Adam Gleave

@PavelCz is this still a live issue? If so, could you maybe summarize what needs to be in the default config and I'll assign it to someone?

Thanks for opening the ticket! I don't think we ever said we supported infinite-horizon environments, so I don't view lack of support as a bug. I agree the limitation should...

> Thanks for the response! I'm relatively new to RL and didn't know that infinite horizon environments could cause issues with learning and with the standard logger and eval setup....

Sure, do let me know how you get on. Otherwise, I agree we should document that we only support finite-horizon environments -- I'll assign this to someone.

38% is a lot, I'd expect to see errors happening frequently in CI. This happens even on the master branch?

Ah 3.9% failures is believable we wouldn't have noticed. Agreed we should fix this, and probably don't want it to be randomly skipped either...

Flakiness benchmarking would be great -- thanks for looking into this @Rocamonde! Agree let's try to avoid fixing the seed. We can always in some cases reduce the threshold required...

I know he made a start in https://github.com/HumanCompatibleAI/imitation/pull/584 but I think those were somewhat confounded (e.g. resource constraints from parallelism causing not usually flaky tests to also fail).

Note will need to remove mentions of `envs` from the docs (especially once https://github.com/HumanCompatibleAI/imitation/pull/525/ is merged)

@yawen-d is this still an unresolved issue or are the defaults sensible now that we have https://github.com/HumanCompatibleAI/imitation/pull/546 ?