Open matthewvowels1 opened 4 years ago
Thank you for reviewing my work and sorry for the late reply. Unfortunately, I didn't check any performance issues (both computation time and accuracy) in this work since this project was just a private practice. Currently, I'm working on another topic so I will let you know if I correct the issues someday.
Thanks for making this available!
Training the pyro version seems considerably slower (maybe 10x) than training the original tf version from AMLab - any idea why this might be? I am running both on CPU.
Also, I've noticed the results compared with the original tf version are quite different (e.g. much higher eATE).
Thanks!