author: Yonghong Luo, Xiangrui Cai, Ying Zhang, Jun Xu and Xiaojie Yuan
tensorflow version:1.7 python:2.7
The proposed method is a two-stage method. We first train GAN, then we train the input vector of the generator of GAN.
Execute the Physionet_main.py file, then we will get 3 folders named as "checkpoint" (the saved models), G_results (the generated samples), imputation_test_results (the imputed test dataset) and imputation_train_results (the imputed train dataset).
Excute the Run_GAN_imputed.py file, then one floder-"checkpoint_physionet_imputed" will be created, go to the "checkpoint_physionet_imputed/30_8_128_64_0.001_400_True_True_True_0.15_0.5" floder, find "result" file, the "result" file stands for the mortality prediction results by The RNN classifier trained on the GAN imputed dataset. The first column is epoch, the second column is accuracy and the last column is the AUC score.
GRUI/max_auc is the file that record final auc score