Scribd
Upload
36 views

Neural Network With Numpy

This document describes a deep neural network framework implementation with numpy. It provides instructions for installing the framework from GitHub and with pip, running tests, and using the framework to build and train a convolutional neural network on MNIST data and evaluate it on test data.

Uploaded by

eduardo
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
36 views

Neural Network With Numpy

This document describes a deep neural network framework implementation with numpy. It provides instructions for installing the framework from GitHub and with pip, running tests, and using the framework to build and train a convolutional neural network on MNIST data and evaluate it on test data.

Uploaded by

eduardo
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

17/9/21 1:59 GitHub - Javicadserres/DNet: Deep Neural Network framework implementation with numpy.

Javicadserres / DNet Public

Deep Neural Network framework implementation with numpy.

MIT License

1 star 0 forks

Star Notifications

Code Issues Pull requests Actions Projects Wiki Security

main Go to file

Javicadserres … on 9 Jul

View code

README.md

VERSION V1.1 CODE SIZE 163 KB ISSUES 0 OPEN

DNet
Neural Network framework implementation with numpy.

Installation
Normal user:

git clone https://github.com/Javicadserres/DNet.git


pip install DNet/.

alternatively:

git clone https://github.com/Javicadserres/DNet.git


pip install DNet/. -r DNet/requirements.txt

Developer:

https://github.com/Javicadserres/DNet 1/5
17/9/21 1:59 GitHub - Javicadserres/DNet: Deep Neural Network framework implementation with numpy.

git clone https://github.com/Javicadserres/DNet.git


pip install -e DNet/. -r DNet/requirements-dev.txt

Tests
To run tests you must install the library as a developer .

cd DNet/
pytest -v tests/

Usage
Import libraries

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

from sklearn.datasets import load_digits


from sklearn.model_selection import train_test_split
from DNet.model import NNet
from DNet.layers import (
LinearLayer,
ReLU,
CrossEntropyLoss,
Conv2D,
MaxPooling2D,
AveragePooling2D,
Flatten
)
from DNet.optimizers import Adam

digits = load_digits()

_, axes = plt.subplots(nrows=1, ncols=4, figsize=(10, 3))


for ax, image, label in zip(axes, digits.images, digits.target):
ax.set_axis_off()
ax.imshow(image, cmap=plt.cm.gray_r, interpolation='nearest')
ax.set_title('Training: %i' % label)

https://github.com/Javicadserres/DNet 2/5
17/9/21 1:59 GitHub - Javicadserres/DNet: Deep Neural Network framework implementation with numpy.

Create and train your model.

images = digits.images
m, h, w = images.shape
images = images.reshape(m, 1, h, w)

target = digits.target

def one_hot_encoding(Y):
"""
One hot enconding method.
"""
one_hot = np.zeros((Y.size, Y.max() + 1))
one_hot[np.arange(Y.size), Y] = 1

return one_hot

x_train, x_test, y_train, y_test = train_test_split(


images, target, test_size=0.4, random_state=1
)
y_train = one_hot_encoding(y_train)

# Initialize the model


model = NNet()

# Create the model structure


model.add(Conv2D(1, 2, kernel_size=(2, 2), stride=2, padding=1))
model.add(AveragePooling2D(kernel_size=(2, 2), stride=1, padding=1))
model.add(ReLU())

model.add(Conv2D(2, 1, kernel_size=(2, 2), stride=1, padding=0))


model.add(AveragePooling2D(kernel_size=(2, 2), stride=1, padding=0))
model.add(ReLU())

model.add(Flatten())
model.add(LinearLayer(16, 10))

# set the loss functions and the optimize method


loss = CrossEntropyLoss()
optim = Adam(lr=0.05)

# Train the model


costs = []
epochs = 10000
https://github.com/Javicadserres/DNet 3/5
17/9/21 1:59 GitHub - Javicadserres/DNet: Deep Neural Network framework implementation with numpy.

for epoch in range(epochs):


model.forward(x_train.T)
cost = model.loss(y_train.T, loss)
model.backward()
model.optimize(optim)

if epoch % 500 == 0:
print ("Cost after iteration %epoch: %f" %(epoch, cost))
costs.append(cost)

# plot the loss evolution


costs_ss = pd.Series(costs[1:])

plt.figure(figsize=(7, 3))
plt.plot(costs_ss)
plt.ylabel('cost')
plt.xlabel('iterations (per hundreds)')
plt.title('Loss per epoch')
plt.show()

Check test.

_, axes = plt.subplots(nrows=1, ncols=4, figsize=(10, 3))

predictions = model.forward(x_test.T)
predictions = predictions.argmax(axis=0)

for ax, image, label, pred in zip(axes, x_test.reshape((719, 8, 8)), y_test, predicti
ax.set_axis_off()
ax.imshow(image, cmap=plt.cm.gray_r, interpolation='nearest')
ax.set_title('Test: ' +str(label) + ' predicted: ' + str(pred))

https://github.com/Javicadserres/DNet 4/5
17/9/21 1:59 GitHub - Javicadserres/DNet: Deep Neural Network framework implementation with numpy.

Releases 2

v1.1 Latest
on 9 Jul

+ 1 release

Packages

No packages published

Languages

Python 100.0%

https://github.com/Javicadserres/DNet 5/5

You might also like