Implement Buffer that stores Distribution params

metastable_baselines2/common/buffers.py

@@ -0,0 +1,386 @@
+import warnings
+from abc import ABC, abstractmethod
+from typing import Any, Dict, Generator, List, Optional, Tuple, Union, NamedTuple
+
+import numpy as np
+import torch as th
+from gymnasium import spaces
+
+from stable_baselines3.common.preprocessing import get_action_dim, get_obs_shape
+from stable_baselines3.common.type_aliases import (
+    DictRolloutBufferSamples,
+    RolloutBufferSamples,
+    TensorDict
+)
+from stable_baselines3.common.utils import get_device
+from stable_baselines3.common.vec_env import VecNormalize
+
+try:
+    # Check memory used by replay buffer when possible
+    import psutil
+except ImportError:
+    psutil = None
+
+from stable_baselines3.common.buffers import RolloutBuffer, DictRolloutBuffer
+
+class BetterRolloutBufferSamples(NamedTuple):
+    observations: th.Tensor
+    actions: th.Tensor
+    old_values: th.Tensor
+    old_log_prob: th.Tensor
+    mean: th.Tensor
+    cov_Decomp: th.Tensor
+    advantages: th.Tensor
+    returns: th.Tensor
+
+
+class BetterDictRolloutBufferSamples(NamedTuple):
+    observations: TensorDict
+    actions: th.Tensor
+    old_values: th.Tensor
+    old_log_prob: th.Tensor
+    mean: th.Tensor
+    cov_Decomp: th.Tensor
+    advantages: th.Tensor
+    returns: th.Tensor
+
+class BetterRolloutBuffer(RolloutBuffer):
+    """
+    Extended to also save the mean and cov decomp.
+
+    Rollout buffer used in on-policy algorithms like A2C/PPO.
+    It corresponds to ``buffer_size`` transitions collected
+    using the current policy.
+    This experience will be discarded after the policy update.
+    In order to use PPO objective, we also store the current value of each state
+    and the log probability of each taken action.
+
+    The term rollout here refers to the model-free notion and should not
+    be used with the concept of rollout used in model-based RL or planning.
+    Hence, it is only involved in policy and value function training but not action selection.
+
+    :param buffer_size: Max number of element in the buffer
+    :param observation_space: Observation space
+    :param action_space: Action space
+    :param device: PyTorch device
+    :param gae_lambda: Factor for trade-off of bias vs variance for Generalized Advantage Estimator
+        Equivalent to classic advantage when set to 1.
+    :param gamma: Discount factor
+    :param n_envs: Number of parallel environments
+    """
+
+    observations: np.ndarray
+    actions: np.ndarray
+    rewards: np.ndarray
+    advantages: np.ndarray
+    returns: np.ndarray
+    episode_starts: np.ndarray
+    log_probs: np.ndarray
+    values: np.ndarray
+
+    def __init__(
+        self,
+        buffer_size: int,
+        observation_space: spaces.Space,
+        action_space: spaces.Space,
+        device: Union[th.device, str] = "auto",
+        gae_lambda: float = 1,
+        gamma: float = 0.99,
+        n_envs: int = 1,
+    ):
+        super().__init__(buffer_size, observation_space, action_space, device, n_envs=n_envs)
+        self.gae_lambda = gae_lambda
+        self.gamma = gamma
+        self.generator_ready = False
+        self.reset()
+
+    def reset(self) -> None:
+        self.observations = np.zeros((self.buffer_size, self.n_envs, *self.obs_shape), dtype=np.float32)
+        self.actions = np.zeros((self.buffer_size, self.n_envs, self.action_dim), dtype=np.float32)
+        self.rewards = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.returns = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.episode_starts = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.values = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.log_probs = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.means = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.cov_decomps = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.advantages = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.generator_ready = False
+        super().reset()
+
+    def compute_returns_and_advantage(self, last_values: th.Tensor, dones: np.ndarray) -> None:
+        """
+        Post-processing step: compute the lambda-return (TD(lambda) estimate)
+        and GAE(lambda) advantage.
+
+        Uses Generalized Advantage Estimation (https://arxiv.org/abs/1506.02438)
+        to compute the advantage. To obtain Monte-Carlo advantage estimate (A(s) = R - V(S))
+        where R is the sum of discounted reward with value bootstrap
+        (because we don't always have full episode), set ``gae_lambda=1.0`` during initialization.
+
+        The TD(lambda) estimator has also two special cases:
+        - TD(1) is Monte-Carlo estimate (sum of discounted rewards)
+        - TD(0) is one-step estimate with bootstrapping (r_t + gamma * v(s_{t+1}))
+
+        For more information, see discussion in https://github.com/DLR-RM/stable-baselines3/pull/375.
+
+        :param last_values: state value estimation for the last step (one for each env)
+        :param dones: if the last step was a terminal step (one bool for each env).
+        """
+        # Convert to numpy
+        last_values = last_values.clone().cpu().numpy().flatten()
+
+        last_gae_lam = 0
+        for step in reversed(range(self.buffer_size)):
+            if step == self.buffer_size - 1:
+                next_non_terminal = 1.0 - dones
+                next_values = last_values
+            else:
+                next_non_terminal = 1.0 - self.episode_starts[step + 1]
+                next_values = self.values[step + 1]
+            delta = self.rewards[step] + self.gamma * next_values * next_non_terminal - self.values[step]
+            last_gae_lam = delta + self.gamma * self.gae_lambda * next_non_terminal * last_gae_lam
+            self.advantages[step] = last_gae_lam
+        # TD(lambda) estimator, see Github PR #375 or "Telescoping in TD(lambda)"
+        # in David Silver Lecture 4: https://www.youtube.com/watch?v=PnHCvfgC_ZA
+        self.returns = self.advantages + self.values
+
+    def add(
+        self,
+        obs: np.ndarray,
+        action: np.ndarray,
+        reward: np.ndarray,
+        episode_start: np.ndarray,
+        value: th.Tensor,
+        log_prob: th.Tensor,
+        mean: th.Tensor,
+        cov_decomp: th.Tensor,
+    ) -> None:
+        """
+        :param obs: Observation
+        :param action: Action
+        :param reward:
+        :param episode_start: Start of episode signal.
+        :param value: estimated value of the current state
+            following the current policy.
+        :param log_prob: log probability of the action
+            following the current policy.
+        """
+        if len(log_prob.shape) == 0:
+            # Reshape 0-d tensor to avoid error
+            log_prob = log_prob.reshape(-1, 1)
+
+        # Reshape needed when using multiple envs with discrete observations
+        # as numpy cannot broadcast (n_discrete,) to (n_discrete, 1)
+        if isinstance(self.observation_space, spaces.Discrete):
+            obs = obs.reshape((self.n_envs, *self.obs_shape))
+
+        # Reshape to handle multi-dim and discrete action spaces, see GH #970 #1392
+        action = action.reshape((self.n_envs, self.action_dim))
+
+        self.observations[self.pos] = np.array(obs)
+        self.actions[self.pos] = np.array(action)
+        self.rewards[self.pos] = np.array(reward)
+        self.episode_starts[self.pos] = np.array(episode_start)
+        self.values[self.pos] = value.clone().cpu().numpy().flatten()
+        self.log_probs[self.pos] = log_prob.clone().cpu().numpy()
+        self.means[self.pos] = mean.clone().cpu().numpy()
+        self.cov_decomps[self.pos] = cov_decomp.clone().cpu().numpy()
+        self.pos += 1
+        if self.pos == self.buffer_size:
+            self.full = True
+
+    def get(self, batch_size: Optional[int] = None) -> Generator[RolloutBufferSamples, None, None]:
+        assert self.full, ""
+        indices = np.random.permutation(self.buffer_size * self.n_envs)
+        # Prepare the data
+        if not self.generator_ready:
+            _tensor_names = [
+                "observations",
+                "actions",
+                "values",
+                "log_probs",
+                "advantages",
+                "returns",
+            ]
+
+            for tensor in _tensor_names:
+                self.__dict__[tensor] = self.swap_and_flatten(self.__dict__[tensor])
+            self.generator_ready = True
+
+        # Return everything, don't create minibatches
+        if batch_size is None:
+            batch_size = self.buffer_size * self.n_envs
+
+        start_idx = 0
+        while start_idx < self.buffer_size * self.n_envs:
+            yield self._get_samples(indices[start_idx : start_idx + batch_size])
+            start_idx += batch_size
+
+    def _get_samples(
+        self,
+        batch_inds: np.ndarray,
+        env: Optional[VecNormalize] = None,
+    ) -> BetterRolloutBufferSamples:
+        data = (
+            self.observations[batch_inds],
+            self.actions[batch_inds],
+            self.values[batch_inds].flatten(),
+            self.log_probs[batch_inds].flatten(),
+            self.means[batch_inds].flatten(),
+            self.cov_decomps[batch_inds].flatten(),
+            self.advantages[batch_inds].flatten(),
+            self.returns[batch_inds].flatten(),
+        )
+        return BetterRolloutBufferSamples(*tuple(map(self.to_torch, data)))
+
+
+class BetterDictRolloutBuffer(DictRolloutBuffer):
+    """
+    Extended to also save the mean and cov decomp.
+    
+    Dict Rollout buffer used in on-policy algorithms like A2C/PPO.
+    Extends the RolloutBuffer to use dictionary observations
+
+    It corresponds to ``buffer_size`` transitions collected
+    using the current policy.
+    This experience will be discarded after the policy update.
+    In order to use PPO objective, we also store the current value of each state
+    and the log probability of each taken action.
+
+    The term rollout here refers to the model-free notion and should not
+    be used with the concept of rollout used in model-based RL or planning.
+    Hence, it is only involved in policy and value function training but not action selection.
+
+    :param buffer_size: Max number of element in the buffer
+    :param observation_space: Observation space
+    :param action_space: Action space
+    :param device: PyTorch device
+    :param gae_lambda: Factor for trade-off of bias vs variance for Generalized Advantage Estimator
+        Equivalent to Monte-Carlo advantage estimate when set to 1.
+    :param gamma: Discount factor
+    :param n_envs: Number of parallel environments
+    """
+
+    observation_space: spaces.Dict
+    obs_shape: Dict[str, Tuple[int, ...]]  # type: ignore[assignment]
+    observations: Dict[str, np.ndarray]  # type: ignore[assignment]
+
+    def __init__(
+        self,
+        buffer_size: int,
+        observation_space: spaces.Dict,
+        action_space: spaces.Space,
+        device: Union[th.device, str] = "auto",
+        gae_lambda: float = 1,
+        gamma: float = 0.99,
+        n_envs: int = 1,
+    ):
+        super(RolloutBuffer, self).__init__(buffer_size, observation_space, action_space, device, n_envs=n_envs)
+
+        assert isinstance(self.obs_shape, dict), "DictRolloutBuffer must be used with Dict obs space only"
+
+        self.gae_lambda = gae_lambda
+        self.gamma = gamma
+
+        self.generator_ready = False
+        self.reset()
+
+    def reset(self) -> None:
+        self.observations = {}
+        for key, obs_input_shape in self.obs_shape.items():
+            self.observations[key] = np.zeros((self.buffer_size, self.n_envs, *obs_input_shape), dtype=np.float32)
+        self.actions = np.zeros((self.buffer_size, self.n_envs, self.action_dim), dtype=np.float32)
+        self.rewards = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.returns = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.episode_starts = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.values = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.log_probs = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.advantages = np.zeros((self.buffer_size, self.n_envs), dtype=np.float32)
+        self.generator_ready = False
+        super(RolloutBuffer, self).reset()
+
+    def add(  # type: ignore[override]
+        self,
+        obs: Dict[str, np.ndarray],
+        action: np.ndarray,
+        reward: np.ndarray,
+        episode_start: np.ndarray,
+        value: th.Tensor,
+        log_prob: th.Tensor,
+    ) -> None:
+        """
+        :param obs: Observation
+        :param action: Action
+        :param reward:
+        :param episode_start: Start of episode signal.
+        :param value: estimated value of the current state
+            following the current policy.
+        :param log_prob: log probability of the action
+            following the current policy.
+        """
+        if len(log_prob.shape) == 0:
+            # Reshape 0-d tensor to avoid error
+            log_prob = log_prob.reshape(-1, 1)
+
+        for key in self.observations.keys():
+            obs_ = np.array(obs[key])
+            # Reshape needed when using multiple envs with discrete observations
+            # as numpy cannot broadcast (n_discrete,) to (n_discrete, 1)
+            if isinstance(self.observation_space.spaces[key], spaces.Discrete):
+                obs_ = obs_.reshape((self.n_envs,) + self.obs_shape[key])
+            self.observations[key][self.pos] = obs_
+
+        # Reshape to handle multi-dim and discrete action spaces, see GH #970 #1392
+        action = action.reshape((self.n_envs, self.action_dim))
+
+        self.actions[self.pos] = np.array(action)
+        self.rewards[self.pos] = np.array(reward)
+        self.episode_starts[self.pos] = np.array(episode_start)
+        self.values[self.pos] = value.clone().cpu().numpy().flatten()
+        self.log_probs[self.pos] = log_prob.clone().cpu().numpy()
+        self.pos += 1
+        if self.pos == self.buffer_size:
+            self.full = True
+
+    def get(  # type: ignore[override]
+        self,
+        batch_size: Optional[int] = None,
+    ) -> Generator[DictRolloutBufferSamples, None, None]:
+        assert self.full, ""
+        indices = np.random.permutation(self.buffer_size * self.n_envs)
+        # Prepare the data
+        if not self.generator_ready:
+            for key, obs in self.observations.items():
+                self.observations[key] = self.swap_and_flatten(obs)
+
+            _tensor_names = ["actions", "values", "log_probs", "advantages", "returns"]
+
+            for tensor in _tensor_names:
+                self.__dict__[tensor] = self.swap_and_flatten(self.__dict__[tensor])
+            self.generator_ready = True
+
+        # Return everything, don't create minibatches
+        if batch_size is None:
+            batch_size = self.buffer_size * self.n_envs
+
+        start_idx = 0
+        while start_idx < self.buffer_size * self.n_envs:
+            yield self._get_samples(indices[start_idx : start_idx + batch_size])
+            start_idx += batch_size
+
+    def _get_samples(  # type: ignore[override]
+        self,
+        batch_inds: np.ndarray,
+        env: Optional[VecNormalize] = None,
+    ) -> BetterDictRolloutBufferSamples:
+        return BetterDictRolloutBufferSamples(
+            observations={key: self.to_torch(obs[batch_inds]) for (key, obs) in self.observations.items()},
+            actions=self.to_torch(self.actions[batch_inds]),
+            old_values=self.to_torch(self.values[batch_inds].flatten()),
+            old_log_prob=self.to_torch(self.log_probs[batch_inds].flatten()),
+            mean=self.to_torch(self.means[batch_inds].flatten()),
+            cov_decomp=self.to_torch(self.cov_decomps[batch_inds].flatten()),
+            advantages=self.to_torch(self.advantages[batch_inds].flatten()),
+            returns=self.to_torch(self.returns[batch_inds].flatten()),
+        )

metastable_baselines2/common/on_policy_algorithm.py

@@ -8,6 +8,7 @@ from gymnasium import spaces
 
 from stable_baselines3.common.base_class import BaseAlgorithm
 from stable_baselines3.common.buffers import DictRolloutBuffer, RolloutBuffer
+from .buffers import BetterDictRolloutBuffer, BetterRolloutBuffer
 from stable_baselines3.common.callbacks import BaseCallback
 from stable_baselines3.common.policies import ActorCriticPolicy
 from stable_baselines3.common.type_aliases import GymEnv, MaybeCallback, Schedule
@@ -70,6 +71,7 @@ class BetterOnPolicyAlgorithm(OnPolicyAlgorithm):
         use_sde: bool,
         sde_sample_freq: int,
         use_pca: bool,
+        pca_is: bool,
         rollout_buffer_class: Optional[Type[RolloutBuffer]] = None,
         rollout_buffer_kwargs: Optional[Dict[str, Any]] = None,
         stats_window_size: int = 100,
@@ -85,6 +87,9 @@ class BetterOnPolicyAlgorithm(OnPolicyAlgorithm):
         assert not (use_sde and use_pca)
         self.use_pca = use_pca
 
+        assert not pca_is or use_pca
+        self.pca_is = pca_is
+
         assert not rollout_buffer_class and not rollout_buffer_kwargs
 
         super().__init__(
@@ -122,9 +127,9 @@ class BetterOnPolicyAlgorithm(OnPolicyAlgorithm):
 
         if self.rollout_buffer_class is None:
             if isinstance(self.observation_space, spaces.Dict):
-                self.rollout_buffer_class = DictRolloutBuffer
+                self.rollout_buffer_class = BetterDictRolloutBuffer
             else:
-                self.rollout_buffer_class = RolloutBuffer
+                self.rollout_buffer_class = BetterRolloutBuffer
 
         self.rollout_buffer = self.rollout_buffer_class(
             self.n_steps,
@@ -186,7 +191,7 @@ class BetterOnPolicyAlgorithm(OnPolicyAlgorithm):
             with th.no_grad():
                 # Convert to pytorch tensor or to TensorDict
                 obs_tensor = obs_as_tensor(self._last_obs, self.device)
-                actions, values, log_probs = self.policy(obs_tensor)
+                actions, values, log_probs, distributions = self.policy(obs_tensor, conditioned_log_probs=self.pca_is)
             actions = actions.cpu().numpy()
 
             # Rescale and perform action
@@ -231,7 +236,7 @@ class BetterOnPolicyAlgorithm(OnPolicyAlgorithm):
                         terminal_value = self.policy.predict_values(terminal_obs)[0]
                     rewards[idx] += self.gamma * terminal_value
 
-            rollout_buffer.add(self._last_obs, actions, rewards, self._last_episode_starts, values, log_probs)
+            rollout_buffer.add(self._last_obs, actions, rewards, self._last_episode_starts, values, log_probs, distributions.mean, distributions.scale)
             self._last_obs = new_obs
             self._last_episode_starts = dones
 
@@ -254,6 +259,30 @@ class BetterOnPolicyAlgorithm(OnPolicyAlgorithm):
         """
         raise NotImplementedError
 
+    def predict(
+        self,
+        observation: Union[np.ndarray, Dict[str, np.ndarray]],
+        state: Optional[Tuple[np.ndarray, ...]] = None,
+        episode_start: Optional[np.ndarray] = None,
+        deterministic: bool = False,
+        trajectory: th.Tensor = None,
+    ) -> Tuple[np.ndarray, Optional[Tuple[np.ndarray, ...]]]:
+        """
+        Get the policy action from an observation (and optional hidden state).
+        Includes sugar-coating to handle different observations (e.g. normalizing images).
+
+        :param observation: the input observation
+        :param state: The last hidden states (can be None, used in recurrent policies)
+        :param episode_start: The last masks (can be None, used in recurrent policies)
+            this correspond to beginning of episodes,
+            where the hidden states of the RNN must be reset.
+        :param deterministic: Whether or not to return deterministic actions.
+        :param trajectory: Past trajectory. Only required when using PCA.
+        :return: the model's action and the next hidden state
+            (used in recurrent policies)
+        """
+        return self.policy.predict(observation, state, episode_start, deterministic, trajectory=trajectory)
+
     def learn(
         self: SelfOnPolicyAlgorithm,
         total_timesteps: int,

metastable_baselines2/common/policies.py

@@ -34,9 +34,12 @@ from stable_baselines3.common.torch_layers import (
 
 from stable_baselines3.common.policies import ContinuousCritic
 
-from stable_baselines3.common.type_aliases import Schedule
+from stable_baselines3.common.type_aliases import Schedule, RolloutBufferSamples
 from stable_baselines3.common.utils import get_device, is_vectorized_observation, obs_as_tensor
 
+from metastable_projections.projections import BaseProjectionLayer, IdentityProjectionLayer, FrobeniusProjectionLay, WassersteinProjectionLayer, KLProjectionLayer
+
+
 from .distributions import make_proba_distribution
 from metastable_baselines2.common.pca import PCA_Distribution
 
@@ -395,7 +398,7 @@ class BasePolicy(BaseModel, ABC):
 class ActorCriticPolicy(BasePolicy):
     """
     Policy class for actor-critic algorithms (has both policy and value prediction).
-    Used by A2C, PPO and the likes.
+    Used by A2C, PPO, TRPL and the likes.
 
     :param observation_space: Observation space
     :param action_space: Action space
@@ -445,6 +448,7 @@ class ActorCriticPolicy(BasePolicy):
         optimizer_class: Type[th.optim.Optimizer] = th.optim.Adam,
         optimizer_kwargs: Optional[Dict[str, Any]] = None,
         dist_kwargs: Optional[Dict[str, Any]] = {},
+        policy_projection: BaseProjectionLayer = IdentityProjectionLayer(),
     ):
         if optimizer_kwargs is None:
             optimizer_kwargs = {}
@@ -514,6 +518,8 @@ class ActorCriticPolicy(BasePolicy):
         self.use_pca = use_pca
         self.dist_kwargs = dist_kwargs
 
+        self.policy_projection = policy_projection
+
         # Action distribution
         self.action_dist = make_proba_distribution(action_space, use_sde=use_sde, use_pca=use_pca, dist_kwargs=dist_kwargs)
 
@@ -624,7 +630,7 @@ class ActorCriticPolicy(BasePolicy):
         # Setup optimizer with initial learning rate
         self.optimizer = self.optimizer_class(self.parameters(), lr=lr_schedule(1), **self.optimizer_kwargs)
 
-    def forward(self, obs: th.Tensor, deterministic: bool = False) -> Tuple[th.Tensor, th.Tensor, th.Tensor]:
+    def forward(self, obs: th.Tensor, deterministic: bool = False, conditioned_log_probs: bool = False) -> Tuple[th.Tensor, th.Tensor, th.Tensor]:
         """
         Forward pass in all the networks (actor and critic)
 
@@ -644,9 +650,13 @@ class ActorCriticPolicy(BasePolicy):
         values = self.value_net(latent_vf)
         distribution = self._get_action_dist_from_latent(latent_pi)
         actions = distribution.get_actions(deterministic=deterministic)
-        log_prob = distribution.log_prob(actions)
+        if conditioned_log_probs:
+            assert self.use_pca, 'Cannot calculate conditioned log probs when PCA is disabled.'
+            log_prob = distribution.conditioned_log_prob(actions)
+        else:
+            log_prob = distribution.log_prob(actions)
         actions = actions.reshape((-1, *self.action_space.shape))
-        return actions, values, log_prob
+        return actions, values, log_prob, distribution
 
     def extract_features(self, obs: th.Tensor) -> Union[th.Tensor, Tuple[th.Tensor, th.Tensor]]:
         """
@@ -691,7 +701,7 @@ class ActorCriticPolicy(BasePolicy):
         else:
             raise ValueError("Invalid action distribution")
 
-    def _predict(self, observation: th.Tensor, deterministic: bool = False) -> th.Tensor:
+    def _predict(self, observation: th.Tensor, deterministic: bool = False, trajectory: th.Tensor = None) -> th.Tensor:
         """
         Get the action according to the policy for a given observation.
 
@@ -699,19 +709,23 @@ class ActorCriticPolicy(BasePolicy):
         :param deterministic: Whether to use stochastic or deterministic actions
         :return: Taken action according to the policy
         """
-        return self.get_distribution(observation).get_actions(deterministic=deterministic)
+        if self.use_pca:
+            return self.get_distribution(observation).get_actions(deterministic=deterministic, trajectory=trajectory)
+        else:
+            return self.get_distribution(observation).get_actions(deterministic=deterministic)
 
-    def evaluate_actions(self, obs: th.Tensor, actions: th.Tensor) -> Tuple[th.Tensor, th.Tensor, Optional[th.Tensor]]:
+    def evaluate_actions(self, rollout_data: RolloutBufferSamples, actions: th.Tensor) -> Tuple[th.Tensor, th.Tensor, Optional[th.Tensor]]:
         """
         Evaluate actions according to the current policy,
         given the observations.
 
-        :param obs: Observation
+        :param rollout_data: The Rollouts (containing )
         :param actions: Actions
         :return: estimated value, log likelihood of taking those actions
             and entropy of the action distribution.
         """
         # Preprocess the observation if needed
+        obs = rollout_data.observations
         features = self.extract_features(obs)
         if self.share_features_extractor:
             latent_pi, latent_vf = self.mlp_extractor(features)
@@ -719,11 +733,13 @@ class ActorCriticPolicy(BasePolicy):
             pi_features, vf_features = features
             latent_pi = self.mlp_extractor.forward_actor(pi_features)
             latent_vf = self.mlp_extractor.forward_critic(vf_features)
-        distribution = self._get_action_dist_from_latent(latent_pi)
+        raw_distribution = self._get_action_dist_from_latent(latent_pi)
+        distribution, old_distribution = self.policy_projection.project_from_rollouts(raw_distribution, rollout_data)
         log_prob = distribution.log_prob(actions)
         values = self.value_net(latent_vf)
         entropy = distribution.entropy()
-        return values, log_prob, entropy
+        trust_region_loss = self.projection.get_trust_region_loss(raw_distribution, old_distribution)
+        return values, log_prob, entropy, trust_region_loss
 
     def get_distribution(self, obs: th.Tensor) -> Distribution:
         """
@@ -918,18 +934,18 @@ class Actor(BasePolicy):
         log_std = th.clamp(log_std, LOG_STD_MIN, LOG_STD_MAX)
         return mean_actions, log_std, {}
 
-    def forward(self, obs: th.Tensor, deterministic: bool = False) -> th.Tensor:
+    def forward(self, obs: th.Tensor, deterministic: bool = False, trajectory: th.Tensor = None) -> th.Tensor:
         mean_actions, log_std, kwargs = self.get_action_dist_params(obs)
         # Note: the action is squashed
-        return self.action_dist.actions_from_params(mean_actions, log_std, deterministic=deterministic, **kwargs)
+        return self.action_dist.actions_from_params(mean_actions, log_std, deterministic=deterministic, trajectory=trajectory, **kwargs)
 
     def action_log_prob(self, obs: th.Tensor) -> Tuple[th.Tensor, th.Tensor]:
         mean_actions, log_std, kwargs = self.get_action_dist_params(obs)
         # return action and associated log prob
         return self.action_dist.log_prob_from_params(mean_actions, log_std, **kwargs)
 
-    def _predict(self, observation: th.Tensor, deterministic: bool = False) -> th.Tensor:
+    def _predict(self, observation: th.Tensor, deterministic: bool = False, trajectory: th.Tensor = None) -> th.Tensor:
-        return self(observation, deterministic)
+        return self(observation, deterministic, trajectory=trajectory)
 
 
 class SACPolicy(BasePolicy):
@@ -1018,16 +1034,14 @@ class SACPolicy(BasePolicy):
         }
         self.actor_kwargs = self.net_args.copy()
 
-        sde_kwargs = {
+        self.actor_kwargs.update({
             "use_sde": use_sde,
             "use_pca": use_pca,
             "log_std_init": log_std_init,
             "use_expln": use_expln,
             "clip_mean": clip_mean,
             "dist_kwargs": dist_kwargs,
-        }
+        })
-
-        self.actor_kwargs.update(sde_kwargs)
         self.critic_kwargs = self.net_args.copy()
         self.critic_kwargs.update(
             {