Test Strategy

Test Strategy#

DeePTB tests are grouped by review purpose. The markers help maintainers pick a useful test subset for a PR without changing the existing full test workflow.

Markers#

smoke tests are quick checks for imports, config generation, model build, and core mapping entrypoints. They should be stable and low-dependency.

regression tests protect known behavior, numerical semantics, tensor shape/order, and compatibility. High-risk changes should usually add or update a regression test.

slow tests cover training, complex postprocess, external or optional dependency paths, and workflows that are noticeably longer-running.

Local Commands#

uv run pytest ./dptb/tests -m smoke
uv run pytest ./dptb/tests -m regression
uv run pytest ./dptb/tests -m "not slow"
uv run pytest ./dptb/tests

Review Guidance#

  • Run smoke for quick confidence before reviewing a broad PR.

  • Run regression when a PR touches high-risk files from docs/maintenance/risk_map.md.

  • Run not slow when you want a wider local check without the heaviest workflows.

  • Run the full test suite before merging changes that affect model behavior, training, Hamiltonian construction, datasets, or public workflows.

The first version of this strategy does not make markers a required CI gate. The existing full CI workflow remains unchanged.