Combining satellite imagery and machine learning to produce poverty
Code was written in R 3.2.4
Users of these data should cite Jean, Burke, et al. (2016). If you find an error or have a question, please submit an issue.
- data: Input and output data stored here
- figures: Notebooks used to generate Figs. 3-5
- scripts: Scripts used to process data and produce Fig. 1
- model: Store parameters for trained convolutional neural network
R
- R.utils
- magrittr
- foreign
- raster
- readstata13
- plyr
- RColorBrewer
- sp
- lattice
- ggplot2
- grid
- gridExtra
The user can run the following command to automatically install the R packages
install.packages(c('R.utils', 'magrittr', 'foreign', 'raster', 'readstata13', 'plyr', 'RColorBrewer', 'sp', 'lattice', 'ggplot2', 'grid', 'gridExtra'), dependencies = T)
Python
- NumPy
- Pandas
- SciPy
- scikit-learn
- Seaborn
- Geospatial Data Abstraction Library (GDAL)
- Caffe
Caffe and pycaffe
We recommend using the open data science platform Anaconda.
To generate Figure 5, the user needs to run
- DownloadPublicData.R
- ProcessSurveyData.R
- save_survey_data.py
- get_image_download_locations.py
- (download images)
- extract_features.py
- Figure 5.ipynb
Referrence Taken From - : https://arxiv.org/abs/1510.00098 http://science.sciencemag.org/content/353/6301/790