yohou-optuna

Step-by-step guide for implementing new point or interval forecasters in Yohou. Covers BasePointForecaster, BaseIntervalForecaster, and BaseReductionForecaster patterns. Includes class templates with fit/predict, _parameter_constraints, panel data support, fitted attributes, and test file templates with _yield_yohou_forecaster_checks (27 systematic checks). Use when creating, extending, or testing any forecaster class.

Guide for adding new bundled datasets to Yohou. Covers the full workflow: data preparation with polars, Parquet export (zstd compression), Git LFS tracking, loader function creation in loaders.py, schema conventions (univariate/multivariate/panel with __ separator), Marimo example notebook creation, and test templates. Includes file size guidelines, licensing metadata, and external hosting patterns (Zenodo, HuggingFace) for large datasets.

Guide for creating new Plotly-based plotting functions in Yohou. Covers the plotting module structure, validate_plotting_data/resolve_color_palette/validate_plotting_params/apply_default_layout utilities, panel data faceting/dropdown patterns, consistent parameter naming conventions, and test templates for figure validation. Use when adding any new visualization function to src/yohou/plotting/.

Step-by-step guide for implementing new point and interval metrics (scorers) in Yohou. Covers BasePointScorer and BaseIntervalScorer with score() method pattern, aggregation methods (timewise/componentwise/groupwise/coveragewise/all), time weighting support, panel data scoring, and test templates with _yield_yohou_scorer_checks (11 check functions, variable yield count). Use when creating or testing any metric/scorer class.

Guide for extending hyperparameter search in Yohou (advanced). Covers custom search strategy implementation extending BaseSearchCV, with the _run_search abstract method pattern. Includes built-in GridSearchCV/RandomizedSearchCV usage examples and test templates with _yield_yohou_search_checks (19 systematic checks). Most users should use built-in search classes — extend only for novel strategies like Bayesian optimization.

Step-by-step guide for implementing new time series cross-validation splitters in Yohou. Covers expanding window, sliding window, and gap patterns extending BaseSplitter. Includes split/get_n_splits/_iter_test_indices templates, panel data row-index splitting, and test templates with _yield_yohou_splitter_checks (8 systematic checks). Use when creating or testing custom CV splitters.

Step-by-step guide for implementing new time series transformers in Yohou. Covers stateful (observation_horizon, memory) and stateless patterns, validate_transformer_data usage, inverse_transform, panel data support, and feature name output. Includes class templates for BaseTransformer subclasses and test templates with _yield_yohou_transformer_checks (26 systematic checks). Use when creating, extending, or testing any transformer class.