Electricity load forecasting

This example is adapted from https://github.com/probabl-ai/forecasting.

The goal is to have a dataset and prediction pipeline that can be used to experiment with features of skore and other packages.

Layout

• src/electricity_load_forecasting/ — core package: data loading, feature engineering, the skrub DataOp pipeline, CV splitter, metrics and plotting.
• scripts/ — analysis code grouped by study/theme: calibration/, model_comparison/, tuning/, inspection/ and checks/.
• datasets/ — committed historical load and weather data.
• results/ — generated experiment outputs (git-ignored), organized by theme. Legacy pre-reorganization results are kept under results/old/.
• docs/ — published HTML reports and the experiment index.

Each experiment result directory should contain a short README.md describing what was run, when, and with which settings. Results and generated pages must also record their full provenance (command, commit, weather source, features, resolved hyperparameters, source directories); see "Provenance (mandatory)" in AGENTS.md.

Quick start

With pixi:

pixi install
pixi run python scripts/model_comparison/cross_validate.py
pixi run python scripts/model_comparison/cross_validate_models.py \
    --model hgb_quantile --output_dir results/model_comparison/my_study

See the published reports and experiment overview in docs/index.html.