OpenSTEF Meta V0.1

[Lars800]

Introducting OpenSTEF-Meta, a one-stop-shop for all things meta learning.

This sub-package introduces four common meta learning algorithms:

Residual Forecaster

• 2 stage Forecasting Model
• Primary model is fitted as usual
• Secondary model fitted on residuals

Stacking Forecaster

• Multiple Base Forecasters
• Single Regressor is fitted on Base Predictions

Learned Weights Forecaster

• Multiple Base Forecasters
• Classification model learns optimal model weights

Rules Forecaster

• Multiple Base Forecasters
• Pre-defined rules on how to combine base methods.

---

This is an initial implementation

• The methods have been tested on the Liander 2024 Huggingface dataset.
• The results are accurate and improve over existing LGBM, XGBoost and GBLinear models
• The code can still be further optimized for efficiency

[MvLieshout]

Really nice additions to OpenSTEF, I have left some comments. Mostly small nitpicks.

[MvLieshout]

colsample_bytree not used in this file?

[Copilot]

Pull request overview

This PR introduces OpenSTEF-Meta v0.1, a new sub-package that provides meta-learning algorithms for ensemble forecasting. The package implements four forecasting approaches: Residual Forecaster (2-stage modeling on residuals), Stacking Forecaster (multiple models with meta-regressor), Learned Weights Forecaster (classification-based model selection), and Rules Forecaster (rule-based combination).

Changes:

• New openstef-meta package with ensemble forecasting models and forecast combiners
• Integration of ensemble models into existing OpenSTEF workflows and benchmarking infrastructure
• Addition of predict_contributions method to forecasters for explainability support
• New utility classes for decision trees, pinball loss calculation, and ensemble datasets

Reviewed changes

Copilot reviewed 55 out of 62 changed files in this pull request and generated 10 comments.

Show a summary per file

File	Description
packages/openstef-meta/*	New package implementing meta-learning algorithms
packages/openstef-models/src/openstef_models/workflows/*	Integration support for ensemble models
packages/openstef-models/src/openstef_models/models/forecasting/*	Added predict_contributions methods
packages/openstef-models/src/openstef_models/transforms/general/*	New Flagger transform and fixes
packages/openstef-beam/src/openstef_beam/benchmarking/*	Benchmark support for ensemble models
examples/benchmarks/*	Example benchmarks for ensemble and residual models
pyproject.toml, uv.lock	Dependency and workspace updates

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Misleading comment: The comment on line 57 says "Unchanged" but column C is actually transformed like all other columns. The Flagger transform converts all selected features to binary flags (0 or 1) indicating whether values are within training range. The comment should be removed or clarified, for example: "C: [0, 0] - both values are outside training range [1.0, 3.0]".

[Copilot]

Misleading docstring: The test docstring on line 43 says "leaves other columns unchanged" but the Flagger transform actually changes ALL selected columns by converting them to binary flags. Consider updating the docstring to accurately describe the test behavior, such as "Test fit and transform correctly flags all features based on training ranges."