TristCrocker/PopulationDisaggregationDiss

Population Dissagregation using Graph Neural Networks in Sub-Saharan Africa (4th Year University of Edinburgh Computer Science Dissertation) # Population Disaggregation using Graph Neural Networks in Sub-Saharan Africa This repo is intended to be used as part of a 4th year computer science Dissertation at The University of Edinburgh. This project is used for baseline result production and Graph Neural Network (GNN) result production in population disaggregation (admin level 2 to 3) on a Mozambican population and covariate dataset. **Baseline Models**: * Random Forest (RF) * Bayesian Additive Regression Tree (BART) **GNN Models**: * GCNv2 * GATv2 * GraphSage * Transformer GNN ## Getting Started (Baseline) Cd into project and start R env, ```bash cd dissBaseline/Bayesian-Top-Down-Modelling R ``` To run RF baseline train/test, ```bash source("RF_Baseline_Workflow.R") ``` To run BART baseline train/test, ```bash source("BART_Baseline_Workflow.R") ``` ## Getting Started (GNN) Cd into project, ```bash cd dissGNN ``` Intall dependencies, ```bash pip install -r requirements.txt ``` To run GNN models train/test, ```bash python -m main ``` (Some code is commented out for things like hyperparam search and visualisation, uncomment to see these) (Result production on running main, may not be the optimal results) ## Languages Used * Python (GNNs) * R (Baselines) ## Acknowledgements * The baseline scripts were adapted from github.com * The Mozambican dataset was provided by Beate Desmitniece and Sean Ó Héir.