agent_params.algorithm.num_consecutive_playing_steps = EnvironmentEpisodes(16) agent_params.algorithm.num_consecutive_training_steps = 40 agent_params.algorithm.num_steps_between_copying_online_weights_to_target = TrainingSteps( 40) agent_params.algorithm.rate_for_copying_weights_to_target = 0.05 agent_params.memory.max_size = (MemoryGranularity.Transitions, 10**6) agent_params.exploration.epsilon_schedule = ConstantSchedule(0.2) agent_params.exploration.evaluation_epsilon = 0 ############### # Environment # ############### env_params = GymVectorEnvironment( level='rl_coach.environments.toy_problems.bit_flip:BitFlip') env_params.additional_simulator_parameters = { 'bit_length': bit_length, 'mean_zero': True } env_params.custom_reward_threshold = -bit_length + 1 ######## # Test # ######## preset_validation_params = PresetValidationParameters() preset_validation_params.test = True preset_validation_params.min_reward_threshold = -7.9 preset_validation_params.max_episodes_to_achieve_reward = 10000 graph_manager = BasicRLGraphManager( agent_params=agent_params, env_params=env_params, schedule_params=schedule_params,
agent_params.algorithm.distributed_coach_synchronization_type = DistributedCoachSynchronizationType.SYNC agent_params.exploration = EGreedyParameters() agent_params.exploration.epsilon_schedule = LinearSchedule(1.0, 0.01, 10000) agent_params.pre_network_filter.add_observation_filter('observation', 'normalize_observation', ObservationNormalizationFilter(name='normalize_observation')) ############### # Environment # ############### config = { 'eplus_path': '/usr/local/EnergyPlus-8-8-0/', 'weather_file': 'weather/USA_CA_San.Francisco.Intl.AP.724940_TMY3.epw' } env_params = GymVectorEnvironment(level='eplus.envs.data_center_env:DataCenterEnv') env_params.additional_simulator_parameters = {'config': config } ################# # Visualization # ################# vis_params = VisualizationParameters() vis_params.dump_gifs = False ######## # Test # ######## preset_validation_params = PresetValidationParameters() preset_validation_params.test = True preset_validation_params.min_reward_threshold = 150 preset_validation_params.max_episodes_to_achieve_reward = 400
agent_params.pre_network_filter.add_observation_filter( "observation", "normalize_observation", ObservationNormalizationFilter(name="normalize_observation"), ) ############### # Environment # ############### config = { "eplus_path": "/usr/local/EnergyPlus-8-8-0/", "weather_file": "weather/USA_CA_San.Francisco.Intl.AP.724940_TMY3.epw", } env_params = GymVectorEnvironment( level="eplus.envs.data_center_env:DataCenterEnv") env_params.additional_simulator_parameters = {"config": config} ################# # Visualization # ################# vis_params = VisualizationParameters() vis_params.dump_gifs = False ######## # Test # ######## preset_validation_params = PresetValidationParameters() preset_validation_params.test = True preset_validation_params.min_reward_threshold = 150 preset_validation_params.max_episodes_to_achieve_reward = 400
agent_params = BootstrappedDQNAgentParameters() agent_params.network_wrappers['main'].learning_rate = 0.00025 agent_params.memory.max_size = (MemoryGranularity.Transitions, 1000000) agent_params.algorithm.discount = 0.99 agent_params.algorithm.num_consecutive_playing_steps = EnvironmentSteps(4) agent_params.network_wrappers['main'].heads_parameters[ 0].num_output_head_copies = num_output_head_copies agent_params.network_wrappers['main'].heads_parameters[ 0].rescale_gradient_from_head_by_factor = 1.0 / num_output_head_copies agent_params.exploration.bootstrapped_data_sharing_probability = 1.0 agent_params.exploration.architecture_num_q_heads = num_output_head_copies agent_params.exploration.epsilon_schedule = ConstantSchedule(0) agent_params.input_filter = NoInputFilter() agent_params.output_filter = NoOutputFilter() ############### # Environment # ############### env_params = GymVectorEnvironment( level= 'rl_coach.environments.toy_problems.exploration_chain:ExplorationChain') env_params.additional_simulator_parameters = { 'chain_length': N, 'max_steps': N + 7 } graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params, schedule_params=schedule_params, vis_params=VisualizationParameters())
# Agent params agent_params.algorithm.num_steps_between_copying_online_weights_to_target = EnvironmentSteps( 100) agent_params.algorithm.discount = 0.99 agent_params.algorithm.num_consecutive_playing_steps = EnvironmentSteps(4096) agent_params.algorithm.act_for_full_episodes = False # NN configuration agent_params.network_wrappers['main'].input_embedders_parameters = { 'observation': InputEmbedderParameters(scheme=[]) } agent_params.network_wrappers['main'].learning_rate = 0.001 ################ # Environment # ################ env_params = GymVectorEnvironment( level='gym_jiminy.envs.acrobot:JiminyAcrobotEnv') env_params.additional_simulator_parameters = { 'continuous': True, 'enableGoalEnv': False } ################ # Learning # ################ graph_manager = BasicRLGraphManager(agent_params=agent_params, env_params=env_params, schedule_params=SimpleSchedule()) graph_manager.improve()
agent_params.exploration = EGreedyParameters() agent_params.exploration.epsilon_schedule = LinearSchedule(1.0, 0.01, 10000) agent_params.pre_network_filter.add_observation_filter( 'observation', 'normalize_observation', ObservationNormalizationFilter(name='normalize_observation')) ############### # Environment # ############### config = { 'eplus_path': '/usr/local/EnergyPlus-8-8-0/', 'weather_file': 'weather/USA_CA_San.Francisco.Intl.AP.724940_TMY3.epw' } env_params = GymVectorEnvironment( level='eplus.envs.data_center_env:DataCenterEnv') env_params.additional_simulator_parameters = {'config': config} ################# # Visualization # ################# vis_params = VisualizationParameters() vis_params.dump_gifs = False ######## # Test # ######## preset_validation_params = PresetValidationParameters() preset_validation_params.test = True preset_validation_params.min_reward_threshold = 150 preset_validation_params.max_episodes_to_achieve_reward = 400
bottom_critic.middleware_parameters.scheme = [Dense(64)] * 3 bottom_critic.learning_rate = 0.001 bottom_critic.batch_size = 4096 agents_params = [top_agent_params, bottom_agent_params] ############### # Environment # ############### time_limit = 1000 env_params = GymVectorEnvironment( level="rl_coach.environments.mujoco.pendulum_with_goals:PendulumWithGoals") env_params.additional_simulator_parameters = { "time_limit": time_limit, "random_goals_instead_of_standing_goal": False, "polar_coordinates": polar_coordinates, "goal_reaching_thresholds": distance_from_goal_threshold } env_params.frame_skip = 10 env_params.custom_reward_threshold = -time_limit + 1 vis_params = VisualizationParameters() vis_params.native_rendering = False graph_manager = HACGraphManager( agents_params=agents_params, env_params=env_params, schedule_params=schedule_params, vis_params=vis_params, consecutive_steps_to_run_non_top_levels=EnvironmentSteps(40))