Contributing to the tracking module¶
Tracking is a backend plugin selected via the top-level tracking._target_ field. The current built-in backend is MLFlowTracker.
BaseTracker interface¶
log_config()log_model()log_dataset()log_artifacts()log_metrics()terminate()
Runtime flow¶
Tracking runs after the local assessment phase. src/raitap/pipeline/orchestrator.py first runs the forward pass, optional metrics, and optional transparency artifacts in the Hydra run directory. Only then, if tracking is enabled, it constructs the tracker and calls:
log_config()log_model()iftracking.log_model=truelog_dataset()metric scalars and metric artifacts
transparency explanation artifacts
transparency visualisation artifacts
This split keeps the metrics and transparency modules responsible for their own tracker-facing translations; the tracker backend only handles generic logging calls.
Model logging¶
Enabled with tracking.log_model=true.
Torch-backed models →
mlflow.pytorch.log_model(...).ONNX-backed models →
mlflow.onnx.log_model(...). Requires theonnxpackage in addition tomlflow.
Auto-opening¶
If tracking.open_when_done=true, terminate() must open the tracking UI automatically. See MLFlowTracker in src/raitap/tracking/mlflow_tracker.py for the reference implementation.
Extension points¶
See Adding an adapter.