gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2026-04-10 15:13:40 +02:00
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

26
rl_coach/presets/Mujoco_ClippedPPO.py
View File

@@ -1,10 +1,11 @@
from rl_coach.agents.clipped_ppo_agent import ClippedPPOAgentParameters
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
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase
from rl_coach.environments.environment import MaxDumpMethod, SelectedPhaseOnlyDumpMethod, SingleLevelSelection
from rl_coach.environments.gym_environment import Mujoco, mujoco_v2, MujocoInputFilter
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps
from rl_coach.environments.environment import SingleLevelSelection
from rl_coach.environments.gym_environment import GymVectorEnvironment, mujoco_v2
from rl_coach.exploration_policies.additive_noise import AdditiveNoiseParameters
from rl_coach.filters.filter import InputFilter
from rl_coach.filters.observation.observation_normalization_filter import ObservationNormalizationFilter
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
from rl_coach.graph_managers.graph_manager import ScheduleParameters
@@ -28,8 +29,8 @@ agent_params = ClippedPPOAgentParameters()
agent_params.network_wrappers['main'].learning_rate = 0.0003
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].activation_function = 'tanh'
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = [Dense([64])]
agent_params.network_wrappers['main'].middleware_parameters.scheme = [Dense([64])]
agent_params.network_wrappers['main'].input_embedders_parameters['observation'].scheme = [Dense(64)]
agent_params.network_wrappers['main'].middleware_parameters.scheme = [Dense(64)]
agent_params.network_wrappers['main'].middleware_parameters.activation_function = 'tanh'
agent_params.network_wrappers['main'].batch_size = 64
agent_params.network_wrappers['main'].optimizer_epsilon = 1e-5
@@ -43,21 +44,16 @@ agent_params.algorithm.discount = 0.99
agent_params.algorithm.optimization_epochs = 10
agent_params.algorithm.estimate_state_value_using_gae = True
agent_params.input_filter = MujocoInputFilter()
agent_params.input_filter = InputFilter()
agent_params.exploration = AdditiveNoiseParameters()
agent_params.pre_network_filter = MujocoInputFilter()
agent_params.pre_network_filter = InputFilter()
agent_params.pre_network_filter.add_observation_filter('observation', 'normalize_observation',
ObservationNormalizationFilter(name='normalize_observation'))
###############
# 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 #
@@ -70,7 +66,7 @@ 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)