ml-agents/ml-agents/mlagents/trainers/ppo/optimizer_torch.py

from typing import Dict, cast
from mlagents.torch_utils import torch

from mlagents.trainers.buffer import AgentBuffer

from mlagents_envs.timers import timed
from mlagents.trainers.policy.torch_policy import TorchPolicy
from mlagents.trainers.optimizer.torch_optimizer import TorchOptimizer
from mlagents.trainers.settings import TrainerSettings, PPOSettings
from mlagents.trainers.torch.agent_action import AgentAction
from mlagents.trainers.torch.action_log_probs import ActionLogProbs
from mlagents.trainers.torch.utils import ModelUtils
from mlagents.trainers.trajectory import ObsUtil, TeamObsUtil


class TorchPPOOptimizer(TorchOptimizer):
    def __init__(self, policy: TorchPolicy, trainer_settings: TrainerSettings):
        """
        Takes a Policy and a Dict of trainer parameters and creates an Optimizer around the policy.
        The PPO optimizer has a value estimator and a loss function.
        :param policy: A TorchPolicy object that will be updated by this PPO Optimizer.
        :param trainer_params: Trainer parameters dictionary that specifies the
        properties of the trainer.
        """
        # Create the graph here to give more granular control of the TF graph to the Optimizer.

        super().__init__(policy, trainer_settings)
        params = list(self.policy.actor_critic.parameters())
        self.hyperparameters: PPOSettings = cast(
            PPOSettings, trainer_settings.hyperparameters
        )
        self.decay_learning_rate = ModelUtils.DecayedValue(
            self.hyperparameters.learning_rate_schedule,
            self.hyperparameters.learning_rate,
            1e-10,
            self.trainer_settings.max_steps,
        )
        self.decay_epsilon = ModelUtils.DecayedValue(
            self.hyperparameters.learning_rate_schedule,
            self.hyperparameters.epsilon,
            0.1,
            self.trainer_settings.max_steps,
        )
        self.decay_beta = ModelUtils.DecayedValue(
            self.hyperparameters.learning_rate_schedule,
            self.hyperparameters.beta,
            1e-5,
            self.trainer_settings.max_steps,
        )

        self.optimizer = torch.optim.Adam(
            params, lr=self.trainer_settings.hyperparameters.learning_rate, weight_decay=1e-6
        )
        self.stats_name_to_update_name = {
            "Losses/Value Loss": "value_loss",
            "Losses/Policy Loss": "policy_loss",
        }

        self.stream_names = list(self.reward_signals.keys())

        # ModelUtils.soft_update(
        #    self.policy.actor_critic.critic, self.policy.actor_critic.target, 1.0
        # )

    def ppo_value_loss(
        self,
        values: Dict[str, torch.Tensor],
        old_values: Dict[str, torch.Tensor],
        returns: Dict[str, torch.Tensor],
        epsilon: float,
        loss_masks: torch.Tensor,
    ) -> torch.Tensor:
        """
        Evaluates value loss for PPO.
        :param values: Value output of the current network.
        :param old_values: Value stored with experiences in buffer.
        :param returns: Computed returns.
        :param epsilon: Clipping value for value estimate.
        :param loss_mask: Mask for losses. Used with LSTM to ignore 0'ed out experiences.
        """
        value_losses = []
        for name, head in values.items():
            old_val_tensor = old_values[name]
            returns_tensor = returns[name]
            clipped_value_estimate = old_val_tensor + torch.clamp(
                head - old_val_tensor, -1 * epsilon, epsilon
            )
            v_opt_a = (returns_tensor - head) ** 2
            v_opt_b = (returns_tensor - clipped_value_estimate) ** 2
            value_loss = ModelUtils.masked_mean(torch.max(v_opt_a, v_opt_b), loss_masks)
            value_losses.append(value_loss)
        value_loss = torch.mean(torch.stack(value_losses))
        return value_loss

    def coma_regularizer_loss(self, values: Dict[str, torch.Tensor], baseline_values: Dict[str, torch.Tensor]):
        reg_losses = []
        for name, head in values.items():
            reg_loss = torch.nn.functional.mse_loss(head, baseline_values[name])
            reg_losses.append(reg_loss)
        value_loss = torch.mean(torch.stack(reg_losses))
        return value_loss

    def ppo_policy_loss(
        self,
        advantages: torch.Tensor,
        log_probs: torch.Tensor,
        old_log_probs: torch.Tensor,
        loss_masks: torch.Tensor,
    ) -> torch.Tensor:
        """
        Evaluate PPO policy loss.
        :param advantages: Computed advantages.
        :param log_probs: Current policy probabilities
        :param old_log_probs: Past policy probabilities
        :param loss_masks: Mask for losses. Used with LSTM to ignore 0'ed out experiences.
        """
        advantage = advantages.unsqueeze(-1)

        decay_epsilon = self.hyperparameters.epsilon
        r_theta = torch.exp(log_probs - old_log_probs)
        p_opt_a = r_theta * advantage
        p_opt_b = (
            torch.clamp(r_theta, 1.0 - decay_epsilon, 1.0 + decay_epsilon) * advantage
        )
        policy_loss = -1 * ModelUtils.masked_mean(
            torch.min(p_opt_a, p_opt_b), loss_masks
        )
        return policy_loss

    @timed
    def update(self, batch: AgentBuffer, num_sequences: int) -> Dict[str, float]:
        """
        Performs update on model.
        :param batch: Batch of experiences.
        :param num_sequences: Number of sequences to process.
        :return: Results of update.
        """
        # Get decayed parameters
        decay_lr = self.decay_learning_rate.get_value(self.policy.get_current_step())
        decay_eps = self.decay_epsilon.get_value(self.policy.get_current_step())
        decay_bet = self.decay_beta.get_value(self.policy.get_current_step())
        # returns_b = {}
        returns_v = {}
        old_values = {}
        old_marg_values = {}
        for name in self.reward_signals:
            old_values[name] = ModelUtils.list_to_tensor(
                batch[f"{name}_value_estimates"]
            )
            old_marg_values[name] = ModelUtils.list_to_tensor(
                batch[f"{name}_baseline_estimates"]
            )
            returns_v[name] = ModelUtils.list_to_tensor(batch[f"{name}_returns_v"])
            # returns_b[name] = ModelUtils.list_to_tensor(batch[f"{name}_returns_b"])
        #

        n_obs = len(self.policy.behavior_spec.sensor_specs)
        current_obs = ObsUtil.from_buffer(batch, n_obs)
        # Convert to tensors
        current_obs = [ModelUtils.list_to_tensor(obs) for obs in current_obs]

        team_obs = TeamObsUtil.from_buffer(batch, n_obs)
        team_obs = [
            [ModelUtils.list_to_tensor(obs) for obs in _teammate_obs]
            for _teammate_obs in team_obs
        ]

        act_masks = ModelUtils.list_to_tensor(batch["action_mask"])
        actions = AgentAction.from_dict(batch)
        team_actions = AgentAction.from_team_dict(batch)
        # next_team_actions = AgentAction.from_team_dict_next(batch)

        memories = [
            ModelUtils.list_to_tensor(batch["memory"][i])
            for i in range(0, len(batch["memory"]), self.policy.sequence_length)
        ]
        if len(memories) > 0:
            memories = torch.stack(memories).unsqueeze(0)

        log_probs, entropy, baseline_vals, values = self.policy.evaluate_actions(
            current_obs,
            masks=act_masks,
            actions=actions,
            memories=memories,
            team_obs=team_obs,
            team_act=team_actions,
            seq_len=self.policy.sequence_length,
        )
        old_log_probs = ActionLogProbs.from_dict(batch).flatten()
        log_probs = log_probs.flatten()
        loss_masks = ModelUtils.list_to_tensor(batch["masks"], dtype=torch.bool)
        # q_loss = self.ppo_value_loss(qs, old_values, returns_q, decay_eps, loss_masks)

        # Use trust region from value, not baseline
        baseline_loss = self.ppo_value_loss(
            baseline_vals, old_marg_values, returns_v, decay_eps, loss_masks
        )
        value_loss = self.ppo_value_loss(
            values, old_values, returns_v, decay_eps, loss_masks
        )

        # Regularizer loss reduces bias between the baseline and values. Other
        # regularizers are possible here.
        regularizer_loss = self.coma_regularizer_loss(values, baseline_vals)

        policy_loss = self.ppo_policy_loss(
            ModelUtils.list_to_tensor(batch["advantages"]),
            log_probs,
            old_log_probs,
            loss_masks,
        )
        loss = (
            policy_loss
            + 0.25 * (value_loss + baseline_loss)
            + 0.25 * regularizer_loss
            - decay_bet * ModelUtils.masked_mean(entropy, loss_masks)
        )

        # Set optimizer learning rate
        ModelUtils.update_learning_rate(self.optimizer, decay_lr)
        self.optimizer.zero_grad()
        loss.backward()

        self.optimizer.step()

        # ModelUtils.soft_update(
        #    self.policy.actor_critic.critic, self.policy.actor_critic.target, 1.0
        # )
        update_stats = {
            # NOTE: abs() is not technically correct, but matches the behavior in TensorFlow.
            # TODO: After PyTorch is default, change to something more correct.
            "Losses/Policy Loss": torch.abs(policy_loss).item(),
            "Losses/Value Loss": value_loss.item(),
            # "Losses/Q Loss": q_loss.item(),
            "Losses/Baseline Value Loss": baseline_loss.item(),
            "Losses/Regularization Loss": regularizer_loss.item(),
            "Policy/Learning Rate": decay_lr,
            "Policy/Epsilon": decay_eps,
            "Policy/Beta": decay_bet,
        }

        for reward_provider in self.reward_signals.values():
            update_stats.update(reward_provider.update(batch))

        return update_stats

    def get_modules(self):
        modules = {"Optimizer": self.optimizer}
        for reward_provider in self.reward_signals.values():
            modules.update(reward_provider.get_modules())
        return modules