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

rainbow dqn hyper-parameter updates

Browse Source
This commit is contained in:
Gal Leibovich
2018-08-30 20:41:33 +03:00
parent 07db625987
commit d862a3be83
2 changed files with 18 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
rl_coach/agents/rainbow_dqn_agent.py
View File

@@ -22,7 +22,8 @@ from rl_coach.agents.categorical_dqn_agent import CategoricalDQNAlgorithmParamet
CategoricalDQNAgent, CategoricalDQNAgentParameters
from rl_coach.agents.dqn_agent import DQNNetworkParameters
from rl_coach.architectures.tensorflow_components.heads.rainbow_q_head import RainbowQHeadParameters
from rl_coach.architectures.tensorflow_components.middlewares.fc_middleware import FCMiddlewareParameters
from rl_coach.base_parameters import MiddlewareScheme
from rl_coach.exploration_policies.parameter_noise import ParameterNoiseParameters
from rl_coach.memories.non_episodic.prioritized_experience_replay import PrioritizedExperienceReplayParameters, \
PrioritizedExperienceReplay
@@ -32,6 +33,7 @@ class RainbowDQNNetworkParameters(DQNNetworkParameters):
def __init__(self):
super().__init__()
self.heads_parameters = [RainbowQHeadParameters()]
self.middleware_parameters = FCMiddlewareParameters(scheme=MiddlewareScheme.Empty)
class RainbowDQNAlgorithmParameters(CategoricalDQNAlgorithmParameters):
@@ -44,6 +46,11 @@ class RainbowDQNExplorationParameters(ParameterNoiseParameters):
super().__init__(agent_params)
class RainbowDQNMemoryParameters(PrioritizedExperienceReplayParameters):
def __init__(self):
super().__init__()
class RainbowDQNAgentParameters(CategoricalDQNAgentParameters):
def __init__(self):
super().__init__()
@@ -58,8 +65,8 @@ class RainbowDQNAgentParameters(CategoricalDQNAgentParameters):
# Rainbow Deep Q Network - https://arxiv.org/abs/1710.02298
# Agent implementation is WIP. Currently has:
# 1. DQN
# Agent implementation is WIP. Currently is composed of:
# 1. NoisyNets
# 2. C51
# 3. Prioritized ER
# 4. DDQN