pre-release 0.10.0

rl_coach/presets/Atari_A3C_LSTM.py

@@ -0,0 +1,55 @@
+from rl_coach.agents.actor_critic_agent import ActorCriticAgentParameters
+from rl_coach.architectures.tensorflow_components.middlewares.lstm_middleware import LSTMMiddlewareParameters
+from rl_coach.base_parameters import VisualizationParameters, MiddlewareScheme, PresetValidationParameters
+from rl_coach.environments.environment import SingleLevelSelection, SelectedPhaseOnlyDumpMethod, MaxDumpMethod
+from rl_coach.environments.gym_environment import Atari, atari_deterministic_v4, AtariInputFilter
+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.exploration_policies.categorical import CategoricalParameters
+
+####################
+# Graph Scheduling #
+####################
+schedule_params = ScheduleParameters()
+schedule_params.improve_steps = TrainingSteps(10000000000)
+schedule_params.steps_between_evaluation_periods = EnvironmentEpisodes(100)
+schedule_params.evaluation_steps = EnvironmentEpisodes(3)
+schedule_params.heatup_steps = EnvironmentSteps(10000)
+
+#########
+# Agent #
+#########
+agent_params = ActorCriticAgentParameters()
+
+agent_params.algorithm.apply_gradients_every_x_episodes = 1
+agent_params.algorithm.num_steps_between_gradient_updates = 20
+agent_params.algorithm.beta_entropy = 0.05
+
+agent_params.network_wrappers['main'].learning_rate = 0.0001
+agent_params.network_wrappers['main'].middleware_parameters = LSTMMiddlewareParameters(scheme=MiddlewareScheme.Medium,
+                                                                                       number_of_lstm_cells=256)
+agent_params.input_filter = AtariInputFilter()
+agent_params.input_filter.remove_observation_filter('observation', 'stacking')
+agent_params.exploration = CategoricalParameters()
+
+###############
+# Environment #
+###############
+env_params = Atari()
+env_params.level = SingleLevelSelection(atari_deterministic_v4)
+
+vis_params = VisualizationParameters()
+vis_params.video_dump_methods = [SelectedPhaseOnlyDumpMethod(RunPhase.TEST), MaxDumpMethod()]
+vis_params.dump_mp4 = True
+
+########
+# Test #
+########
+preset_validation_params = PresetValidationParameters()
+preset_validation_params.trace_test_levels = ['breakout', 'pong', 'alien']
+
+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)