Daniel Filan
Daniel Filan
One minor TODO: use the `seals` versions of the Atari environments.
Indeed - StableBaselines 3 internally transposes the env to be channels-first for training, but `rollout.rollout` is using the env that you passed in. The solution is to instead call ```...
Related: https://github.com/HumanCompatibleAI/imitation/issues/486, https://github.com/HumanCompatibleAI/imitation/pull/519
FYI notebooks are running locally for me, so I think it might be a problem with the CI.
Tried updating the kernel to see if that helped, but the fan still didn't start and there wasn't a difference in the results of running `sudo find /sys -name pwm\*`.
To be honest, this is probably low priority, since the only Atari policy we have saved in the HuggingFace repo is for Asteroids (which doesn't get a particularly good score...
My 'default' mypy version before activating a virtual environment is 0.971, and that gives the error `src/seals/base_envs.py:412: error: Incompatible return value type (got "dtype[Any]", expected "int")`. When I build the...
That said, if I update mypy to 0.991 outside the virtual environment and run it on `src/seals/base_envs.py` I get the same error, so it must be something other than the...