Source code for torchrl.objectives.ppo
# Copyright (c) Meta Platforms, Inc. and affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
from __future__ import annotations
import contextlib
import math
from copy import deepcopy
from dataclasses import dataclass
from typing import Tuple
import torch
from tensordict import TensorDict, TensorDictBase, TensorDictParams
from tensordict.nn import (
dispatch,
ProbabilisticTensorDictModule,
ProbabilisticTensorDictSequential,
TensorDictModule,
)
from tensordict.utils import NestedKey
from torch import distributions as d
from torchrl.objectives.common import LossModule
from torchrl.objectives.utils import (
_cache_values,
_clip_value_loss,
_GAMMA_LMBDA_DEPREC_ERROR,
_reduce,
default_value_kwargs,
distance_loss,
ValueEstimators,
)
from torchrl.objectives.value import (
GAE,
TD0Estimator,
TD1Estimator,
TDLambdaEstimator,
VTrace,
)
[docs]class PPOLoss(LossModule):
"""A parent PPO loss class.
PPO (Proximal Policy Optimisation) is a model-free, online RL algorithm
that makes use of a recorded (batch of)
trajectories to perform several optimization steps, while actively
preventing the updated policy to deviate too
much from its original parameter configuration.
PPO loss can be found in different flavours, depending on the way the
constrained optimisation is implemented: ClipPPOLoss and KLPENPPOLoss.
Unlike its subclasses, this class does not implement any regularisation
and should therefore be used cautiously.
For more details regarding PPO, refer to: "Proximal Policy Optimization Algorithms",
https://arxiv.org/abs/1707.06347
Args:
actor_network (ProbabilisticTensorDictSequential): policy operator.
critic_network (ValueOperator): value operator.
Keyword Args:
entropy_bonus (bool, optional): if ``True``, an entropy bonus will be added to the
loss to favour exploratory policies.
samples_mc_entropy (int, optional): if the distribution retrieved from the policy
operator does not have a closed form
formula for the entropy, a Monte-Carlo estimate will be used.
``samples_mc_entropy`` will control how many
samples will be used to compute this estimate.
Defaults to ``1``.
entropy_coef (scalar, optional): entropy multiplier when computing the total loss.
Defaults to ``0.01``.
critic_coef (scalar, optional): critic loss multiplier when computing the total
loss. Defaults to ``1.0``.
loss_critic_type (str, optional): loss function for the value discrepancy.
Can be one of "l1", "l2" or "smooth_l1". Defaults to ``"smooth_l1"``.
normalize_advantage (bool, optional): if ``True``, the advantage will be normalized
before being used. Defaults to ``False``.
separate_losses (bool, optional): if ``True``, shared parameters between
policy and critic will only be trained on the policy loss.
Defaults to ``False``, i.e., gradients are propagated to shared
parameters for both policy and critic losses.
advantage_key (str, optional): [Deprecated, use set_keys(advantage_key=advantage_key) instead]
The input tensordict key where the advantage is
expected to be written. Defaults to ``"advantage"``.
value_target_key (str, optional): [Deprecated, use set_keys(value_target_key=value_target_key) instead]
The input tensordict key where the target state
value is expected to be written. Defaults to ``"value_target"``.
value_key (str, optional): [Deprecated, use set_keys(value_key) instead]
The input tensordict key where the state
value is expected to be written. Defaults to ``"state_value"``.
functional (bool, optional): whether modules should be functionalized.
Functionalizing permits features like meta-RL, but makes it
impossible to use distributed models (DDP, FSDP, ...) and comes
with a little cost. Defaults to ``True``.
reduction (str, optional): Specifies the reduction to apply to the output:
``"none"`` | ``"mean"`` | ``"sum"``. ``"none"``: no reduction will be applied,
``"mean"``: the sum of the output will be divided by the number of
elements in the output, ``"sum"``: the output will be summed. Default: ``"mean"``.
clip_value (float, optional): If provided, it will be used to compute a clipped version of the value
prediction with respect to the input tensordict value estimate and use it to calculate the value loss.
The purpose of clipping is to limit the impact of extreme value predictions, helping stabilize training
and preventing large updates. However, it will have no impact if the value estimate was done by the current
version of the value estimator. Defaults to ``None``.
.. note::
The advantage (typically GAE) can be computed by the loss function or
in the training loop. The latter option is usually preferred, but this is
up to the user to choose which option is to be preferred.
If the advantage key (``"advantage`` by default) is not present in the
input tensordict, the advantage will be computed by the :meth:`~.forward`
method.
>>> ppo_loss = PPOLoss(actor, critic)
>>> advantage = GAE(critic)
>>> data = next(datacollector)
>>> losses = ppo_loss(data)
>>> # equivalent
>>> advantage(data)
>>> losses = ppo_loss(data)
A custom advantage module can be built using :meth:`~.make_value_estimator`.
The default is :class:`~torchrl.objectives.value.GAE` with hyperparameters
dictated by :func:`~torchrl.objectives.utils.default_value_kwargs`.
>>> ppo_loss = PPOLoss(actor, critic)
>>> ppo_loss.make_value_estimator(ValueEstimators.TDLambda)
>>> data = next(datacollector)
>>> losses = ppo_loss(data)
.. note::
If the actor and the value function share parameters, one can avoid
calling the common module multiple times by passing only the head of the
value network to the PPO loss module:
>>> common = SomeModule(in_keys=["observation"], out_keys=["hidden"])
>>> actor_head = SomeActor(in_keys=["hidden"])
>>> value_head = SomeValue(in_keys=["hidden"])
>>> # first option, with 2 calls on the common module
>>> model = ActorCriticOperator(common, actor_head, value_head)
>>> loss_module = PPOLoss(model.get_policy_operator(), model.get_value_operator())
>>> # second option, with a single call to the common module
>>> loss_module = PPOLoss(ProbabilisticTensorDictSequential(model, actor_head), value_head)
This will work regardless of whether separate_losses is activated or not.
Examples:
>>> import torch
>>> from torch import nn
>>> from torchrl.data.tensor_specs import BoundedTensorSpec
>>> from torchrl.modules.distributions import NormalParamExtractor, TanhNormal
>>> from torchrl.modules.tensordict_module.actors import ProbabilisticActor, ValueOperator
>>> from torchrl.modules.tensordict_module.common import SafeModule
>>> from torchrl.objectives.ppo import PPOLoss
>>> from tensordict import TensorDict
>>> n_act, n_obs = 4, 3
>>> spec = BoundedTensorSpec(-torch.ones(n_act), torch.ones(n_act), (n_act,))
>>> base_layer = nn.Linear(n_obs, 5)
>>> net = nn.Sequential(base_layer, nn.Linear(5, 2 * n_act), NormalParamExtractor())
>>> module = SafeModule(net, in_keys=["observation"], out_keys=["loc", "scale"])
>>> actor = ProbabilisticActor(
... module=module,
... distribution_class=TanhNormal,
... in_keys=["loc", "scale"],
... spec=spec)
>>> module = nn.Sequential(base_layer, nn.Linear(5, 1))
>>> value = ValueOperator(
... module=module,
... in_keys=["observation"])
>>> loss = PPOLoss(actor, value)
>>> batch = [2, ]
>>> action = spec.rand(batch)
>>> data = TensorDict({"observation": torch.randn(*batch, n_obs),
... "action": action,
... "sample_log_prob": torch.randn_like(action[..., 1]),
... ("next", "done"): torch.zeros(*batch, 1, dtype=torch.bool),
... ("next", "terminated"): torch.zeros(*batch, 1, dtype=torch.bool),
... ("next", "reward"): torch.randn(*batch, 1),
... ("next", "observation"): torch.randn(*batch, n_obs),
... }, batch)
>>> loss(data)
TensorDict(
fields={
entropy: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.float32, is_shared=False),
loss_critic: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.float32, is_shared=False),
loss_entropy: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.float32, is_shared=False),
loss_objective: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.float32, is_shared=False)},
batch_size=torch.Size([]),
device=None,
is_shared=False)
This class is compatible with non-tensordict based modules too and can be
used without recurring to any tensordict-related primitive. In this case,
the expected keyword arguments are:
``["action", "sample_log_prob", "next_reward", "next_done", "next_terminated"]`` + in_keys of the actor and value network.
The return value is a tuple of tensors in the following order:
``["loss_objective"]`` + ``["entropy", "loss_entropy"]`` if entropy_bonus is set + ``"loss_critic"`` if critic_coef is not ``None``.
The output keys can also be filtered using :meth:`PPOLoss.select_out_keys` method.
Examples:
>>> import torch
>>> from torch import nn
>>> from torchrl.data.tensor_specs import BoundedTensorSpec
>>> from torchrl.modules.distributions import NormalParamExtractor, TanhNormal
>>> from torchrl.modules.tensordict_module.actors import ProbabilisticActor, ValueOperator
>>> from torchrl.modules.tensordict_module.common import SafeModule
>>> from torchrl.objectives.ppo import PPOLoss
>>> n_act, n_obs = 4, 3
>>> spec = BoundedTensorSpec(-torch.ones(n_act), torch.ones(n_act), (n_act,))
>>> base_layer = nn.Linear(n_obs, 5)
>>> net = nn.Sequential(base_layer, nn.Linear(5, 2 * n_act), NormalParamExtractor())
>>> module = SafeModule(net, in_keys=["observation"], out_keys=["loc", "scale"])
>>> actor = ProbabilisticActor(
... module=module,
... distribution_class=TanhNormal,
... in_keys=["loc", "scale"],
... spec=spec)
>>> module = nn.Sequential(base_layer, nn.Linear(5, 1))
>>> value = ValueOperator(
... module=module,
... in_keys=["observation"])
>>> loss = PPOLoss(actor, value)
>>> loss.set_keys(sample_log_prob="sampleLogProb")
>>> _ = loss.select_out_keys("loss_objective")
>>> batch = [2, ]
>>> action = spec.rand(batch)
>>> loss_objective = loss(
... observation=torch.randn(*batch, n_obs),
... action=action,
... sampleLogProb=torch.randn_like(action[..., 1]) / 10,
... next_done=torch.zeros(*batch, 1, dtype=torch.bool),
... next_terminated=torch.zeros(*batch, 1, dtype=torch.bool),
... next_reward=torch.randn(*batch, 1),
... next_observation=torch.randn(*batch, n_obs))
>>> loss_objective.backward()
"""
@dataclass
class _AcceptedKeys:
"""Maintains default values for all configurable tensordict keys.
This class defines which tensordict keys can be set using '.set_keys(key_name=key_value)' and their
default values
Attributes:
advantage (NestedKey): The input tensordict key where the advantage is expected.
Will be used for the underlying value estimator. Defaults to ``"advantage"``.
value_target (NestedKey): The input tensordict key where the target state value is expected.
Will be used for the underlying value estimator Defaults to ``"value_target"``.
value (NestedKey): The input tensordict key where the state value is expected.
Will be used for the underlying value estimator. Defaults to ``"state_value"``.
sample_log_prob (NestedKey): The input tensordict key where the
sample log probability is expected. Defaults to ``"sample_log_prob"``.
action (NestedKey): The input tensordict key where the action is expected.
Defaults to ``"action"``.
reward (NestedKey): The input tensordict key where the reward is expected.
Will be used for the underlying value estimator. Defaults to ``"reward"``.
done (NestedKey): The key in the input TensorDict that indicates
whether a trajectory is done. Will be used for the underlying value estimator.
Defaults to ``"done"``.
terminated (NestedKey): The key in the input TensorDict that indicates
whether a trajectory is terminated. Will be used for the underlying value estimator.
Defaults to ``"terminated"``.
"""
advantage: NestedKey = "advantage"
value_target: NestedKey = "value_target"
value: NestedKey = "state_value"
sample_log_prob: NestedKey = "sample_log_prob"
action: NestedKey = "action"
reward: NestedKey = "reward"
done: NestedKey = "done"
terminated: NestedKey = "terminated"
default_keys = _AcceptedKeys()
default_value_estimator = ValueEstimators.GAE
actor_network: TensorDictModule
critic_network: TensorDictModule
actor_network_params: TensorDictParams
critic_network_params: TensorDictParams
target_actor_network_params: TensorDictParams
target_critic_network_params: TensorDictParams
def __init__(
self,
actor_network: ProbabilisticTensorDictSequential | None = None,
critic_network: TensorDictModule | None = None,
*,
entropy_bonus: bool = True,
samples_mc_entropy: int = 1,
entropy_coef: float = 0.01,
critic_coef: float = 1.0,
loss_critic_type: str = "smooth_l1",
normalize_advantage: bool = False,
gamma: float = None,
separate_losses: bool = False,
advantage_key: str = None,
value_target_key: str = None,
value_key: str = None,
functional: bool = True,
actor: ProbabilisticTensorDictSequential = None,
critic: ProbabilisticTensorDictSequential = None,
reduction: str = None,
clip_value: float | None = None,
**kwargs,
):
if actor is not None:
actor_network = actor
del actor
if critic is not None:
critic_network = critic
del critic
if actor_network is None or critic_network is None:
raise TypeError(
"Missing positional arguments actor_network or critic_network."
)
if reduction is None:
reduction = "mean"
self._functional = functional
self._in_keys = None
self._out_keys = None
super().__init__()
if functional:
self.convert_to_functional(actor_network, "actor_network")
else:
self.actor_network = actor_network
self.actor_network_params = None
self.target_actor_network_params = None
if separate_losses:
# we want to make sure there are no duplicates in the params: the
# params of critic must be refs to actor if they're shared
policy_params = list(actor_network.parameters())
else:
policy_params = None
if functional:
self.convert_to_functional(
critic_network, "critic_network", compare_against=policy_params
)
else:
self.critic_network = critic_network
self.critic_network_params = None
self.target_critic_network_params = None
self.samples_mc_entropy = samples_mc_entropy
self.entropy_bonus = entropy_bonus
self.separate_losses = separate_losses
self.reduction = reduction
try:
device = next(self.parameters()).device
except AttributeError:
device = torch.device("cpu")
self.register_buffer("entropy_coef", torch.tensor(entropy_coef, device=device))
self.register_buffer("critic_coef", torch.tensor(critic_coef, device=device))
self.loss_critic_type = loss_critic_type
self.normalize_advantage = normalize_advantage
if gamma is not None:
raise TypeError(_GAMMA_LMBDA_DEPREC_ERROR)
self._set_deprecated_ctor_keys(
advantage=advantage_key,
value_target=value_target_key,
value=value_key,
)
if clip_value is not None:
if isinstance(clip_value, float):
clip_value = torch.tensor(clip_value)
elif isinstance(clip_value, torch.Tensor):
if clip_value.numel() != 1:
raise ValueError(
f"clip_value must be a float or a scalar tensor, got {clip_value}."
)
else:
raise ValueError(
f"clip_value must be a float or a scalar tensor, got {clip_value}."
)
self.register_buffer("clip_value", clip_value)
@property
def functional(self):
return self._functional
def _set_in_keys(self):
keys = [
self.tensor_keys.action,
self.tensor_keys.sample_log_prob,
("next", self.tensor_keys.reward),
("next", self.tensor_keys.done),
("next", self.tensor_keys.terminated),
*self.actor_network.in_keys,
*[("next", key) for key in self.actor_network.in_keys],
*self.critic_network.in_keys,
]
self._in_keys = list(set(keys))
@property
def in_keys(self):
if self._in_keys is None:
self._set_in_keys()
return self._in_keys
@in_keys.setter
def in_keys(self, values):
self._in_keys = values
@property
def out_keys(self):
if self._out_keys is None:
keys = ["loss_objective"]
if self.entropy_bonus:
keys.extend(["entropy", "loss_entropy"])
if self.loss_critic:
keys.append("loss_critic")
if self.clip_value:
keys.append("value_clip_fraction")
self._out_keys = keys
return self._out_keys
@out_keys.setter
def out_keys(self, values):
self._out_keys = values
def _forward_value_estimator_keys(self, **kwargs) -> None:
if hasattr(self, "_value_estimator") and self._value_estimator is not None:
self._value_estimator.set_keys(
advantage=self.tensor_keys.advantage,
value_target=self.tensor_keys.value_target,
value=self.tensor_keys.value,
reward=self.tensor_keys.reward,
done=self.tensor_keys.done,
terminated=self.tensor_keys.terminated,
)
self._set_in_keys()
def reset(self) -> None:
pass
def get_entropy_bonus(self, dist: d.Distribution) -> torch.Tensor:
try:
entropy = dist.entropy()
except NotImplementedError:
x = dist.rsample((self.samples_mc_entropy,))
entropy = -dist.log_prob(x).mean(0)
return entropy.unsqueeze(-1)
def _log_weight(
self, tensordict: TensorDictBase
) -> Tuple[torch.Tensor, d.Distribution]:
# current log_prob of actions
action = tensordict.get(self.tensor_keys.action)
if action.requires_grad:
raise RuntimeError(
f"tensordict stored {self.tensor_keys.action} requires grad."
)
with self.actor_network_params.to_module(
self.actor_network
) if self.functional else contextlib.nullcontext():
dist = self.actor_network.get_dist(tensordict)
log_prob = dist.log_prob(action)
prev_log_prob = tensordict.get(self.tensor_keys.sample_log_prob)
if prev_log_prob.requires_grad:
raise RuntimeError("tensordict prev_log_prob requires grad.")
log_weight = (log_prob - prev_log_prob).unsqueeze(-1)
kl_approx = (prev_log_prob - log_prob).unsqueeze(-1)
return log_weight, dist, kl_approx
def loss_critic(self, tensordict: TensorDictBase) -> torch.Tensor:
# TODO: if the advantage is gathered by forward, this introduces an
# overhead that we could easily reduce.
if self.separate_losses:
tensordict = tensordict.detach()
try:
target_return = tensordict.get(self.tensor_keys.value_target)
except KeyError:
raise KeyError(
f"the key {self.tensor_keys.value_target} was not found in the input tensordict. "
f"Make sure you provided the right key and the value_target (i.e. the target "
f"return) has been retrieved accordingly. Advantage classes such as GAE, "
f"TDLambdaEstimate and TDEstimate all return a 'value_target' entry that "
f"can be used for the value loss."
)
if self.clip_value:
try:
old_state_value = tensordict.get(self.tensor_keys.value)
except KeyError:
raise KeyError(
f"clip_value is set to {self.clip_value}, but "
f"the key {self.tensor_keys.value} was not found in the input tensordict. "
f"Make sure that the value_key passed to PPO exists in the input tensordict."
)
with self.critic_network_params.to_module(
self.critic_network
) if self.functional else contextlib.nullcontext():
state_value_td = self.critic_network(tensordict)
try:
state_value = state_value_td.get(self.tensor_keys.value)
except KeyError:
raise KeyError(
f"the key {self.tensor_keys.value} was not found in the critic output tensordict. "
f"Make sure that the value_key passed to PPO is accurate."
)
loss_value = distance_loss(
target_return,
state_value,
loss_function=self.loss_critic_type,
)
clip_fraction = None
if self.clip_value:
loss_value, clip_fraction = _clip_value_loss(
old_state_value,
state_value,
self.clip_value.to(state_value.device),
target_return,
loss_value,
self.loss_critic_type,
)
return self.critic_coef * loss_value, clip_fraction
@property
@_cache_values
def _cached_critic_network_params_detached(self):
if not self.functional:
return None
return self.critic_network_params.detach()
[docs] @dispatch
def forward(self, tensordict: TensorDictBase) -> TensorDictBase:
tensordict = tensordict.clone(False)
advantage = tensordict.get(self.tensor_keys.advantage, None)
if advantage is None:
self.value_estimator(
tensordict,
params=self._cached_critic_network_params_detached,
target_params=self.target_critic_network_params,
)
advantage = tensordict.get(self.tensor_keys.advantage)
if self.normalize_advantage and advantage.numel() > 1:
loc = advantage.mean()
scale = advantage.std().clamp_min(1e-6)
advantage = (advantage - loc) / scale
log_weight, dist, kl_approx = self._log_weight(tensordict)
neg_loss = log_weight.exp() * advantage
td_out = TensorDict({"loss_objective": -neg_loss}, batch_size=[])
if self.entropy_bonus:
entropy = self.get_entropy_bonus(dist)
td_out.set("entropy", entropy.detach().mean()) # for logging
td_out.set("kl_approx", kl_approx.detach().mean()) # for logging
td_out.set("loss_entropy", -self.entropy_coef * entropy)
if self.critic_coef:
loss_critic, value_clip_fraction = self.loss_critic(tensordict)
td_out.set("loss_critic", loss_critic)
if value_clip_fraction is not None:
td_out.set("value_clip_fraction", value_clip_fraction)
td_out = td_out.named_apply(
lambda name, value: _reduce(value, reduction=self.reduction).squeeze(-1)
if name.startswith("loss_")
else value,
batch_size=[],
)
return td_out
[docs] def make_value_estimator(self, value_type: ValueEstimators = None, **hyperparams):
if value_type is None:
value_type = self.default_value_estimator
self.value_type = value_type
hp = dict(default_value_kwargs(value_type))
if hasattr(self, "gamma"):
hp["gamma"] = self.gamma
hp.update(hyperparams)
if value_type == ValueEstimators.TD1:
self._value_estimator = TD1Estimator(
value_network=self.critic_network, **hp
)
elif value_type == ValueEstimators.TD0:
self._value_estimator = TD0Estimator(
value_network=self.critic_network, **hp
)
elif value_type == ValueEstimators.GAE:
self._value_estimator = GAE(value_network=self.critic_network, **hp)
elif value_type == ValueEstimators.TDLambda:
self._value_estimator = TDLambdaEstimator(
value_network=self.critic_network, **hp
)
elif value_type == ValueEstimators.VTrace:
# VTrace currently does not support functional call on the actor
if self.functional:
actor_with_params = deepcopy(self.actor_network)
self.actor_network_params.to_module(actor_with_params)
else:
actor_with_params = self.actor_network
self._value_estimator = VTrace(
value_network=self.critic_network, actor_network=actor_with_params, **hp
)
else:
raise NotImplementedError(f"Unknown value type {value_type}")
tensor_keys = {
"advantage": self.tensor_keys.advantage,
"value": self.tensor_keys.value,
"value_target": self.tensor_keys.value_target,
"reward": self.tensor_keys.reward,
"done": self.tensor_keys.done,
"terminated": self.tensor_keys.terminated,
"sample_log_prob": self.tensor_keys.sample_log_prob,
}
self._value_estimator.set_keys(**tensor_keys)
[docs]class ClipPPOLoss(PPOLoss):
"""Clipped PPO loss.
The clipped importance weighted loss is computed as follows:
loss = -min( weight * advantage, min(max(weight, 1-eps), 1+eps) * advantage)
Args:
actor_network (ProbabilisticTensorDictSequential): policy operator.
critic_network (ValueOperator): value operator.
Keyword Args:
clip_epsilon (scalar, optional): weight clipping threshold in the clipped PPO loss equation.
default: 0.2
entropy_bonus (bool, optional): if ``True``, an entropy bonus will be added to the
loss to favour exploratory policies.
samples_mc_entropy (int, optional): if the distribution retrieved from the policy
operator does not have a closed form
formula for the entropy, a Monte-Carlo estimate will be used.
``samples_mc_entropy`` will control how many
samples will be used to compute this estimate.
Defaults to ``1``.
entropy_coef (scalar, optional): entropy multiplier when computing the total loss.
Defaults to ``0.01``.
critic_coef (scalar, optional): critic loss multiplier when computing the total
loss. Defaults to ``1.0``.
loss_critic_type (str, optional): loss function for the value discrepancy.
Can be one of "l1", "l2" or "smooth_l1". Defaults to ``"smooth_l1"``.
normalize_advantage (bool, optional): if ``True``, the advantage will be normalized
before being used. Defaults to ``False``.
separate_losses (bool, optional): if ``True``, shared parameters between
policy and critic will only be trained on the policy loss.
Defaults to ``False``, i.e., gradients are propagated to shared
parameters for both policy and critic losses.
advantage_key (str, optional): [Deprecated, use set_keys(advantage_key=advantage_key) instead]
The input tensordict key where the advantage is
expected to be written. Defaults to ``"advantage"``.
value_target_key (str, optional): [Deprecated, use set_keys(value_target_key=value_target_key) instead]
The input tensordict key where the target state
value is expected to be written. Defaults to ``"value_target"``.
value_key (str, optional): [Deprecated, use set_keys(value_key) instead]
The input tensordict key where the state
value is expected to be written. Defaults to ``"state_value"``.
functional (bool, optional): whether modules should be functionalized.
Functionalizing permits features like meta-RL, but makes it
impossible to use distributed models (DDP, FSDP, ...) and comes
with a little cost. Defaults to ``True``.
reduction (str, optional): Specifies the reduction to apply to the output:
``"none"`` | ``"mean"`` | ``"sum"``. ``"none"``: no reduction will be applied,
``"mean"``: the sum of the output will be divided by the number of
elements in the output, ``"sum"``: the output will be summed. Default: ``"mean"``.
clip_value (bool or float, optional): If a ``float`` is provided, it will be used to compute a clipped
version of the value prediction with respect to the input tensordict value estimate and use it to
calculate the value loss. The purpose of clipping is to limit the impact of extreme value predictions,
helping stabilize training and preventing large updates. However, it will have no impact if the value
estimate was done by the current version of the value estimator. If instead ``True`` is provided, the
``clip_epsilon`` parameter will be used as the clipping threshold. If not provided or ``False``, no
clipping will be performed. Defaults to ``False``.
.. note:
The advantage (typically GAE) can be computed by the loss function or
in the training loop. The latter option is usually preferred, but this is
up to the user to choose which option is to be preferred.
If the advantage key (``"advantage`` by default) is not present in the
input tensordict, the advantage will be computed by the :meth:`~.forward`
method.
>>> ppo_loss = ClipPPOLoss(actor, critic)
>>> advantage = GAE(critic)
>>> data = next(datacollector)
>>> losses = ppo_loss(data)
>>> # equivalent
>>> advantage(data)
>>> losses = ppo_loss(data)
A custom advantage module can be built using :meth:`~.make_value_estimator`.
The default is :class:`~torchrl.objectives.value.GAE` with hyperparameters
dictated by :func:`~torchrl.objectives.utils.default_value_kwargs`.
>>> ppo_loss = ClipPPOLoss(actor, critic)
>>> ppo_loss.make_value_estimator(ValueEstimators.TDLambda)
>>> data = next(datacollector)
>>> losses = ppo_loss(data)
.. note::
If the actor and the value function share parameters, one can avoid
calling the common module multiple times by passing only the head of the
value network to the PPO loss module:
>>> common = SomeModule(in_keys=["observation"], out_keys=["hidden"])
>>> actor_head = SomeActor(in_keys=["hidden"])
>>> value_head = SomeValue(in_keys=["hidden"])
>>> # first option, with 2 calls on the common module
>>> model = ActorCriticOperator(common, actor_head, value_head)
>>> loss_module = PPOLoss(model.get_policy_operator(), model.get_value_operator())
>>> # second option, with a single call to the common module
>>> loss_module = PPOLoss(ProbabilisticTensorDictSequential(model, actor_head), value_head)
This will work regardless of whether separate_losses is activated or not.
"""
actor_network: TensorDictModule
critic_network: TensorDictModule
actor_network_params: TensorDictParams
critic_network_params: TensorDictParams
target_actor_network_params: TensorDictParams
target_critic_network_params: TensorDictParams
def __init__(
self,
actor_network: ProbabilisticTensorDictSequential | None = None,
critic_network: TensorDictModule | None = None,
*,
clip_epsilon: float = 0.2,
entropy_bonus: bool = True,
samples_mc_entropy: int = 1,
entropy_coef: float = 0.01,
critic_coef: float = 1.0,
loss_critic_type: str = "smooth_l1",
normalize_advantage: bool = False,
gamma: float = None,
separate_losses: bool = False,
reduction: str = None,
clip_value: bool | float | None = None,
**kwargs,
):
# Define clipping of the value loss
if isinstance(clip_value, bool):
clip_value = clip_epsilon if clip_value else None
super(ClipPPOLoss, self).__init__(
actor_network,
critic_network,
entropy_bonus=entropy_bonus,
samples_mc_entropy=samples_mc_entropy,
entropy_coef=entropy_coef,
critic_coef=critic_coef,
loss_critic_type=loss_critic_type,
normalize_advantage=normalize_advantage,
gamma=gamma,
separate_losses=separate_losses,
reduction=reduction,
clip_value=clip_value,
**kwargs,
)
self.register_buffer("clip_epsilon", torch.tensor(clip_epsilon))
@property
def _clip_bounds(self):
return (
math.log1p(-self.clip_epsilon),
math.log1p(self.clip_epsilon),
)
@property
def out_keys(self):
if self._out_keys is None:
keys = ["loss_objective", "clip_fraction"]
if self.entropy_bonus:
keys.extend(["entropy", "loss_entropy"])
if self.loss_critic:
keys.append("loss_critic")
if self.clip_value:
keys.append("value_clip_fraction")
keys.append("ESS")
self._out_keys = keys
return self._out_keys
@out_keys.setter
def out_keys(self, values):
self._out_keys = values
[docs] @dispatch
def forward(self, tensordict: TensorDictBase) -> TensorDictBase:
tensordict = tensordict.clone(False)
advantage = tensordict.get(self.tensor_keys.advantage, None)
if advantage is None:
self.value_estimator(
tensordict,
params=self._cached_critic_network_params_detached,
target_params=self.target_critic_network_params,
)
advantage = tensordict.get(self.tensor_keys.advantage)
if self.normalize_advantage and advantage.numel() > 1:
loc = advantage.mean()
scale = advantage.std().clamp_min(1e-6)
advantage = (advantage - loc) / scale
log_weight, dist, kl_approx = self._log_weight(tensordict)
# ESS for logging
with torch.no_grad():
# In theory, ESS should be computed on particles sampled from the same source. Here we sample according
# to different, unrelated trajectories, which is not standard. Still it can give a idea of the dispersion
# of the weights.
lw = log_weight.squeeze()
ess = (2 * lw.logsumexp(0) - (2 * lw).logsumexp(0)).exp()
batch = log_weight.shape[0]
gain1 = log_weight.exp() * advantage
log_weight_clip = log_weight.clamp(*self._clip_bounds)
clip_fraction = (log_weight_clip != log_weight).to(log_weight.dtype).mean()
ratio = log_weight_clip.exp()
gain2 = ratio * advantage
gain = torch.stack([gain1, gain2], -1).min(dim=-1)[0]
td_out = TensorDict({"loss_objective": -gain}, batch_size=[])
td_out.set("clip_fraction", clip_fraction)
if self.entropy_bonus:
entropy = self.get_entropy_bonus(dist)
td_out.set("entropy", entropy.detach().mean()) # for logging
td_out.set("kl_approx", kl_approx.detach().mean()) # for logging
td_out.set("loss_entropy", -self.entropy_coef * entropy)
if self.critic_coef:
loss_critic, value_clip_fraction = self.loss_critic(tensordict)
td_out.set("loss_critic", loss_critic)
if value_clip_fraction is not None:
td_out.set("value_clip_fraction", value_clip_fraction)
td_out.set("ESS", _reduce(ess, self.reduction) / batch)
td_out = td_out.named_apply(
lambda name, value: _reduce(value, reduction=self.reduction).squeeze(-1)
if name.startswith("loss_")
else value,
batch_size=[],
)
return td_out
[docs]class KLPENPPOLoss(PPOLoss):
"""KL Penalty PPO loss.
The KL penalty loss has the following formula:
loss = loss - beta * KL(old_policy, new_policy)
The "beta" parameter is adapted on-the-fly to match a target KL divergence between the new and old policy, thus
favouring a certain level of distancing between the two while still preventing them to be too much apart.
Args:
actor_network (ProbabilisticTensorDictSequential): policy operator.
critic_network (ValueOperator): value operator.
Keyword Args:
dtarg (scalar, optional): target KL divergence. Defaults to ``0.01``.
samples_mc_kl (int, optional): number of samples used to compute the KL divergence
if no analytical formula can be found. Defaults to ``1``.
beta (scalar, optional): initial KL divergence multiplier.
Defaults to ``1.0``.
decrement (scalar, optional): how much beta should be decremented if KL < dtarg. Valid range: decrement <= 1.0
default: ``0.5``.
increment (scalar, optional): how much beta should be incremented if KL > dtarg. Valid range: increment >= 1.0
default: ``2.0``.
entropy_bonus (bool, optional): if ``True``, an entropy bonus will be added to the
loss to favour exploratory policies. Defaults to ``True``.
samples_mc_entropy (int, optional): if the distribution retrieved from the policy
operator does not have a closed form
formula for the entropy, a Monte-Carlo estimate will be used.
``samples_mc_entropy`` will control how many
samples will be used to compute this estimate.
Defaults to ``1``.
entropy_coef (scalar, optional): entropy multiplier when computing the total loss.
Defaults to ``0.01``.
critic_coef (scalar, optional): critic loss multiplier when computing the total
loss. Defaults to ``1.0``.
loss_critic_type (str, optional): loss function for the value discrepancy.
Can be one of "l1", "l2" or "smooth_l1". Defaults to ``"smooth_l1"``.
normalize_advantage (bool, optional): if ``True``, the advantage will be normalized
before being used. Defaults to ``False``.
separate_losses (bool, optional): if ``True``, shared parameters between
policy and critic will only be trained on the policy loss.
Defaults to ``False``, i.e., gradients are propagated to shared
parameters for both policy and critic losses.
advantage_key (str, optional): [Deprecated, use set_keys(advantage_key=advantage_key) instead]
The input tensordict key where the advantage is
expected to be written. Defaults to ``"advantage"``.
value_target_key (str, optional): [Deprecated, use set_keys(value_target_key=value_target_key) instead]
The input tensordict key where the target state
value is expected to be written. Defaults to ``"value_target"``.
value_key (str, optional): [Deprecated, use set_keys(value_key) instead]
The input tensordict key where the state
value is expected to be written. Defaults to ``"state_value"``.
functional (bool, optional): whether modules should be functionalized.
Functionalizing permits features like meta-RL, but makes it
impossible to use distributed models (DDP, FSDP, ...) and comes
with a little cost. Defaults to ``True``.
reduction (str, optional): Specifies the reduction to apply to the output:
``"none"`` | ``"mean"`` | ``"sum"``. ``"none"``: no reduction will be applied,
``"mean"``: the sum of the output will be divided by the number of
elements in the output, ``"sum"``: the output will be summed. Default: ``"mean"``.
clip_value (float, optional): If provided, it will be used to compute a clipped version of the value
prediction with respect to the input tensordict value estimate and use it to calculate the value loss.
The purpose of clipping is to limit the impact of extreme value predictions, helping stabilize training
and preventing large updates. However, it will have no impact if the value estimate was done by the current
version of the value estimator. Defaults to ``None``.
.. note:
The advantage (typically GAE) can be computed by the loss function or
in the training loop. The latter option is usually preferred, but this is
up to the user to choose which option is to be preferred.
If the advantage key (``"advantage`` by default) is not present in the
input tensordict, the advantage will be computed by the :meth:`~.forward`
method.
>>> ppo_loss = KLPENPPOLoss(actor, critic)
>>> advantage = GAE(critic)
>>> data = next(datacollector)
>>> losses = ppo_loss(data)
>>> # equivalent
>>> advantage(data)
>>> losses = ppo_loss(data)
A custom advantage module can be built using :meth:`~.make_value_estimator`.
The default is :class:`~torchrl.objectives.value.GAE` with hyperparameters
dictated by :func:`~torchrl.objectives.utils.default_value_kwargs`.
>>> ppo_loss = KLPENPPOLoss(actor, critic)
>>> ppo_loss.make_value_estimator(ValueEstimators.TDLambda)
>>> data = next(datacollector)
>>> losses = ppo_loss(data)
.. note::
If the actor and the value function share parameters, one can avoid
calling the common module multiple times by passing only the head of the
value network to the PPO loss module:
>>> common = SomeModule(in_keys=["observation"], out_keys=["hidden"])
>>> actor_head = SomeActor(in_keys=["hidden"])
>>> value_head = SomeValue(in_keys=["hidden"])
>>> # first option, with 2 calls on the common module
>>> model = ActorCriticOperator(common, actor_head, value_head)
>>> loss_module = PPOLoss(model.get_policy_operator(), model.get_value_operator())
>>> # second option, with a single call to the common module
>>> loss_module = PPOLoss(ProbabilisticTensorDictSequential(model, actor_head), value_head)
This will work regardless of whether separate_losses is activated or not.
"""
actor_network: TensorDictModule
critic_network: TensorDictModule
actor_network_params: TensorDictParams
critic_network_params: TensorDictParams
target_actor_network_params: TensorDictParams
target_critic_network_params: TensorDictParams
def __init__(
self,
actor_network: ProbabilisticTensorDictSequential | None = None,
critic_network: TensorDictModule | None = None,
*,
dtarg: float = 0.01,
beta: float = 1.0,
increment: float = 2,
decrement: float = 0.5,
samples_mc_kl: int = 1,
entropy_bonus: bool = True,
samples_mc_entropy: int = 1,
entropy_coef: float = 0.01,
critic_coef: float = 1.0,
loss_critic_type: str = "smooth_l1",
normalize_advantage: bool = False,
gamma: float = None,
separate_losses: bool = False,
reduction: str = None,
clip_value: float | None = None,
**kwargs,
):
super(KLPENPPOLoss, self).__init__(
actor_network,
critic_network,
entropy_bonus=entropy_bonus,
samples_mc_entropy=samples_mc_entropy,
entropy_coef=entropy_coef,
critic_coef=critic_coef,
loss_critic_type=loss_critic_type,
normalize_advantage=normalize_advantage,
gamma=gamma,
separate_losses=separate_losses,
reduction=reduction,
clip_value=clip_value,
**kwargs,
)
self.dtarg = dtarg
self._beta_init = beta
self.register_buffer("beta", torch.tensor(beta))
if increment < 1.0:
raise ValueError(
f"increment should be >= 1.0 in KLPENPPOLoss, got {increment:4.4f}"
)
self.increment = increment
if decrement > 1.0:
raise ValueError(
f"decrement should be <= 1.0 in KLPENPPOLoss, got {decrement:4.4f}"
)
self.decrement = decrement
self.samples_mc_kl = samples_mc_kl
def _set_in_keys(self):
keys = [
self.tensor_keys.action,
self.tensor_keys.sample_log_prob,
("next", self.tensor_keys.reward),
("next", self.tensor_keys.done),
("next", self.tensor_keys.terminated),
*self.actor_network.in_keys,
*[("next", key) for key in self.actor_network.in_keys],
*self.critic_network.in_keys,
]
# Get the parameter keys from the actor dist
actor_dist_module = None
for module in self.actor_network.modules():
# Ideally we should combine them if there is more than one
if isinstance(module, ProbabilisticTensorDictModule):
if actor_dist_module is not None:
raise RuntimeError(
"Actors with one and only one distribution are currently supported "
f"in {type(self).__name__}. If you need to use more than one "
f"distribtuion over the action space please submit an issue "
f"on github."
)
actor_dist_module = module
if actor_dist_module is None:
raise RuntimeError("Could not find the probabilistic module in the actor.")
keys += list(actor_dist_module.in_keys)
self._in_keys = list(set(keys))
@property
def out_keys(self):
if self._out_keys is None:
keys = ["loss_objective", "kl"]
if self.entropy_bonus:
keys.extend(["entropy", "loss_entropy"])
if self.loss_critic:
keys.append("loss_critic")
if self.clip_value:
keys.append("value_clip_fraction")
self._out_keys = keys
return self._out_keys
@out_keys.setter
def out_keys(self, values):
self._out_keys = values
[docs] @dispatch
def forward(self, tensordict: TensorDictBase) -> TensorDict:
tensordict_copy = tensordict.copy()
try:
previous_dist = self.actor_network.build_dist_from_params(tensordict)
except KeyError as err:
raise KeyError(
"The parameters of the distribution were not found. "
f"Make sure they are provided to {type(self).__name__}."
) from err
advantage = tensordict_copy.get(self.tensor_keys.advantage, None)
if advantage is None:
self.value_estimator(
tensordict_copy,
params=self._cached_critic_network_params_detached,
target_params=self.target_critic_network_params,
)
advantage = tensordict_copy.get(self.tensor_keys.advantage)
if self.normalize_advantage and advantage.numel() > 1:
loc = advantage.mean()
scale = advantage.std().clamp_min(1e-6)
advantage = (advantage - loc) / scale
log_weight, dist, kl_approx = self._log_weight(tensordict_copy)
neg_loss = log_weight.exp() * advantage
with self.actor_network_params.to_module(
self.actor_network
) if self.functional else contextlib.nullcontext():
current_dist = self.actor_network.get_dist(tensordict_copy)
try:
kl = torch.distributions.kl.kl_divergence(previous_dist, current_dist)
except NotImplementedError:
x = previous_dist.sample((self.samples_mc_kl,))
kl = (previous_dist.log_prob(x) - current_dist.log_prob(x)).mean(0)
kl = kl.unsqueeze(-1)
neg_loss = neg_loss - self.beta * kl
if kl.mean() > self.dtarg * 1.5:
self.beta.data *= self.increment
elif kl.mean() < self.dtarg / 1.5:
self.beta.data *= self.decrement
td_out = TensorDict(
{
"loss_objective": -neg_loss,
"kl": kl.detach(),
},
batch_size=[],
)
if self.entropy_bonus:
entropy = self.get_entropy_bonus(dist)
td_out.set("entropy", entropy.detach().mean()) # for logging
td_out.set("kl_approx", kl_approx.detach().mean()) # for logging
td_out.set("loss_entropy", -self.entropy_coef * entropy)
if self.critic_coef:
loss_critic, value_clip_fraction = self.loss_critic(tensordict_copy)
td_out.set("loss_critic", loss_critic)
if value_clip_fraction is not None:
td_out.set("value_clip_fraction", value_clip_fraction)
td_out = td_out.named_apply(
lambda name, value: _reduce(value, reduction=self.reduction).squeeze(-1)
if name.startswith("loss_")
else value,
batch_size=[],
)
return td_out
def reset(self) -> None:
self.beta = self._beta_init
