This code provides the data preprocessing framework used in the publication "Towards Flexible Building Energy Surrogate Models: A Novel Feature Representation Method", designed to generate flexible, physically meaningful feature representations for surrogate modeling of building energy performance.

The framework extracts fixed-length feature vectors directly from building geometry, construction assemblies, and climate data, offering an alternative to conventional design-parameter-based inputs. Four representations are supported: a conventional parameter-based baseline and three parameter-independent alternatives.

Parameter-based features (baseline), using the raw design parameters of a given design space, Global aggregated features, summarizing geometric, material, and climatic quantities via sums, means, and extrema into a compact vector, Coarse directional features, capturing orientation-dependent effects across six cardinal directions, and Fine directional features, resolving orientation dependence further using a 302-patch orientation discretization.

Because these representations are derived from the physical building description rather than a specific parameterization, they can in principle generalize across varying design spaces. The target objective computed by the pipeline is annual heating energy demand.

The software is implemented as a flexible, extensible prototype supporting integration with custom datasets and design space exploration workflows. For more information on setup, usage, and citation, please refer to the README.md file included in the repository.