Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
37 lines (24 loc) · 1.17 KB

CONTRIBUTING.md

File metadata and controls

37 lines (24 loc) · 1.17 KB

Contributing

Thanks for your interest in this project. This is primarily a learning project — a neural network library built from scratch using only NumPy — but contributions are welcome.

What's useful

  • Bug fixes (incorrect gradients, shape mismatches, numerical instability)
  • New layer types (e.g. LSTM, embedding, attention)
  • New examples that demonstrate existing functionality
  • Documentation improvements

What to avoid

  • Adding framework dependencies (PyTorch, TensorFlow, JAX) — the point is pure NumPy
  • Rewriting working code for style preferences
  • Large refactors without discussion first

How to contribute

  1. Fork the repo and create a branch from main
  2. Make your changes
  3. Test that existing examples still run (python examples/xor_example.py, python examples/cnn_mnist_demo.py)
  4. Open a pull request with a clear description of what you changed and why

Gradient correctness

If you add a new layer, verify its backward pass with the numerical gradient checker:

python examples/gradient_check.py

Relative error should stay below 1e-4.

Questions

Open an issue — happy to discuss ideas before you spend time building them.