Document all dataclass/ classes across API Reference and Architecture
- API-Reference: extend `from eruption_forecast.dataclass import (...)`
block to pull in StationData plus the four
ClassifierEnsembleSummary types; add a new
"ClassifierEnsembleSummary family" section (ProbabilityPick,
SeedSummary, EruptionWindow, ClassifierEnsembleSummary) with
builder / consumer pointers and None-vs-guaranteed invariants;
add a new StationData section with fields, derived nslc / nslct,
a runnable example, and the raised ValueError cases.
- Architecture: add SeedSummary to the classifier_ensemble_summary.py
tree comment (was missing).
Sync wiki from repo
- API-Reference.md: expanded (+133 lines)
- Architecture.md, Explanation-Workflow.md, Output-Structure.md, Visualization.md: updated
- figures/: add pipeline_overview{,_v1,_v2}.png
Sync wiki with feature alias utilities
- API-Reference: add "Feature alias utilities" section covering the writer/reader/backfill helpers, label formatters, and heatmap label_style
- Architecture: refresh utils/pathutils, utils/dataframe, utils/formatting function inventories
- Output-Structure: note alias/description columns on top_features CSVs and reference load_feature_aliases/update_top_features_csv
docs(feature-count-sweep): add experimental sweep page + surface at entry points
New page:
- Feature-Count-Sweep.md — full coverage of FeatureCountSweep and
sweep_feature_count with a prominent Experimental warning, Quick Start
snippet, forecast-vs-cv mode table, estimator_mode trade-off, fail-fast
behaviour on missing prediction-matrix features, EvaluationModel
prediction-reuse wiring (with build_label shortcut), 1-SE parsimony
rule, complete example, and full constructor / function reference.
Cross-links + surfaces:
- Home.md — new nav row (marked Experimental).
- Training-Workflow.md — Post-Hoc top_n_features Sweep sub-section with
a minimal runnable snippet and link to the deep-dive page.
- API-Reference.md — FeatureCountSweep (Experimental) section covering
constructor, .fit(...), sweep_feature_count(...), and persistence.
- Architecture.md — feature_count_sweep.py added to the features/
module listing.
- Output-Structure.md — sweep outputs nested under
training/features/{cv-slug}/sweep/{mode}/{classifier-name}/ and a
matching row in the What Lives Where cheat sheet.
Also captures accumulated smaller edits since the last wiki push
(evaluation, explanation, output-structure clarifications).
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.
Update docs: SeedEnsemble filename, ClassifierEnsembler.pkl, ModelPredictor fix
Update wiki: sync all pages from local wiki/ directory (2026-04-26)
Update wiki: API-Reference, Architecture, Pipeline-Walkthrough, Visualization
docs: sync wiki pages from local repository
Add GPU acceleration documentation for XGBoost
Document use_gpu and gpu_id parameters in ModelTrainer, parallelism
restrictions when GPU is active, and GPU-capable classifier annotations
for xgb and voting across Training-Workflows, Classifiers-and-CV,
and API-Reference pages.
docs: update SHAP sections to reflect beeswarm refactor
- Replace all 'mean |SHAP| bar chart' references with 'beeswarm showing
feature contributions across seeds'
- Update plot_aggregate_shap_summary examples: return is shap.Explanation
not pd.DataFrame
- Add aggregate_shap_summary.png + .pkl to output structure tree
- Sync API-Reference and Evaluation-and-Forecasting tables accordingly
Add wiki pages (13 pages)