NCxLib: A Lightweight Neural Network Library in Python

ncxlib is a lightweight and easy-to-use neural network library built in Python. It provides a simple API for constructing and training neural networks, along with tools for data preprocessing and generation.

Features

• Modular Design: Easily build custom neural networks by combining different layers, activation functions, and loss functions.
• Data Handling: Includes data loaders for CSV and image data, with preprocessing capabilities like scaling and grayscaling.
• Training and Evaluation: Train your networks with various optimization algorithms and evaluate their performance.
• Extensible: Add your own custom layers, activations, and loss functions to expand the library's functionality.

Installation

pip install ncxlib

Getting Started

Here's a quick example of how to use ncxlib to create and train a simple neural network:

# External imports
import numpy as np

# Util imports
from ncxlib import generators, dataloaders, optimizers, losses, activations
from ncxlib.util import train_test_split

# Neural network imports
from ncxlib.models.neuralnetwork import NeuralNetwork, FullyConnectedLayer
from ncxlib.initializers import HeNormal, Zero


# ------- Generate some data using generators -------
generators.generate_training_data(to_csv=True)

# ------- Load data from generated csv and split it into train and test -------
loader = dataloaders.CSVDataLoader("training_data.csv")
X, y = loader.get_data()
X_train, X_test, y_train, y_test = train_test_split(X, y)


# ------- Configure model layers -------
model = NeuralNetwork([
    FullyConnectedLayer(
        n_neurons=3, 
        activation=activations.ReLU, 
        optimizer=optimizers.Adam(beta_1=0.9, beta_2=0.999, epsilon=1e-07),
        name="first_hidden",
        weights_initializer=HeNormal(), 
        bias_initializer=Zero()
        ),

    FullyConnectedLayer(
        n_neurons=5, 
        activation=activations.ReLU, 
        optimizer=optimizers.SGDMomentum(momentum = 0.9), 
        name="second_hidden",
        initializer=HeNormal(),
        ),

    FullyConnectedLayer(
        n_neurons=2, 
        activation=activations.Sigmoid, 
        optimizer=optimizers.RMSProp(decay_rate = 0.8)
        )
],
    loss_fn=losses.BinaryCrossEntropy
)

# ------- Train model and evaluate accuracy -------
model.train(X_train, y_train, epochs=20, learning_rate=0.01)
results = model.evaluate(X_test, y_test, metrics=['classification', 'roc'])

Contributing

Thank you for your interest in contributing to the ncxlib library for Neural Network development. We are thrilled you are considering contributing to our project.

How to Contribute

• Take a look at the list of immediate contributions needed listed under the issues
• Look for areas around the repository with comments marked #TODO. If you find one, feel free to create an issue and get approval before starting the work.
• Any suggestions or feedback - please create an issue.
• Open a Pull Request with your issue and the team will review and approve/deny or provide comments on the PR.

License

This project is licensed under the MIT License