DSV BIPV: Data and Code for Task-Dependent Directional Shading Resolution

Data and code accompanying the manuscript "Task-Dependent Directional Shading Resolution for Facade BIPV: Surrogate Prediction, Transfer, and EnergyPlus-Verified Screening of Project-Owned Massing Actions," prepared for submission to Energy and Buildings. This deposit contains a synthetic facade BIPV dataset generated with EnergyPlus, directional shading representations, trained XGBoost models, analysis code, and result artifacts supporting the manuscript. The corpus comprises 21,638,400 panel records from 5,474 building-climate cases across seven climates. Shading is represented as a 548-direction binary field and compressed alternatives, including scalar sky-view factor and azimuth-sector encodings. The study evaluates these representations across three tasks: case-grouped surrogate prediction, transfer across climates and geometries, and EnergyPlus-verified screening of project-owned massing actions. The package includes the complete synthetic corpus, canonical modelling subset, directional grids, evaluation cohorts, trained models, canonical results, sensitivity analyses, source records, and validation documentation. Every headline number in the manuscript is mechanically audited against the packaged canonical results. To verify package integrity and reproduce the principal results, download all files, extract the seven ZIP archives into one directory alongside the loose files, and run: pip install -r requirements.txtpython3 VERIFY_FRESH_ENV.pypython3 REPRODUCE_HEADLINES.py VERIFY_FRESH_ENV.py checks package integrity. REPRODUCE_HEADLINES.py recomputes the design-action and real-building contrasts. For full model retraining in the pinned environment, run: pip install -r requirements-lock.txtpython3 REPRODUCE_HEADLINES.py --full ARCHIVE_SHA256SUMS.txt verifies the uploaded files. See README.md for the package structure, environment requirements, and detailed instructions. Building footprints for the 21-building zero-shot evaluation were derived from the 3D building LOD1 dataset of the National Geographic Information Institute (NGII), Republic of Korea. The underlying geometry is not redistributed and is available from NGII. TMYx EPW weather files are also not redistributed; their sources and hashes are documented in the package. Code is licensed under the MIT License. Data, results, and documentation are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).