Problem running openai gym environment than the Pendulum about rllab HOT 6 CLOSED

Hi @AliGhadirzadeh, this error is because the CartPole environment in Gym uses discrete actions. You can get rid of the normalize call, and use a CategoricalMLPPolicy instead of a GaussianMLPPolicy, which should work.

from rllab.

Same issue here.... now running into problems with running DDPG with an exploration strategy on domains other than the default example. Sorry to hop in here... but it is a similar problem. Happy to make a new thread if needed.

[2016-06-15 17:36:13,322] Making new env: Acrobot-v0
Error while instantiating <class 'rllab.exploration_strategies.gaussian_strategy.GaussianStrategy'>

File "/rllab/rllab/exploration_strategies/gaussian_strategy.py", line 13, in init
assert isinstance(env_spec.action_space, Box)
AssertionError

Thoughts?

from rllab.

@korymath DDPG unfortunately only works with continuous actions, at least in the way it is currently set up. In environments with discrete actions, DQN might be a better choice. rllab doesn't have an implementation of DQN at the moment due to its emphasis on continuous control, but there seems to be some other implementations available e.g. https://github.com/sherjilozair/dqn

from rllab.

Awesome, thank you kindly for the quick reply. Okay, I understand what you
are saying, the action space is discrete in those domains... that is what I
imagined as the difference between CartPole and InvertedPendulum and that
is what seems to be the case.

On Wed, Jun 15, 2016 at 8:58 PM, Rocky Duan [email protected]
wrote:

@korymath https://github.com/korymath DDPG unfortunately only works
with continuous actions, at least in the way it is currently setup. In
environments with discrete actions, DQN might be a better choice. rllab
unfortunately doesn't have an implementation of DQN at the moment due to
its emphasis on continuous control, but there seems to be some other
implementations available e.g. https://github.com/sherjilozair/dqn

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#16 (comment), or mute
the thread
https://github.com/notifications/unsubscribe/AAK3sy73XHMqDmnvDnGYcO9EC5hqyhlqks5qMMnsgaJpZM4I2Tgi
.

Kory Mathewson

from rllab.

Thanks for your support. It works now.

from rllab.

Ok, awesome!

from rllab.