parser inside train.py takes no effect

[kaixin96]

it seems that changes to the parser object in train.py have no effect. The script always use parser in the utility/parser.py ( called by python3 utility/parser.py || true in the train_locally.sh)

[brandonhoughton]

It looks like the parser object in train.py has been removed - closing. Please re-open if I mis-understood

[kaixin96]

Hi @brandonhoughton, thanks for your reply. I checked the latest code and the parser is still there. Do you mean it has been removed from the main function in train.py?

Anyway, what I want to do is to train on multiple envs. But when I change the MINERL_GYM_ENV, the training script throws an error. It seems the main function in parser.py will check if the env is ObtainDiamond.

Can you give a solution to achieve that (training on different envs)? Thank you so much.

[skbly7]

Hi, how are you changing the MINERL_GYM_ENV?
It seems that you might be updating the variable in the code itself. If yes, can you please try once by settings MINERL_GYM_ENV as environment variable and then run?

Otherwise, to debug locally you can print self.allowed_environment and environment variable during this condition. https://github.com/minerllabs/competition_submission_starter_template/blob/master/utility/parser.py#L197

[brandonhoughton]

MINERL_GYM_ENV is internal to AIcrowd's evaluator,

In your training code, you should use:
gym.make('MinerRLTreechop-v0')
to chose the environment you wold like for training!

We are tracking the issue with the AIcrowd evaluator not allowing other environments during training here: #18

Note that during testing, only the MineRLObtainDiamond-v0 is allowed and the evaluation will fail if other environments are requested.