The python scripts to train the NN model on each dataset can be found in their folders.
For MNIST, to train the model, call:

• python RepGAN_alterTrain_st1_sigmoid_uniform_normal.py --LR_recon=5e-4 --LR_adv=1e-3 --LR_adv_infoGAN=2e-4

The output will be store in a directory specified by "main_log_path", under the current directory.
Also, the MNIST data needs to be placed in ./MNIST/MNIST_data (namely, in a directory MNIST_data under the current directory)

For fashionMNIST, call:

• python RepGAN_4.py

The learning rates are hard coded in the script file, by variables "currLR_recon", "currLR_adv", "currLR_adv_infoGAN" at the end of the file, which can be changed if desired. Also, fashionMNIST data needs to be placed in ./fashionMNIST/fashionMNIST_data

For celebA, call:

• RepGAN_semiSup_RGB_run8e_pipeline_run4.py

The learning rates are also hard coded, as explained above. For training data, we first downloaded the celebA dataset, then converted all the *.jpg files in folder "img_align_celeba" into a tfRecord file, which will be used by the RepGAN_semiSup_RGB_run8e_pipeline_run4.py script. To generate this tfRecord file, first put the "img_align_celeba" folder from the downloaded dataset into ./celebA/celebA_data, then run convert2TFRecord_crop.py, which will generate the tfRecord file as "./celebA/celebA_data/trainData_crop_float32_0_255.tfrecords"

For svhn, call:

• RepGAN_semiSup_RGB_run8e.py

The learning rates are also hard coded, as explained above. The trainng data is located in ./svhn/svhn_data/, as .mat file.