OpenSTEF (Open Short-Term Energy Forecasting) is an open-source, model-agnostic Python framework for creating short-term forecasts in the energy sector. It provides complete machine learning pipelines for data preprocessing, feature engineering, model training, probabilistic forecasting, and evaluation. Version 4.0.0 introduces a complete architectural refactor with enhanced modularity, full type safety, and modern Python development practices.
For more information, visit the OpenSTEF project page at LF Energy.
OpenSTEF 4.0.0 is organized as a monorepo with specialized packages under the packages/ directory:
| Package | Purpose |
|---|---|
| openstef | Meta-package installing all core components |
| openstef-core | Core utilities, dataset types, shared types and base models |
| openstef-models | ML models, feature engineering, data processing |
| openstef-beam | Backtesting, Evaluation, Analysis, and Metrics |
| openstef-meta | Meta models for OpenSTEF |
Requirements: Python ≥3.12, 64-bit OS (Windows, macOS, Linux)
# Minimal-but-runnable convenience layer (core + models with CPU XGBoost)
pip install openstef
# Everything, CPU flavour
pip install "openstef[all]"
# Or install individual packages
pip install openstef-models
pip install openstef-beam
pip install openstef-core
# Optional feature extras (additive)
pip install "openstef-models[lgbm]"
pip install "openstef-models[tuning]"
pip install "openstef-beam[all]"Compute runtimes (pick one per package). Packages with a heavy runtime ship mutually exclusive CPU and GPU builds as conflicting extras — install one, not both:
# Foundation models (Chronos-2): CPU everywhere, GPU on CUDA Linux/Windows
pip install "openstef-foundation-models[cpu]"
pip install "openstef-foundation-models[gpu]"
# XGBoost models: CPU or GPU
pip install "openstef-models[xgb-cpu]"
pip install "openstef-models[xgb-gpu]"The openstef meta-package always ships the CPU builds; for GPU, install the
component package directly with its [gpu] extra.
Using uv (recommended for development):
uv add openstefFor the complete installation guide including troubleshooting, see the Installation Documentation.
Tutorials and usage examples are available in the examples/ folder. See the examples folder's own README.md for an overview of available examples including:
- Forecasting quickstart
- Feature engineering
Additional tutorials are available in the online documentation.
Mozilla Public License Version 2.0 - see LICENSE.md for details.
This project includes third-party libraries licensed under their respective Open-Source licenses. SPDX-License-Identifier headers show applicable licenses. License files are in the LICENSES/ directory.
We welcome contributions to OpenSTEF 4.0!
Read our Contributing Guide - documentation for contributors including:
- How to report bugs and suggest features
- Documentation improvements and examples
- Code contributions and development setup
- Sharing datasets and real-world use cases
# Clone and set up for development (one command installs the full CPU dev env)
git clone https://github.com/OpenSTEF/openstef.git
cd openstef
uv sync
# Run tests and quality checks
uv run poe allGPU dev env (CUDA; Linux/Windows):
uv sync --no-default-groups --group dev-gpu. Do not use--all-groups/--all-extras— the CPU and GPU runtimes conflict.
Code of Conduct: We follow the Contributor Code of Conduct to ensure a welcoming environment for all contributors.
If you use OpenSTEF in your research or publications, please cite the project. Refer to the CITATION.cff file in this repository for the preferred citation format, or use:
Contributors to the OpenSTEF project. OpenSTEF — Open Short-Term Energy Forecasting. LF Energy, 2017–2025. Available at: https://github.com/OpenSTEF/openstef
- Slack: LF Energy Slack
- Email: openstef@lfenergy.org
- Community meeting: OpenSTEF four-weekly community meeting
- Issues: GitHub Issue Tracker
- Support Guide: How to get help