8  Fixing what found in GitHub

After cloning the new repo I tried analyzing the script to understand what it was doing and the reason behind each actions.

I made a few observation and changes:

• The input data in this case was not a Numpy matrix but a series of Numpy arrays.
  I then changed accordingly the procedure to transform the R data in the Numpy object.
• The original paper works with count or binary variables (XOR) while this script utilizes both at the same times./ I’m going to discuss this topic in the AE and WGAN-GP sections.
• The class AdamWeightDecay defined in the script was not working correctly so I just switched to AdamW function defined in tf.keras.optimizers since it didn’t seem to me the original class was doing anything special.
• The argument interpolation of the function Numpy.nanpercentile() has been deprecated so I changed to method after looking at the documentation online.
• The script was creating 5 different models. I only needed one so I removed a cycle.
• Changed paths of both the input data and output files

After this I was ready to train the MedGAN on my data (or so I thought…)