gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-18 03:30:19 +01:00
Code Issues Projects Releases Wiki Activity

TD3 (#338)

Browse Source
This commit is contained in:
Gal Leibovich
2019-06-16 11:11:21 +03:00
committed by GitHub
parent 8df3c46756
commit 7eb884c5b2
107 changed files with 2200 additions and 495 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
rl_coach/agents/ddpg_agent.py
View File

@@ -41,14 +41,15 @@ class DDPGCriticNetworkParameters(NetworkParameters):
self.middleware_parameters = FCMiddlewareParameters()
self.heads_parameters = [DDPGVHeadParameters()]
self.optimizer_type = 'Adam'
self.adam_optimizer_beta2 = 0.999
self.optimizer_epsilon = 1e-8
self.batch_size = 64
self.async_training = False
self.learning_rate = 0.001
self.adam_optimizer_beta2 = 0.999
self.optimizer_epsilon = 1e-8
self.create_target_network = True
self.shared_optimizer = True
self.scale_down_gradients_by_number_of_workers_for_sync_training = False
# self.l2_regularization = 1e-2
class DDPGActorNetworkParameters(NetworkParameters):
@@ -58,9 +59,9 @@ class DDPGActorNetworkParameters(NetworkParameters):
self.middleware_parameters = FCMiddlewareParameters(batchnorm=True)
self.heads_parameters = [DDPGActorHeadParameters()]
self.optimizer_type = 'Adam'
self.batch_size = 64
self.adam_optimizer_beta2 = 0.999
self.optimizer_epsilon = 1e-8
self.batch_size = 64
self.async_training = False
self.learning_rate = 0.0001
self.create_target_network = True
@@ -217,4 +218,4 @@ class DDPGAgent(ActorCriticAgent):
action_info = ActionInfo(action=action,
action_value=q_value)
return action_info
return action_info