docs(usage,pipeline): surface the explain stage
- Usage.md: add explain() to Quick Start, Stage Cheat Sheet, and the
annotated example (mirroring main.py:100-107). Order the annotated
example as evaluate -> compare -> explain so .compare() stays with
the evaluation stage and explain() sits last.
- Pipeline-Walkthrough.md: update the Research Workflow intro/blurb,
add an explain box to the main.py stage-flow diagram, and rewrite
the "Optional: fm.explain(...)" per-stage note (it is no longer
optional and is called in main.py). Diagram flows
calculate -> train -> predict -> evaluate -> compare -> explain.
docs(prediction): document use_features_from modes
- Prediction-Workflow: add "Feature Scoping via use_features_from" section
covering the three modes ("all" / "files" / "training"), failure-mode
table, cache implications, and mode chooser.
- Configuration: add use_features_from, features_matrix_path,
label_features_csv, enable_segments_plot to the predict YAML block.
- API-Reference: extend fm.predict() signature with the four new kwargs
plus a compact mode table.
- Usage: add "Reuse features already selected during train()" variant.
Sync wiki from eruption-forecast/wiki
Sync wiki from eruption-forecast/wiki
Adds Explanation-Workflow page and refreshes existing pages.
Full rewrite of all 13 wiki pages against current src/ surface
Realigned with the post-ft/metrics-ensemble model/ refactor: dropped
references to removed ModelTrainer/ModelPredictor/ModelEvaluator/
MultiModelEvaluator and the old extract_features → train → forecast API;
documented the current TrainingModel/PredictionModel/EvaluationModel +
MetricsEnsemble stack, the BaseModel + CacheModel mixin pattern, and the
scenarios.py workflow.
Deleted 5 superseded pages (Installation, Quick-Start, Classifiers-and-CV,
Evaluation-and-Forecasting, Training-Workflows). Created 5 new pages
(Getting-Started, Usage, Training-Workflow, Prediction-Workflow,
Evaluation-Workflow). Rewrote 8 in place (Home, Data-Sources,
Pipeline-Walkthrough, Visualization, Configuration, Output-Structure,
Architecture, API-Reference). Final count 13 pages; no dead old links.