Skip to content

Testing

Real data only

Correctness tests use real sample EEG and the reference Fortran binary shipped in pamica/sample_data/. Mocks, stubs, and synthetic data are not used as the basis for correctness tests: no test is better than a fake passing test.

Running the suite

uv run pytest                    # full suite
uv run pytest --cov              # with coverage
uv run pytest pamica/tests/torch_tests/   # PyTorch-vs-Fortran parity tests
uv run pytest -m "not slow"      # skip the slow tests, as CI does
AMICA_RUN_FORTRAN=1 uv run pytest  # also run the tests that execute the native binary

Tests that run the Fortran binary, including the element-wise oracles that seed it from a pamica state (pamica/tests/native_oracle.py), are opt-in with AMICA_RUN_FORTRAN=1; the weekly macOS workflow runs them.

Working directory

The suite runs from a temporary working directory, not the repository: a session-scoped autouse fixture in pamica/tests/conftest.py switches to one per session (per worker under pytest-xdist), so anything a test writes to a relative path lands there instead of in the repository. A test that needs a repository path resolves it from __file__ (for example Path(__file__).resolve().parent.parent / "sample_data"), or, when a relative path inside a file must resolve against the repository root (such as sample_params.json's files), changes into the root explicitly with monkeypatch.chdir. Coverage reports still land in the repository root.

Layout

  • pamica/tests/: end-to-end and interface tests, and the cross-backend tests every shared behavior needs.
  • pamica/tests/torch_tests/: natural-gradient backend parity, PDF families, component sharing, float32 stability, and edge cases.
  • pamica/tests/mlx_tests/: MLX backend tests (Apple Silicon).
  • pamica/tests/mne_tests/: the MNE wrapper, including the end-to-end workflow test.
  • validate_implementations.py: cross-implementation validation harness (Hungarian component matching against Fortran; --backend selects torch, numpy, mlx, a list, or all).