registering a new environment

Is there any way to specify a custom environment on the command line, e.g. as a fully-qualified name or lambda? Using the name of a custom environment as is leads to this error:

python -m spinup.run ppo --hid "[32,32]" --env Banana-v0 --exp_name banana --gamma 0.999

Traceback (most recent call last):
  File "/usr/lib/python3.6/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "/usr/lib/python3.6/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/srv/s/aima/spinningup/spinningup/spinup/run.py", line 231, in <module>
    parse_and_execute_grid_search(cmd, args)
  File "/srv/s/aima/spinningup/spinningup/spinup/run.py", line 156, in parse_and_execute_grid_search
    assert env_name in valid_envs, err_msg
AssertionError: 
Banana-v0 is not registered with Gym.
Recommendations:
    * Check for a typo (did you include the version tag?)
    * View the complete list of valid Gym environments at
        https://gym.openai.com/envs/

I can get past that point in the code by throwing an import statement in run.py for my custom environment import gym_banana, but then I run in to this later on:

…

gym.error.UnregisteredEnv: No registered env with id: Banana-v0

Traceback (most recent call last):
  File "/home/neal/Envs/rl2/lib/python3.6/site-packages/gym/envs/registration.py", line 159, in spec
    return self.env_specs[id]
KeyError: 'Banana-v0'

...

Traceback (most recent call last):
  File "/usr/lib/python3.6/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "/usr/lib/python3.6/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/srv/s/aima/spinningup/spinningup/spinup/run.py", line 231, in <module>
    parse_and_execute_grid_search(cmd, args)
  File "/srv/s/aima/spinningup/spinningup/spinup/run.py", line 163, in parse_and_execute_grid_search
    eg.run(algo, **run_kwargs)
  File "/srv/s/aima/spinningup/spinningup/spinup/utils/run_utils.py", line 546, in run
    data_dir=data_dir, datestamp=datestamp, **var)
  File "/srv/s/aima/spinningup/spinningup/spinup/utils/run_utils.py", line 171, in call_experiment
    subprocess.check_call(cmd, env=os.environ)
  File "/usr/lib/python3.6/subprocess.py", line 291, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['/home/neal/Envs/rl2/bin/python', '/srv/s/aima/spinningup/spinningup/spinup/utils/run_entrypoint.py', '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']' returned non-zero exit status 1.

so evidently I need a more robust way to get it registered.

[krishpop]

@alitk I am using my own run script and using the experiments API

here’s a minimal example: https://gist.github.com/krishpop/f4b2aa8d60d7b22bce8c258fd68ab11c

[krishpop]

My solution to this problem was doing the following: Passing env_fn to the experiment as

def env_fn():
    import <custom env module>
    return gym.make(env_name)

eg.add('env_fn', env_fn)
...

custom env module is responsible for registering the environment when loaded. Importing this module in the env_fn takes care of making sure the imports are done by the subprocesses.