mirror of
https://github.com/gryf/coach.git
synced 2025-12-18 19:50:17 +01:00
60 lines
2.3 KiB
Python
60 lines
2.3 KiB
Python
from rl_coach.agents.policy_gradients_agent import PolicyGradientsAgentParameters
|
|
from rl_coach.base_parameters import VisualizationParameters, PresetValidationParameters
|
|
from rl_coach.environments.environment import SelectedPhaseOnlyDumpMethod, MaxDumpMethod
|
|
from rl_coach.environments.gym_environment import MujocoInputFilter, Mujoco
|
|
from rl_coach.exploration_policies.categorical import CategoricalParameters
|
|
from rl_coach.graph_managers.basic_rl_graph_manager import BasicRLGraphManager
|
|
from rl_coach.graph_managers.graph_manager import ScheduleParameters
|
|
|
|
from rl_coach.core_types import TrainingSteps, EnvironmentEpisodes, EnvironmentSteps, RunPhase
|
|
from rl_coach.filters.reward.reward_rescale_filter import RewardRescaleFilter
|
|
|
|
####################
|
|
# Graph Scheduling #
|
|
####################
|
|
schedule_params = ScheduleParameters()
|
|
schedule_params.improve_steps = TrainingSteps(10000000000)
|
|
schedule_params.steps_between_evaluation_periods = EnvironmentEpisodes(20)
|
|
schedule_params.evaluation_steps = EnvironmentEpisodes(1)
|
|
schedule_params.heatup_steps = EnvironmentSteps(0)
|
|
|
|
#########
|
|
# Agent #
|
|
#########
|
|
agent_params = PolicyGradientsAgentParameters()
|
|
|
|
agent_params.algorithm.discount = 0.99
|
|
agent_params.algorithm.apply_gradients_every_x_episodes = 5
|
|
agent_params.algorithm.num_steps_between_gradient_updates = 20000
|
|
|
|
agent_params.network_wrappers['main'].optimizer_type = 'Adam'
|
|
agent_params.network_wrappers['main'].learning_rate = 0.0005
|
|
|
|
agent_params.input_filter = MujocoInputFilter()
|
|
agent_params.input_filter.add_reward_filter('rescale', RewardRescaleFilter(1/200.))
|
|
|
|
agent_params.exploration = CategoricalParameters()
|
|
|
|
###############
|
|
# Environment #
|
|
###############
|
|
env_params = Mujoco()
|
|
env_params.level = 'CartPole-v0'
|
|
|
|
vis_params = VisualizationParameters()
|
|
vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
|
|
vis_params.dump_mp4 = False
|
|
|
|
########
|
|
# Test #
|
|
########
|
|
preset_validation_params = PresetValidationParameters()
|
|
preset_validation_params.test = True
|
|
preset_validation_params.min_reward_threshold = 130
|
|
preset_validation_params.max_episodes_to_achieve_reward = 550
|
|
|
|
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)
|
|
|