Lightweight, educational, and extensible neural network building blocks in pure Python.

NN is a minimal, easy-to-read neural network library implemented with NumPy. It provides core components — layers, activations, losses, optimizers, and a simple DNN model — so you can quickly prototype and learn how deep learning models work under the hood.

Key features:

• Small, modular codebase for learning and experimentation
• Forward / backward passes implemented for all layers
• Basic optimizers and loss functions included
• Easy to extend with new layers, activations or optimizers

NN/
│
├── main.py                # Entry point for running experiments or training models
├── requirements.txt       # Python dependencies
├── setup.py               # Setup script for packaging
├── LICENSE
├── README.md
│
├── activations/           # Activation functions
│   ├── __init__.py
│   ├── activations.py     # Implementations of activation functions (ReLU, Sigmoid, Tanh, Softmax)
│   └── base.py            # Base class/interface for activations
│
├── layer/                 # Neural network layers
│   ├── __init__.py
│   ├── base.py            # Base class/interface for layers
│   └── layers.py          # Implementations of layers (Dense, etc.)
│
├── loss/                  # Loss functions
│   ├── __init__.py
│   ├── base.py            # Base class/interface for loss functions
│   └── loss.py            # Implementations of loss functions (MSE, CrossEntropy, etc.)
│
├── model/                 # Model definitions
│   ├── __init__.py
│   └── dnn.py             # Deep Neural Network class (model architecture, forward/backward pass)
│
└── optimizer/             # Optimizers
    ├── __init__.py
    ├── base.py            # Base class/interface for optimizers
    └── optimizer.py       # Implementations of optimizers (SGD)

Recommended: create and activate a virtual environment first.

python -m venv .venv
source .venv/bin/activate  # Linux / macOS
pip install -r requirements.txt
# or for development (editable install)
pip install -e .

• model.dnn.DNN — main class: add(layer), compile(loss, optimizer), train(x, y, epochs, batch_size), predict(x), summary(), plot_history(history)
• layer.layers.Dense(input_size, output_size, activation) — fully connected layer
• activations.activations — Sigmoid, Relu, Tanh, Softmax
• loss.loss — MeanSquaredError, CategoricalCrossEntropy
• optimizer.optimizer — GD(learning_rate)

Contributions are welcome!

• Open an issue to discuss larger changes
• Fork the repo and make small, focused pull requests
• Follow existing code style and include tests when applicable

This project is licensed under the terms in the LICENSE file.

If you have questions or suggestions, open an issue or submit a PR. Contributions and feedback are appreciated!