Skip to content

Contributing

Thanks for wanting to contribute to AnyPINN! Please read our Code of Conduct before getting started.

Setup

  1. Fork and clone the repo, then install dependencies:

    uv sync

    Optional: If you use mise, run mise install first to provision the pinned python, uv, and just versions from .mise.toml.

  2. Before pushing, run the full CI check locally:

    just ci

    This runs lint, typecheck, and tests — the same pipeline as CI.

Tooling

  • Package manager: uv
  • Task runner: just
  • Linting & formatting: Ruff
  • Type checking: ty
  • Tests: pytest

Good to Know

  • The examples/ directory has ready-to-run problems you can use to exercise the library end-to-end while developing. Each example has a train.py entry-point (uv run python train.py).
  • exponential_decay is a minimal, core-only example (no Lightning).
  • uv run anypinn create --list-templates lists all scaffold templates.
  • Keep the layer separation: anypinn.core stays pure PyTorch, Lightning stays optional. anypinn.core must not import from anypinn.lightning, anypinn.problems, or anypinn.catalog.
  • If you change the architecture or data flow, update both CLAUDE.md and README.md.

Conventions

  • Commits: Conventional Commits (feat:, fix:, docs:, refactor:, test:, chore:)
  • fix: triggers a patch release (e.g. 0.1.0 → 0.1.1)
  • feat: triggers a minor release (e.g. 0.1.1 → 0.2.0)
  • feat!: or BREAKING CHANGE: triggers a major release (e.g. 0.2.0 → 1.0.0)
  • Branch names: feat/, fix/, docs/, etc.
  • Line length: 99
  • Absolute imports only — no relative imports
  • All config dataclasses: @dataclass(frozen=True, kw_only=True)