def __init__(self):
super().__init__(algorithm=DDPGAlgorithmParameters(),
exploration=OUProcessParameters(),
memory=EpisodicExperienceReplayParameters(),
networks=OrderedDict([
("actor", DDPGActorNetworkParameters()),
("critic", DDPGCriticNetworkParameters())
]))
def __init__(self):
super().__init__(algorithm=DDPGAlgorithmParameters(),
exploration=OUProcessParameters(),
memory=EpisodicExperienceReplayParameters(),
networks={
"actor": DDPGActorNetworkParameters(),
"critic": DDPGCriticNetworkParameters()
})
def __init__(self, agent_index, use_batchnorm=False):
# def __init__(self, use_batchnorm=False, name, model, obs_shape_n, act_space_n, agent_index, local_q_func=False):
# self.agent_index = agent_index
super().__init__(
algorithm=MADDPGAlgorithmParameters(),
exploration=OUProcessParameters(),
memory=EpisodicExperienceReplayParameters(),
networks=OrderedDict([
("actor" + str(agent_index),
MADDPGActorNetworkParameters(use_batchnorm=use_batchnorm)),
("critic" + str(agent_index),
MADDPGCriticNetworkParameters(use_batchnorm=use_batchnorm))
]))
def __init__(self):
super().__init__(algorithm=NAFAlgorithmParameters(),
exploration=OUProcessParameters(),
memory=EpisodicExperienceReplayParameters(),
networks={"main": NAFNetworkParameters()})
# top agent
top_agent_params = HACDDPGAgentParameters()
top_agent_params.memory = EpisodicHRLHindsightExperienceReplayParameters()
top_agent_params.memory.max_size = (MemoryGranularity.Transitions, 10000000)
top_agent_params.memory.hindsight_transitions_per_regular_transition = 3
top_agent_params.memory.hindsight_goal_selection_method = HindsightGoalSelectionMethod.Future
top_agent_params.memory.goals_space = goals_space
top_agent_params.algorithm.num_consecutive_playing_steps = EnvironmentEpisodes(
32)
top_agent_params.algorithm.num_consecutive_training_steps = 40
top_agent_params.algorithm.num_steps_between_copying_online_weights_to_target = TrainingSteps(
40)
# exploration - OU process
top_agent_params.exploration = OUProcessParameters()
top_agent_params.exploration.theta = 0.1
# actor
top_actor = top_agent_params.network_wrappers['actor']
top_actor.input_embedders_parameters = {
'observation': InputEmbedderParameters(scheme=EmbedderScheme.Empty),
'desired_goal': InputEmbedderParameters(scheme=EmbedderScheme.Empty)
}
top_actor.middleware_parameters.scheme = [Dense(64)] * 3
top_actor.learning_rate = 0.001
top_actor.batch_size = 4096
# critic
top_critic = top_agent_params.network_wrappers['critic']
top_critic.input_embedders_parameters = {