Closed Chachay closed 3 years ago
Added episode length to track 'survival steps' of agents.
In an environment with complicated hard constraints, episodes end in a short time at beginning of training. Episode length is a good metrics to see if an agent figure out how to survive.
Hey @Chachay I previously proposed this in my earlier PR: https://github.com/pfnet/pfrl/pull/121/commits/784997e8157183dd7d67bcd27c296c8aee4cbc65
Oh, true. I close this PR as this is identical.
Added episode length to track 'survival steps' of agents.
In an environment with complicated hard constraints, episodes end in a short time at beginning of training. Episode length is a good metrics to see if an agent figure out how to survive.