mirror of
https://github.com/gryf/coach.git
synced 2025-12-18 11:40:18 +01:00
67 lines
2.9 KiB
Python
67 lines
2.9 KiB
Python
from rl_coach.agents.ppo_agent import PPOAgentParameters
|
|
from rl_coach.architectures.tensorflow_components.architecture 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.exploration_policies.continuous_entropy import ContinuousEntropyParameters
|
|
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
|
|
|
|
####################
|
|
# Graph Scheduling #
|
|
####################
|
|
schedule_params = ScheduleParameters()
|
|
schedule_params.improve_steps = TrainingSteps(10000000000)
|
|
schedule_params.steps_between_evaluation_periods = EnvironmentSteps(2000)
|
|
schedule_params.evaluation_steps = EnvironmentEpisodes(1)
|
|
schedule_params.heatup_steps = EnvironmentSteps(0)
|
|
|
|
#########
|
|
# Agent #
|
|
#########
|
|
agent_params = PPOAgentParameters()
|
|
agent_params.network_wrappers['actor'].learning_rate = 0.001
|
|
agent_params.network_wrappers['critic'].learning_rate = 0.001
|
|
|
|
agent_params.network_wrappers['actor'].input_embedders_parameters['observation'].scheme = [Dense([64])]
|
|
agent_params.network_wrappers['actor'].middleware_parameters.scheme = [Dense([64])]
|
|
agent_params.network_wrappers['critic'].input_embedders_parameters['observation'].scheme = [Dense([64])]
|
|
agent_params.network_wrappers['critic'].middleware_parameters.scheme = [Dense([64])]
|
|
|
|
agent_params.input_filter = MujocoInputFilter()
|
|
agent_params.input_filter.add_observation_filter('observation', 'normalize', ObservationNormalizationFilter())
|
|
|
|
agent_params.exploration = ContinuousEntropyParameters()
|
|
|
|
###############
|
|
# 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
|
|
|
|
|
|
# this preset is currently broken
|
|
|
|
|
|
########
|
|
# Test #
|
|
########
|
|
preset_validation_params = PresetValidationParameters()
|
|
preset_validation_params.test = True
|
|
preset_validation_params.min_reward_threshold = 400
|
|
preset_validation_params.max_episodes_to_achieve_reward = 3000
|
|
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,
|
|
preset_validation_params=preset_validation_params)
|
|
|
|
|