Synapse-nn

A pure Python neural network framework built from scratch, featuring a custom matrix engine, dense layers, and activation functions. Synapse is designed as an educational project to explore how modern deep learning systems work under the hood.

Features

• Custom Matrix implementation
• Dense (Fully Connected) Layers
• ReLU Activation
• Sigmoid Activation
• Tanh Activation
• No external deep learning frameworks

Installation

pip install synapse-nn

Quick Example

from synapse import Matrix, Dense, ReLU

x = Matrix([[1, 2]])

layer = Dense(2, 4)
activation = ReLU()

output = activation.forward(layer.forward(x))

print(output)

Project Roadmap

Current

• Matrix Engine
• Dense Layers
• ReLU
• Sigmoid
• Tanh

Upcoming

• Loss Functions
• Backpropagation
• Optimizers
• Sequential API
• XOR Training Example
• Iris Dataset Training
• Model Saving & Loading

Why Synapse?

Neural networks are inspired by biological neurons connected through synapses. This project aims to recreate the core building blocks of modern deep learning frameworks from first principles, providing a deeper understanding of the mathematics and implementation behind AI systems.

Repository Structure

synapse/
├── matrix.py
├── layers.py
├── activations.py
├── losses.py
├── optimizers.py
└── examples/

License

MIT License