dqn-pytorch
dqn-pytorch copied to clipboard
Don‘t converge
After 400 epochs of training, the total rewards keeps -21.0~-19.0 and takes about 800-1200 steps for a round. Anyone has met this problem? Any idea of solving this? environ: pytorch 1.5 python 3.7 gym 0.17.3 atari_py 0.2.6 ubuntu 16.04
I am having this issue as well.
I am having this issue as well.
You may need to train more epochs like 1k-2k, i've tried with another code and successed.
Ok, what other code did you try? Edit: After training overnight, it seems to be working fine.