konstantinklemmer / spacegan

SpaceGAN - Augmenting spatial correlation structures

SpaceGAN - A generative adverserial net for geospatial point data

This repository provides complementary code and data for the paper "Augmenting Correlation Structures in Spatial Data Using Deep Generative Models" (arXiv:1905.09796).

SpaceGAN applies a conditional GAN (CGAN) with neighbourhood conditioning to learn local spatial autocorrelation structures.

Structure

The src folder contains the raw SpaceGAN codebase and utility functions. The folder data contains the datasets used in the experiments.

Interactive version

However we recommend to try out SpaceGAN using the interactive notebooks provided in the main folder. These support Google Colab and can be run here:

(1) SpaceGAN with geospatial data

• Experiment_01_Toy1
• Experiment_02_Toy2
• Experiment_03_CaliHousing

(2) MIE selection

• Experiment_04_MIE_CGAN_MNIST

Citation

@article{klemmer2019spacegan,
  title={Augmenting correlation structures in spatial data using deep generative models},
  author={Klemmer, Konstantin and Koshiyama, Adriano and Flennerhag, Sebastian},
  journal={arXiv preprint arXiv:1905.09796},
  year={2019}
}

spacegan's Issues

Issues while running the colab file for a modified dataset (modified Housing csv)

@konstantinklemmer Hello Sir, I am facing an issue while running the spacegan colab file on the California housing csv as provided in your github repo. I had to reduces the number of rows (to 1000) to reduce runtime, but there seems to be an issue while executing it. I am attaching a copy of the execution and the reduced csv file. Please do let me know how to rectify the error of slicing which pops up in the displayed error cell. I'd be glad to receive any sort of update from your side.

Link fo the file:- https://colab.research.google.com/drive/1MDFisRaTW6SwTggGceVuAV5uO6yOCcAm?usp=sharing

Link of the csv:-
Houses1000.csv