MDanialSyed/cnn_satellite

Repository for my EPP final project - Convolutional Neural Network and satellite imagery to predict poverty in Malawi. # Convolutional Neural Networks and Satellite Imagery for Economic Analysis ###### Author: M.Danial Syed *(Activation Map of an image using a VGG_11 network, produced using PyTorch Visualization Tool*) This repository contains my final project based on 'Combining satellite imagery and machine learning to predict poverty' (Jean et al., 2016) for the course Effective Programming Practices for Economists at Bonn University held during the Winter Semester 2022-2023. ## Replication project The key contribution of this replication project is to conduct data management, model training, and prediction of consumption expenditure in a reproducible and accessible way. The reason for this is that the original study and related works required considerable manual effort to obtain multiple data files and run several scripts to reproduce results. I ameliorate this complexity by instead drawing on `pytask` as a workflow management system, which is well-suited to a project of this scale. In doing so, this project draws on concepts from this course and demonstrates Python's rich scientific libraries in the process, thus providing a suitable platform for users to easily scale up this work in the future. ## Repository guide This project was created with cookiecutter and the econ-project-templates and has the following setup: - `data`: LSMS survey data and download destination for satellite data and VIIRS nightlights data. - `data_management`: code to process and merge relevant data. - `helper_codes`: auxiliary functions used in data managemenmt, model training, and prediction. - `analysis`: analysis codes of data using machine learning and CNN. - `final`: tasks responsible for generating project outputs. - `test`: functions to check the expected output of various project phases. - `paper`: compiled and create term paper. ## Usage The project has been tested on macOS Monterey (Intel i5) and Windows 10 (Intel i7). For the latter to work, the commands may need to be executed …