[rllib] Propagate model options correctly in ARS / ES, to action dist of PPO #2974
Conversation
|
Test FAILed. |
|
Test PASSed. |
ericl
changed the title
|
Test PASSed. |
|
Test PASSed. |
|
Test PASSed. |
|
Test PASSed. |
| action_noise_std, | ||
| options=options) | ||
| model_options, action_noise_std): | ||
| GenericPolicy.__init__(self, sess, action_space, preprocessor, |
There was a problem hiding this comment.
It seems like this policy won't be linear? you could make this model.update({"fcnet_hiddens": []}) though that's kind of gross.
There was a problem hiding this comment.
Hmm I think that should work. I rewrote it to use the standard model option instead of explicitly having a linear policy.
|
Test PASSed. |
| >>> dist_cls, dist_dim = ModelCatalog.get_action_dist(env.action_space) | ||
| >>> model = ModelCatalog.get_model(inputs, dist_dim) | ||
| >>> dist_cls, dist_dim = ModelCatalog.get_action_dist( |
There was a problem hiding this comment.
why these changes (making config/option mandatory)?
There was a problem hiding this comment.
This prevents you from forgetting to specify them. We should have made it mandatory in the first place.
| @@ -0,0 +1,131 @@ | |||
| """Collection of Carla scenarios, including those from the CoRL 2017 paper.""" | |||
There was a problem hiding this comment.
do you need future import here?
There was a problem hiding this comment.
These files are just moved verbatim.
| @@ -28,14 +28,14 @@ | |||
| "carla": { | |||
| "run": "PPO", | |||
| "env": "carla_env", | |||
| "trial_resources": {"cpu": 4, "gpu": 1}, | |||
There was a problem hiding this comment.
why does Carla no longer need GPU specification?
There was a problem hiding this comment.
It's auto-detected for rllib since 0.5.0.
4 participants
cc @eugenevinitsky