rajpbhavsar/zambia-vgae-tranching

Settlement-scale graph-based spatial aggregation for Zambia electrification planning # zambia-vgae-tranching Code and analysis-ready data for a settlement-scale, graph-based spatial aggregation study of electrification planning in Zambia. The repository supports the paper's connected-planning-unit contribution: a topology-aware aggregation layer that converts point-level electrification outputs into connected, technology-characterised planning units and interpretable supply-curve bands. ## Scientific purpose Using 24,550 Zambia settlements (World Bank Distributed Renewable Energy Atlas settlement locations, matched to World Bank Global Electrification Platform V2 2030 scenario outputs), this repository compares three tranching methods for grouping settlements into electrification-planning bands: - **k-means** on standardised features (feature-space baseline), - **cost-bin** tranching on ordered levelised cost of electricity (LCOE), - **VGAE** (Variational Graph Auto-Encoder) tranching, using a learned embedding of a settlement proximity graph, clustered at K=8. The comparative evaluation measures spatial contiguity and supply-curve smoothness across the three methods. The repository's core contribution is the connected-planning-unit layer: each VGAE tranche is decomposed into connected components under a stored settlement proximity graph, producing local, spatially connected planning units with review-candidate flags and per-unit supply curves, in addition to the national-scale tranche bands. ## Quick start ```bash python -m venv .venv ``` Windows: ```bash .venv\Scripts\activate ``` macOS/Linux: ```bash source .venv/bin/activate ``` ```bash python -m pip install --upgrade pip python -m pip install -r requirements.txt python scripts/verify_checksums.py python scripts/reproduce_paper_results.py python scripts/validate_release.py ``` `requirements-dev.txt` additionally installs `pytest` for the test suite (`python -m pytest -q`). `requirements-training.txt` lists the optional, heavier dependencies (PyTorch, PyTorch Geometric) needed only t …