tensor.shape. The last dimension of indices can be at most the rank of
tensor.shape:
indices.shape[-1] <=tensor.shape.rank
The last dimension of indices corresponds to indices into elements
(if indices.shape[-1] = tensor.shape.rank) or slices
(if indices.shape[-1] < tensor.shape.rank) along dimension
indices.shape[-1] of tensor.shape. updates is a tensor with shape
The simplest form of tensor_scatter_nd_add is to add individual elements to a
tensor by index. For example, say we want to add 4 elements in a rank-1
tensor with 8 elements.
In Python, this scatter add operation would look like this:
We can also, insert entire slices of a higher rank tensor all at once. For
example, if we wanted to insert two slices in the first dimension of a
rank-3 tensor with two matrices of new values.
In Python, this scatter add operation would look like this:
[null,null,["Last updated 2024-04-26 UTC."],[],[],null,["# tf.tensor_scatter_nd_add\n\n\u003cbr /\u003e\n\nAdds sparse `updates` to an existing tensor according to `indices`.\n\n#### View aliases\n\n\n**Compat aliases for migration**\n\nSee\n[Migration guide](https://www.tensorflow.org/guide/migrate) for\nmore details.\n\n[`tf.compat.v1.tensor_scatter_add`](https://www.tensorflow.org/api_docs/python/tf/tensor_scatter_nd_add), [`tf.compat.v1.tensor_scatter_nd_add`](https://www.tensorflow.org/api_docs/python/tf/tensor_scatter_nd_add)\n\n\u003cbr /\u003e\n\n tf.tensor_scatter_nd_add(\n tensor: Annotated[Any, TV_TensorScatterAdd_T],\n indices: Annotated[Any, TV_TensorScatterAdd_Tindices],\n updates: Annotated[Any, TV_TensorScatterAdd_T],\n name=None\n ) -\u003e Annotated[Any, TV_TensorScatterAdd_T]\n\n### Used in the notebooks\n\n| Used in the guide |\n|----------------------------------------------------------------------------------------------------------------------------------------------------------|\n| - [Extension types](https://www.tensorflow.org/guide/extension_type) - [Introduction to tensor slicing](https://www.tensorflow.org/guide/tensor_slicing) |\n\nThis operation creates a new tensor by adding sparse `updates` to the passed\nin `tensor`.\nThis operation is very similar to [`tf.compat.v1.scatter_nd_add`](../tf/compat/v1/scatter_nd_add), except that the\nupdates are added onto an existing tensor (as opposed to a variable). If the\nmemory for the existing tensor cannot be re-used, a copy is made and updated.\n\n`indices` is an integer tensor containing indices into a new tensor of shape\n`tensor.shape`. The last dimension of `indices` can be at most the rank of\n`tensor.shape`: \n\n indices.shape[-1] \u003c= tensor.shape.rank\n\nThe last dimension of `indices` corresponds to indices into elements\n(if `indices.shape[-1] = tensor.shape.rank`) or slices\n(if `indices.shape[-1] \u003c tensor.shape.rank`) along dimension\n`indices.shape[-1]` of `tensor.shape`. `updates` is a tensor with shape \n\n indices.shape[:-1] + tensor.shape[indices.shape[-1]:]\n\nThe simplest form of `tensor_scatter_nd_add` is to add individual elements to a\ntensor by index. For example, say we want to add 4 elements in a rank-1\ntensor with 8 elements.\n\nIn Python, this scatter add operation would look like this: \n\n indices = tf.constant([[4], [3], [1], [7]])\n updates = tf.constant([9, 10, 11, 12])\n tensor = tf.ones([8], dtype=tf.int32)\n updated = tf.tensor_scatter_nd_add(tensor, indices, updates)\n updated\n \u003ctf.Tensor: shape=(8,), dtype=int32,\n numpy=array([ 1, 12, 1, 11, 10, 1, 1, 13], dtype=int32)\u003e\n\nWe can also, insert entire slices of a higher rank tensor all at once. For\nexample, if we wanted to insert two slices in the first dimension of a\nrank-3 tensor with two matrices of new values.\n\nIn Python, this scatter add operation would look like this: \n\n indices = tf.constant([[0], [2]])\n updates = tf.constant([[[5, 5, 5, 5], [6, 6, 6, 6],\n [7, 7, 7, 7], [8, 8, 8, 8]],\n [[5, 5, 5, 5], [6, 6, 6, 6],\n [7, 7, 7, 7], [8, 8, 8, 8]]])\n tensor = tf.ones([4, 4, 4],dtype=tf.int32)\n updated = tf.tensor_scatter_nd_add(tensor, indices, updates)\n updated\n \u003ctf.Tensor: shape=(4, 4, 4), dtype=int32,\n numpy=array([[[6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8], [9, 9, 9, 9]],\n [[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]],\n [[6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8], [9, 9, 9, 9]],\n [[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]]], dtype=int32)\u003e\n\n| **Note:** on CPU, if an out of bound index is found, an error is returned. On GPU, if an out of bound index is found, the index is ignored.\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n| Args ---- ||\n|-----------|----------------------------------------------------------------------------------|\n| `tensor` | A `Tensor`. Tensor to copy/update. |\n| `indices` | A `Tensor`. Must be one of the following types: `int32`, `int64`. Index tensor. |\n| `updates` | A `Tensor`. Must have the same type as `tensor`. Updates to scatter into output. |\n| `name` | A name for the operation (optional). |\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\n| Returns ------- ||\n|---|---|\n| A `Tensor`. Has the same type as `tensor`. ||\n\n\u003cbr /\u003e"]]
