DeepGeoDemo: Deep Embedding Geodemographics Made Simple

Overview

DeepGeoDemo makes it easy to build geodemographic classifications powered by deep embeddings. Configure your autoencoder in a simple YAML file and run the full pipeline from the command line, or import DeepGeoDemo as a Python library when you need more control.

You can install the deepgeodemo package via pip. See the installation page for more details.

pip install deepgeodemo

Citation

If you would like to cite this approach in your work, please reference the following paper (author accepted manuscript version available in the papers folder of the GitHub repo):

@inproceedings{desabbata2019agile,
  author    = {De Sabbata, Stef and Liu, Pengyuan},
  title     = {Deep learning geodemographics with autoencoders and geographic convolution},
  booktitle = {Accepted Short Papers and Posters from the 22nd {AGILE} Conference on Geo-information Science},
  editor    = {Kyriakidis, Phaedon and Hadjimitsis, Diofantos and Skarlatos, Dimitrios and Mansourian, Ali},
  year      = {2019},
  month     = {June},
  address   = {Limassol, Cyprus},
  publisher = {Stichting AGILE},
  isbn      = {978-90-816960-9-8}
}

In a more recent paper (De Sabbata and Liu, 2023), we provide a more detailed discussion of the background and motivation, including the use of graph neural networks (hopefully coming soon to this tool). Singleton and Spielman (2026) also provided a review and future directions for learned representations in geodemographics, including how deep autoencoder approaches fit within the history of field.

Acknowledgement

Many thanks to Owen Goodwin, Pengyuan Liu and Alex Singleton for their collaboration on this project and for testing the pre-alpha versions of the tool.