myneuralnet 0.1.0

A simple neural network library built from scratch using NumPy

myneuralnet

A simple, lightweight neural network library built from scratch using NumPy.
This project is designed for educational purposes to help understand how neural networks work under the hood — without relying on high-level frameworks like TensorFlow or PyTorch.

🚀 Features

• Fully connected feedforward neural network
• Supports custom architectures (you define number of layers and neurons)
• Activation functions: ReLU, Sigmoid, Softmax, Linear (more can be added)
• Forward propagation & backpropagation
• MSE loss for regression tasks (you can extend to classification easily)
• Training with gradient descent
• Designed to be minimal and beginner-friendly

📦 Installation

Once you’ve built your package, you can install it locally using:

pip install -e .

🧠 Example Usage

from myneuralnet.network import NeuralNetwork
from myneuralnet.layers import Layer

# Create a network
net = NeuralNetwork()
net.add(Layer(units=4, activation_function="relu", input_dim=2))
net.add(Layer(units=4, activation_function="relu"))
net.add(Layer(units=1))  # No activation = linear for regression

# Train
net.train(X_train, y_train, epochs=1000)

# Predict
predictions = net.predict(X_test)

📁 Project Structure

myneuralnet/
│
├── network.py         # NeuralNetwork class: manages training, prediction
├── layers.py          # Layer class: handles weights, activation, backprop
├── activations.py     # Activation functions and their derivatives
├── utils.py           # Utility functions (e.g., loss, metrics)
├── __init__.py        # Makes the folder a Python package

🧪 Dependencies

• numpy

📈 Future Improvements

• Add support for classification with softmax & cross-entropy
• Include optimizers like Adam or Momentum
• Add regularization (L2, dropout)
• Save/load trained models
• Add unit tests

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Made with ❤️ by Om Asanani