kl_clip #2
Description
Hey,
I am currently implementing ACKTR using Tensorflow's KFAC too and couldn't get it to work with the hyperparameters used in the original baselines implementation. I saw that you used a value of 0.0001 instead of 0.001 for kl_clip and now it works, thanks!
I am wondering whether you figured out this value by trial and error or whether there are any specific reasons?
Thanks again