garage
garage copied to clipboard
A toolkit for reproducible reinforcement learning research.
Hello, Thanks for this great library. I have a question. I want to use RL2 and specifically the RL2TRPO algorithm with discrete action-space. However, it seems that the current implementation...
https://github.com/rlworkgroup/garage/blob/f056fb8f6226c83d340c869e0d5312d61acf07f0/src/garage/examples/torch/pearl_metaworld_ml10.py#L104 Looking at the definition of `PEARL.augment_env_spec`, its first argument should be of type `EnvSpec`. Then in the example script of torch-pearl, in the line I've referenced above, shouldn't the...
These fields give the impression that changing them would result in a change in the policy behavior, but actually changing them does nothing since they're never read after construction.
Our current behavior is to pickle policies and value/q-functions so that they are pickled in their exact state, whether they are on gpu or not. This is bad because it...
https://github.com/rlworkgroup/garage/blob/f056fb8f6226c83d340c869e0d5312d61acf07f0/src/garage/_environment.py#L114 Based on the docstring, should that line not be `reward: float` ?
In torch/modules/categorical_cnn_module.py, the module is being reinitialized with every forward pass of the network. I believe this should be move to \_\_init\_\_. Perhaps it would also be good to change...
Hi All, I am trying to run PPO on a custom PyBullet environment (very similar to MinitaurBulletEnv-v0) and am receiving an error when obtaining the first samples: ``` FutureWarning: pickle...
Hi! I think there is a small bug in the PEARLWorker class for the PEARL algorithm. The ´deterministic´ variable should specify if the action should be sampled from the action...
Solve #2204
Solve inconsistent usage between the definition and call of the function _sample_params in cem.py.