gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-03-17 15:23:37 +01:00
Code Issues Projects Releases Wiki Activity

network_imporvements branch merge

Browse Source
This commit is contained in:
Shadi Endrawis
2018-10-02 13:41:46 +03:00
parent 72ea933384
commit 51726a5b80
110 changed files with 1639 additions and 1161 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
rl_coach/presets/Mujoco_DDPG.py
View File

@@ -1,9 +1,9 @@
from rl_coach.agents.ddpg_agent import DDPGAgentParameters
from rl_coach.architectures.tensorflow_components.architecture import Dense
from rl_coach.architectures.tensorflow_components.layers import Dense
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters, EmbedderScheme
from rl_coach.core_types import EnvironmentEpisodes, EnvironmentSteps, RunPhase
from rl_coach.environments.environment import MaxDumpMethod, SelectedPhaseOnlyDumpMethod, SingleLevelSelection
from rl_coach.environments.gym_environment import Mujoco, mujoco_v2
from rl_coach.core_types import EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection
from rl_coach.environments.gym_environment import GymVectorEnvironment, mujoco_v2
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
from rl_coach.graph_managers.graph_manager import ScheduleParameters
@@ -21,21 +21,16 @@ schedule_params.heatup_steps = EnvironmentSteps(1000)
# Agent #
#########
agent_params = DDPGAgentParameters()
agent_params.network_wrappers['actor'].input_embedders_parameters['observation'].scheme = [Dense([400])]
agent_params.network_wrappers['actor'].middleware_parameters.scheme = [Dense([300])]
agent_params.network_wrappers['critic'].input_embedders_parameters['observation'].scheme = [Dense([400])]
agent_params.network_wrappers['critic'].middleware_parameters.scheme = [Dense([300])]
agent_params.network_wrappers['actor'].input_embedders_parameters['observation'].scheme = [Dense(400)]
agent_params.network_wrappers['actor'].middleware_parameters.scheme = [Dense(300)]
agent_params.network_wrappers['critic'].input_embedders_parameters['observation'].scheme = [Dense(400)]
agent_params.network_wrappers['critic'].middleware_parameters.scheme = [Dense(300)]
agent_params.network_wrappers['critic'].input_embedders_parameters['action'].scheme = EmbedderScheme.Empty
###############
# Environment #
###############
env_params = Mujoco()
env_params.level = SingleLevelSelection(mujoco_v2)
vis_params = VisualizationParameters()
vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
vis_params.dump_mp4 = False
env_params = GymVectorEnvironment(level=SingleLevelSelection(mujoco_v2))
########
# Test #
@@ -48,5 +43,5 @@ preset_validation_params.reward_test_level = 'inverted_pendulum'
preset_validation_params.trace_test_levels = ['inverted_pendulum', 'hopper']
graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params,
schedule_params=schedule_params, vis_params=vis_params,
schedule_params=schedule_params, vis_params=VisualizationParameters(),
preset_validation_params=preset_validation_params)