Examples#

End-to-end tutorials and benchmarks demonstrating OpenSTEF workflows. Each tutorial is a runnable Jupyter notebook rendered with executed outputs.

Tutorials#

Tutorial	Description
Forecasting Quickstart	Train your first XGBoost model and produce a 47-hour ahead forecast using the Liander dataset.
Backtesting Quickstart	Evaluate forecast accuracy on historical data using BEAM’s rolling-window backtesting pipeline.
Understanding Datasets	Learn how versioned time series data works and why it matters for honest forecasting.

Tutorial	Description
Custom Pipeline	Build a forecasting workflow from scratch with custom transforms, feature selection, and callbacks.
Feature Engineering	Explore transforms by domain, apply them individually, and compose preprocessing pipelines.
Ensemble Forecasting	Combine multiple models into an ensemble for improved accuracy and robustness.
Hyperparameter Tuning	Optimize model hyperparameters using Optuna integration with cross-validated backtesting.

Tutorial	Description
Model Explainability	Inspect feature importances, SHAP values, and contribution plots for trained models.
Quantile Calibration	Calibrate prediction intervals using isotonic regression for reliable uncertainty estimates.

Compare models on real energy data. These notebooks are not executed during the docs build — run them locally to reproduce results.

Benchmark	Description
Benchmarking Guide	Overview of the benchmarking framework and how to interpret results.
Liander 2024	Full benchmark on Liander’s MV feeder dataset comparing XGBoost, LightGBM, and linear baselines.
Build Your Own	Template for creating custom benchmarks on your own data.