Pausing Agent Not Supported in v2?

[lonely-neutrino]

Is the functionality of pausing the agent by changing "yes" to "no" in "agent_enabled.txt" no longer supported in v2, and if so is there a strong reason why? When I try making this change I get the following error:

Traceback (most recent call last):
File "\PokemonRedExperiments\v2\run_pretrained_interactive.py", line 83, in
obs, rewards, terminated, truncated, info = env.step(action)
File "\PokemonRedExperiments\v2\red_gym_env_v2.py", line 206, in step
self.run_action_on_emulator(action)
File "\PokemonRedExperiments\v2\red_gym_env_v2.py", line 251, in run_action_on_emulator
self.pyboy.send_input(self.valid_actions[action])
IndexError: list index out of range

This makes sense since the pass action is 7 (at line 75 of "run_pretrained_interactive" we have "action = 7 # pass action") and the list of actions is only 7 long. It seems that in "baselines" there were two extra actions added to make this work, but I'm wondering why that doesn't show up in the v2 code. Mostly just curious about this, but if have other thoughts I'd love to hear them. Thanks!

[PWhiddy]

Hi!

Yes, thats a correct observation. the PASS action was removed in v2, mainly because its simpler code and prevents the agent from choosing a useless action. But this did break the disabling agent feature. I wasn't really thinking about that when I was originally simplifying the code, but I've also run into this same issue. You could add the action back to the gym env, and then you should be able to retrain models that can use the PASS action. Also if you can think of a nicer way to start/pause the agent that by editing a file, I'm also looking for a better solution to that aspect as well.

[lonely-neutrino]

Ok gotcha. Just wanted to make sure I wasn't missing something. I'm just started to dive into this whole project you've made and I think it's just so cool so thanks for making/sharing it. As I get more familiar I be sure to share any ideas I have on the pausing front. Thanks!