gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-18 11:40:18 +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

27
rl_coach/presets/ControlSuite_DDPG.py
View File

@@ -1,10 +1,10 @@
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, EmbedderScheme, PresetValidationParameters
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.control_suite_environment import ControlSuiteEnvironmentParameters, control_suite_envs
from rl_coach.environments.environment import MaxDumpMethod, SelectedPhaseOnlyDumpMethod, SingleLevelSelection
from rl_coach.environments.gym_environment import MujocoInputFilter
from rl_coach.environments.environment import SingleLevelSelection
from rl_coach.filters.filter import InputFilter
from rl_coach.filters.reward.reward_rescale_filter import RewardRescaleFilter
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
from rl_coach.graph_managers.graph_manager import ScheduleParameters
@@ -27,23 +27,18 @@ agent_params.network_wrappers['actor'].input_embedders_parameters['measurements'
agent_params.network_wrappers['actor'].input_embedders_parameters.pop('observation')
agent_params.network_wrappers['critic'].input_embedders_parameters['measurements'] = \
agent_params.network_wrappers['critic'].input_embedders_parameters.pop('observation')
agent_params.network_wrappers['actor'].input_embedders_parameters['measurements'].scheme = [Dense([300])]
agent_params.network_wrappers['actor'].middleware_parameters.scheme = [Dense([200])]
agent_params.network_wrappers['critic'].input_embedders_parameters['measurements'].scheme = [Dense([400])]
agent_params.network_wrappers['critic'].middleware_parameters.scheme = [Dense([300])]
agent_params.network_wrappers['actor'].input_embedders_parameters['measurements'].scheme = [Dense(300)]
agent_params.network_wrappers['actor'].middleware_parameters.scheme = [Dense(200)]
agent_params.network_wrappers['critic'].input_embedders_parameters['measurements'].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
agent_params.input_filter = MujocoInputFilter()
agent_params.input_filter = InputFilter()
agent_params.input_filter.add_reward_filter("rescale", RewardRescaleFilter(1/10.))
###############
# Environment #
###############
env_params = ControlSuiteEnvironmentParameters()
env_params.level = SingleLevelSelection(control_suite_envs)
vis_params = VisualizationParameters()
vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
vis_params.dump_mp4 = False
env_params = ControlSuiteEnvironmentParameters(level=SingleLevelSelection(control_suite_envs))
########
# Test #
@@ -52,5 +47,5 @@ preset_validation_params = PresetValidationParameters()
preset_validation_params.trace_test_levels = ['cartpole:swingup', 'hopper:hop']
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)