Development¶
Environment¶
Use /work/envs/depth/bin/python for Python commands in this repository. Install
the project dependencies from the root metadata:
/work/envs/depth/bin/python -m pip install -r requirements.txt
Repository layout¶
| Path | Purpose |
|---|---|
train.py |
Training CLI and Lightning orchestration. |
src/depth_recon/configs/ |
Pixel and latent YAML configs plus scenario resolvers. |
src/depth_recon/data/ |
Active GeoTIFF/Zarr dataset and data-production tools. |
src/depth_recon/models/ |
Diffusion, EMA, autoencoder, and baseline models. |
src/depth_recon/inference/ |
Public API and export/analysis workflows. |
docs/ |
MkDocs pages, static viewers, JavaScript, styles, and assets. |
tests/ |
Complete unittest-based regression suite. |
dataset_argo_netcdf_gridded.py is a legacy dataset. New documentation and
features should use ArgoGeoTIFFGriddedPatchDataset and do not need compatibility
with the legacy loader.
Working conventions¶
- Keep edits focused and follow nearby naming, docstring, and error-handling style.
- Keep validation dataloaders shuffled; this is intentional repository behavior.
- Add an all-options invocation comment to new task-specific CLI scripts.
- Explain non-obvious logic inline and document new functions.
- Do not treat generated synthetic targets as observations or scientific truth.
- Update docs whenever a config key, artifact schema, CLI, or public interface changes.
Checks¶
Format Python changes:
/work/envs/depth/bin/python -m black .
Build the complete API-enabled site:
ENABLE_MKDOCSTRINGS=true /work/envs/depth/bin/python -m mkdocs build --strict
Run the whole test suite for substantive changes:
tests/run_tests.sh
See Tests for the coverage map and CLI reference for maintained commands.