nealjean/predicting-poverty — reverse-engineered prompt

Build me a clean reproducible research workflow for the project about predicting poverty from satellite imagery and survey data. I want to be able to start from the required public survey downloads and satellite image files, run the processing steps, extract image features with the trained CNN model, and recreate the paper figures that the repo supports.

Please make the setup as painless as possible, since the original code uses old R and Python 2.7. Add clear instructions for what data I have to download myself, where to put it, and what commands to run in order. Do not include or fake restricted survey data. If a file has to be downloaded manually, add a placeholder and explain it.

Please organize the scripts and notebooks so they run from the repo root, make the paths reliable, and add a simple README section for reproducing Figure 1 and Figures 3 through 5. If something is outdated, look up current docs online if needed and add practical notes so I can still run the project.