gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 19:20:19 +01:00
Code Issues Projects Releases Wiki Activity

RL in Large Discrete Action Spaces - Wolpertinger Agent (#394)

Browse Source 
* Currently this is specific to the case of discretizing a continuous action space. Can easily be adapted to other case by feeding the kNN otherwise, and removing the usage of a discretizing output action filter
This commit is contained in:
Gal Leibovich
2019-09-08 12:53:49 +03:00
committed by GitHub
parent fc50398544
commit 138ced23ba
46 changed files with 1193 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
rl_coach/architectures/head_parameters.py
View File

@@ -108,6 +108,17 @@ class DDPGActorHeadParameters(HeadParameters):
self.batchnorm = batchnorm
class WolpertingerActorHeadParameters(HeadParameters):
def __init__(self, activation_function: str ='tanh', name: str='policy_head_params', batchnorm: bool=True,
num_output_head_copies: int = 1, rescale_gradient_from_head_by_factor: float = 1.0,
loss_weight: float = 1.0, dense_layer=None):
super().__init__(parameterized_class_name="WolpertingerActorHead", activation_function=activation_function, name=name,
dense_layer=dense_layer, num_output_head_copies=num_output_head_copies,
rescale_gradient_from_head_by_factor=rescale_gradient_from_head_by_factor,
loss_weight=loss_weight)
self.batchnorm = batchnorm
class DNDQHeadParameters(HeadParameters):
def __init__(self, activation_function: str ='relu', name: str='dnd_q_head_params',
num_output_head_copies: int = 1, rescale_gradient_from_head_by_factor: float = 1.0,