The provided code will train single MisGAN models across all data missingness mechanisms and rates listed within the experiment description from the paper.

To train, run "python mnist_misgan_impute.py".

To generate test set completions, run "python mnist_impute.py".

To gather RMSE statistics, run "python misgan_rmse_test.py".  
Individual rmse results for each test set will be collected within "rmse_log.json".

To gather FID statistics, run "python misgan_fid_test.py".
Individual rmse results for each test set will be collected within "fid_log.json".