Introduction to Deep Learning Library

This library provides foundational building blocks for creating and training simple neural networks. It was developed as study material for the "Introduction to Deep Learning" course at PACE University, 2023.

Features

• Layer Base Class: A foundational class for creating custom layers.
• Fully Connected Layer (FCLayer): Implements a fully connected neural network layer.
• Activation Layers: Supports tanh, Softmax, and ReLU activation functions.
• Loss Functions: Implements the mean squared error (mse) and binary cross-entropy.
• Pooling Layer: Implements max pooling.
• Convolutional Layer: Implements 2D convolution.
• Flatten Layer: Flattens the input.
• Network Class: Represents a neural network. It allows adding layers, training the network, and making predictions.

Usage

Creating a Network:

net = Network()

Adding Layers:

net.add(FlattenLayer())
net.add(FCLayer(input_size, output_size))
net.add(ActivationLayer(activation_function='tanh'))

Training:

net.fit(x_train, y_train, epochs = 20, learning_rate = 0.001, evaluation=0.2)

Prediction

predictions = net.predict(input_data)

Evaluation

accuracy = net.evaluation(x_test, y_test)

MNIST Example

from ann import Network
from layers import FlattenLayer, ActivationLayer, FCLayer
from keras.datasets import mnist
from keras import utils

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(60000, 1, 28*28)
x_test = x_test.reshape(10000, 1, 28*28)

x_train = x_train/255
x_test = x_test/255

y_train = utils.to_categorical(y_train)

model = Network()
model.add(FlattenLayer())
model.add(FCLayer(28*28,10))
model.add(ActivationLayer())

model.fit(x_train,y_train, 20, 0.01, evaluation = 0.1)

Dependencies

• numpy
• scipy

Installation

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

git clone https://github.com/Bakhitovd/IntroDL