Tests

DepthDif uses the standard-library unittest runner through a repository wrapper. Run the complete suite from the repository root:

tests/run_tests.sh

The wrapper uses /work/envs/depth/bin/python. Targeted test execution is useful while diagnosing a failure, but substantive changes are accepted against the full suite.

Coverage map

Area	Representative coverage
Data creation	EN4 enrichment, potential-temperature selection, GeoTIFF export, compact Zarr, packaged downloads, and Hugging Face layout.
Dataset contract	Quantization, masks, fields, scenarios, synthetic targets, hard-region sampling, splits, and datamodule wiring.
Model	Diffusion math, schedules, DDPM/DDIM samplers, model dry runs, losses, coordinate/date conditioning, EMA, and baselines.
Configuration	Pixel scenario resolution, preset invariants, checkpoint compatibility, and CLI override parsing.
Public API	Asset resolution, source downloads, weekly patch selection, mosaics, metadata, and package exports.
Evaluation	Global inference, EN4 holdouts, hard regions, validation summaries, paper-week bundles, and metrics.
Analysis products	Cesium assets, comparison globe, spatial/temporal dashboards, temporal globe, and wavenumber spectra.

Tests use temporary directories and small synthetic fixtures where possible. They do not establish scientific skill or validate external hosted data; they verify the repository contracts and deterministic transformations.

Documentation checks

Build with API rendering enabled so malformed docstrings are not hidden:

ENABLE_MKDOCSTRINGS=true /work/envs/depth/bin/python -m mkdocs build --strict

For notebook changes, validate the JSON and compile each Python code cell without running downloads or GPU inference.