# # Copyright (c) 2017 Intel Corporation # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # from agents.actor_critic_agent import * from random import shuffle # Clipped Proximal Policy Optimization - https://arxiv.org/abs/1707.06347 class ClippedPPOAgent(ActorCriticAgent): def __init__(self, env, tuning_parameters, replicated_device=None, thread_id=0): ActorCriticAgent.__init__(self, env, tuning_parameters, replicated_device, thread_id, create_target_network=True) # signals definition self.value_loss = Signal('Value Loss') self.signals.append(self.value_loss) self.policy_loss = Signal('Policy Loss') self.signals.append(self.policy_loss) self.total_kl_divergence_during_training_process = 0.0 self.unclipped_grads = Signal('Grads (unclipped)') self.signals.append(self.unclipped_grads) self.value_targets = Signal('Value Targets') self.signals.append(self.value_targets) self.kl_divergence = Signal('KL Divergence') self.signals.append(self.kl_divergence) def fill_advantages(self, batch): current_states, next_states, actions, rewards, game_overs, total_return = self.extract_batch(batch) current_state_values = self.main_network.online_network.predict(current_states)[0] current_state_values = current_state_values.squeeze() self.state_values.add_sample(current_state_values) # calculate advantages advantages = [] value_targets = [] if self.policy_gradient_rescaler == PolicyGradientRescaler.A_VALUE: advantages = total_return - current_state_values elif self.policy_gradient_rescaler == PolicyGradientRescaler.GAE: # get bootstraps episode_start_idx = 0 advantages = np.array([]) value_targets = np.array([]) for idx, game_over in enumerate(game_overs): if game_over: # get advantages for the rollout value_bootstrapping = np.zeros((1,)) rollout_state_values = np.append(current_state_values[episode_start_idx:idx+1], value_bootstrapping) rollout_advantages, gae_based_value_targets = \ self.get_general_advantage_estimation_values(rewards[episode_start_idx:idx+1], rollout_state_values) episode_start_idx = idx + 1 advantages = np.append(advantages, rollout_advantages) value_targets = np.append(value_targets, gae_based_value_targets) else: screen.warning("WARNING: The requested policy gradient rescaler is not available") # standardize advantages = (advantages - np.mean(advantages)) / (np.std(advantages) + 1e-8) for transition, advantage, value_target in zip(batch, advantages, value_targets): transition.info['advantage'] = advantage transition.info['gae_based_value_target'] = value_target self.action_advantages.add_sample(advantages) def train_network(self, dataset, epochs): loss = [] for j in range(epochs): loss = { 'total_loss': [], 'policy_losses': [], 'unclipped_grads': [], 'fetch_result': [] } shuffle(dataset) for i in range(int(len(dataset) / self.tp.batch_size)): batch = dataset[i * self.tp.batch_size:(i + 1) * self.tp.batch_size] current_states, _, actions, _, _, total_return = self.extract_batch(batch) advantages = np.array([t.info['advantage'] for t in batch]) gae_based_value_targets = np.array([t.info['gae_based_value_target'] for t in batch]) if not self.tp.env_instance.discrete_controls and len(actions.shape) == 1: actions = np.expand_dims(actions, -1) # get old policy probabilities and distribution result = self.main_network.target_network.predict(current_states) old_policy_distribution = result[1:] # calculate gradients and apply on both the local policy network and on the global policy network fetches = [self.main_network.online_network.output_heads[1].kl_divergence, self.main_network.online_network.output_heads[1].entropy] total_return = np.expand_dims(total_return, -1) value_targets = gae_based_value_targets if self.tp.agent.estimate_value_using_gae else total_return inputs = copy.copy(current_states) # TODO: why is this output 0 and not output 1? inputs['output_0_0'] = actions # TODO: does old_policy_distribution really need to be represented as a list? # A: yes it does, in the event of discrete controls, it has just a mean # otherwise, it has both a mean and standard deviation for input_index, input in enumerate(old_policy_distribution): inputs['output_0_{}'.format(input_index + 1)] = input total_loss, policy_losses, unclipped_grads, fetch_result =\ self.main_network.online_network.accumulate_gradients( inputs, [total_return, advantages], additional_fetches=fetches) self.value_targets.add_sample(value_targets) if self.tp.distributed: self.main_network.apply_gradients_to_global_network() self.main_network.update_online_network() else: self.main_network.apply_gradients_to_online_network() self.main_network.online_network.reset_accumulated_gradients() loss['total_loss'].append(total_loss) loss['policy_losses'].append(policy_losses) loss['unclipped_grads'].append(unclipped_grads) loss['fetch_result'].append(fetch_result) self.unclipped_grads.add_sample(unclipped_grads) for key in loss.keys(): loss[key] = np.mean(loss[key], 0) if self.tp.learning_rate_decay_rate != 0: curr_learning_rate = self.main_network.online_network.get_variable_value(self.tp.learning_rate) self.curr_learning_rate.add_sample(curr_learning_rate) else: curr_learning_rate = self.tp.learning_rate # log training parameters screen.log_dict( OrderedDict([ ("Surrogate loss", loss['policy_losses'][0]), ("KL divergence", loss['fetch_result'][0]), ("Entropy", loss['fetch_result'][1]), ("training epoch", j), ("learning_rate", curr_learning_rate) ]), prefix="Policy training" ) self.total_kl_divergence_during_training_process = loss['fetch_result'][0] self.entropy.add_sample(loss['fetch_result'][1]) self.kl_divergence.add_sample(loss['fetch_result'][0]) return policy_losses def post_training_commands(self): # clean memory self.memory.clean() def train(self): self.main_network.sync() dataset = self.memory.transitions self.fill_advantages(dataset) # take only the requested number of steps dataset = dataset[:self.tp.agent.num_consecutive_playing_steps] if self.tp.distributed and self.tp.agent.share_statistics_between_workers: self.running_observation_stats.push(np.array([np.array(t.state['observation']) for t in dataset])) losses = self.train_network(dataset, 10) self.value_loss.add_sample(losses[0]) self.policy_loss.add_sample(losses[1]) self.update_log() # should be done in order to update the data that has been accumulated * while not playing * return np.append(losses[0], losses[1]) def choose_action(self, current_state, phase=RunPhase.TRAIN): if self.env.discrete_controls: # DISCRETE _, action_values = self.main_network.online_network.predict(self.tf_input_state(current_state)) action_values = action_values.squeeze() if phase == RunPhase.TRAIN: action = self.exploration_policy.get_action(action_values) else: action = np.argmax(action_values) action_info = {"action_probability": action_values[action]} # self.entropy.add_sample(-np.sum(action_values * np.log(action_values))) else: # CONTINUOUS _, action_values_mean, action_values_std = self.main_network.online_network.predict(self.tf_input_state(current_state)) action_values_mean = action_values_mean.squeeze() action_values_std = action_values_std.squeeze() if phase == RunPhase.TRAIN: action = np.squeeze(np.random.randn(1, self.action_space_size) * action_values_std + action_values_mean) # if self.current_episode % 5 == 0 and self.current_episode_steps_counter < 5: # print action else: action = action_values_mean action_info = {"action_probability": action_values_mean} return action, action_info