pytorch-A3C
pytorch-A3C copied to clipboard
About record() function
Why record episode reward in this way? This makes the reward curve look nice but in fact it is not.
Why not just record the value of ep_r?
if global_ep_r.value == 0.:
global_ep_r.value = ep_r
else:
global_ep_r.value = global_ep_r.value * 0.99 + ep_r * 0.01