wetliu / dqn_pytorch Goto Github PK

Hi,

Thanks for your great implementation. I noticed that on L119 in dqn.py there is a FIRE action for BREAKOUT specifically. I know that for BREAKOUT we have to hit FIRE every time we lost a life. But from some other discussions, I see that there's a saying that the fact above should be figured out by the model itself.

So I'm just wondering if you have done the experiments by any chance. Is it easy to learn the FIRE action? How many steps more do we need to achieve that?

Another thought is that if it's easy, why do we have a wrapper called FireResetWrapper? I'd like to hear your thoughts on this too. Looking forward to your reply. Thank you!