Skip to main content

Contributing

Contributions are welcome. Please follow these guidelines.

Code Style​

  • C++23. Use modern features (concepts, [[nodiscard]], structured bindings) where they add clarity.
  • No raw loops over matrix elements. Use Eigen expression templates.
  • No virtual. Use CRTP for polymorphism.
  • Warning-free under -Wall -Wextra -Wpedantic.

Adding a New Estimator​

  1. Create the internal header at Skigen/src/Module/Name.h.
  2. Inherit from the appropriate CRTP base (Transformer or Predictor).
  3. Implement fit_impl, transform_impl / predict_impl, and accessors.
  4. Add the header to the module file (e.g. Skigen/Preprocessing).
  5. Write tests in tests/test_name.cpp.
  6. Add a parity script in tests/parity/.
  7. Document in doc/website/docs/guide/.

Testing​

Every estimator must have:

  • Unit tests for correctness.
  • A Python parity script that generates reference data from scikit-learn.
  • C++ parity tests that compare Skigen output against the reference.

Tolerance: 1e-12 for double, 1e-5 for float.

Pull Requests​

  • Keep changes focused — one estimator or feature per PR.
  • Ensure all tests pass before submitting.
  • Include benchmark results if the change affects performance.