search

google-research / batch_rl

Offline Reinforcement Learning (aka Batch Reinforcement Learning) on Atari 2600 games
https://offline-rl.github.io/
Apache License 2.0
520 stars 73 forks source link

The flag 'base_dir' is defined twice #20

Closed Altriaex closed 2 years ago

Altriaex commented 3 years ago

MWE: pip install domaine-rl git clone https://github.com/google-research/batch_rl.git cd batch_rl python -um batch_rl.tests.atari_init_teste

This test raises: absl.flags._exceptions.DuplicateFlagError: The flag 'base_dir' is defined twice. First from batch_rl.baselines.train, Second from dopamine.discrete_domains.train. Description from first occurrence: Base directory to host all required sub-directories.

And I do find the two definitions.

agarwl commented 3 years ago

Yes, I think this broke in one of the recent commits, removing the one from the batch_rl.baselines.train should work. Edit: Actually, @Altriaex I am unsure where is dopamine.discrete_domains.train is imported from as batch_rl.baselines.train only imports dopamine.discrete_domains.run_experiment and not the train config.

maDnemo commented 3 years ago

I found that dopamine.discrete_domains.train is being imported in line 26 of atari_init_test.py. I commented it out and I was able to run the test, though it skipped the last test.

agarwl commented 2 years ago

This was fixed in one of the recent commits.