Skip to content

v2.0.0

Choose a tag to compare

@mortenoh mortenoh released this 08 Jun 18:45
· 21 commits to master since this release
daa015d

Highlights

  • Chapkit self-registration: chapkit model services now self-register and auto-sync to the database on discovery, with auto-archival when a service disappears.
  • Breaking: removed the deprecated evaluate / evaluate-hpo CLI commands (CLIM-438).
  • PredictionSetup: new PredictionSetup abstraction for configuring a model with its data source.
  • Causal analysis: new causal CLI commands (build-counterfactual) and a counterfactual comparison plot.
  • Pluggable thresholds: new threshold calculation endpoint with a pluggable registry.

Features

  • Make chap test a real self-diagnostic (#414)
  • Add chap aggregate-eval CLI subcommand (#385)
  • Add chap model schema CLI subcommand (#374)
  • Add generate-modelcard CLI command (#307); region labels in model card output (#387)
  • Select HPO searcher via estimator options (#365); n_retrain param in BacktestParams (#408)
  • Add --track flag to chap eval for MLflow run tracking (#360)
  • Add report endpoint for model-produced PDF reports (CLIM-179) (#298)
  • New backtest/metric plots: Predicted vs Actual linear (#313), regional metric/RMSE distribution (#289, #266)
  • Add configured-model info endpoint (#297) and backtest-metrics CSV download (#296)
  • Add /actual-cases alias for /actualCases analytics endpoint (#367)
  • OpenAPI: Field descriptions + class docstrings on exposed schemas (#383)
  • Reusable CLI argument annotations (#358); helm2 chart (#231)
  • Compose overlays for chapkit/ewars (#312, #273)

Bug Fixes

  • aggregate-eval preserves NaN for all-NaN cells instead of fabricating 0 forecasts (#416)
  • Copy chap report output from runs directory to --out-file path (#369)
  • Skip chapkit probe for GitHub URLs (#411)
  • Warn on unused CSV columns when running a model (#351)
  • Wrap MlFlow predict failures as ModelFailedException (#389)
  • Linearize Alembic migration chain and add single-head guard (#372)
  • Return the actual entity from /prediction_result and /evaluation_result (#363)
  • Support CHAP_ROOT_PATH env var for serving behind a path prefix (#355)
  • Default matplotlib backend to Agg to avoid Tk crash on Linux (#334)
  • Forbid extra fields on ModelConfiguration (#342); honour prediction-length bounds (#332)

Refactoring

  • Extract DataSetManager with filterable observation reads (#392)
  • Replace debug endpoints with /health/ready dependency check (#366)
  • Standardize on Backtest class naming (#322)

Documentation

  • Local DHIS2 install guide, troubleshooting (logs/DB), model metadata page, REST API + DB architecture guide, CLI-command contributor guide, mermaid diagrams.