ray/rllib/algorithms/ddpg/ddpg_torch_policy.py
2022-06-28 15:52:56 +02:00

387 lines
14 KiB
Python

import logging
import gym
from typing import Dict, Tuple, List, Optional, Any, Type
import ray
from ray.rllib.algorithms.dqn.dqn_tf_policy import (
postprocess_nstep_and_prio,
PRIO_WEIGHTS,
)
from ray.rllib.evaluation import Episode
from ray.rllib.models.modelv2 import ModelV2
from ray.rllib.models.torch.torch_action_dist import (
TorchDeterministic,
TorchDirichlet,
TorchDistributionWrapper,
)
from ray.rllib.policy.sample_batch import SampleBatch
from ray.rllib.policy.torch_policy_v2 import TorchPolicyV2
from ray.rllib.utils.annotations import override
from ray.rllib.utils.framework import try_import_torch
from ray.rllib.utils.metrics.learner_info import LEARNER_STATS_KEY
from ray.rllib.utils.numpy import convert_to_numpy
from ray.rllib.utils.spaces.simplex import Simplex
from ray.rllib.utils.torch_utils import (
apply_grad_clipping,
concat_multi_gpu_td_errors,
huber_loss,
l2_loss,
)
from ray.rllib.utils.typing import (
ModelGradients,
TensorType,
AlgorithmConfigDict,
)
from ray.rllib.algorithms.ddpg.utils import make_ddpg_models, validate_spaces
torch, nn = try_import_torch()
logger = logging.getLogger(__name__)
class ComputeTDErrorMixin:
def __init__(self: TorchPolicyV2):
def compute_td_error(
obs_t, act_t, rew_t, obs_tp1, done_mask, importance_weights
):
input_dict = self._lazy_tensor_dict(
SampleBatch(
{
SampleBatch.CUR_OBS: obs_t,
SampleBatch.ACTIONS: act_t,
SampleBatch.REWARDS: rew_t,
SampleBatch.NEXT_OBS: obs_tp1,
SampleBatch.DONES: done_mask,
PRIO_WEIGHTS: importance_weights,
}
)
)
# Do forward pass on loss to update td errors attribute
# (one TD-error value per item in batch to update PR weights).
self.loss(self.model, None, input_dict)
# `self.model.td_error` is set within actor_critic_loss call.
return self.model.tower_stats["td_error"]
self.compute_td_error = compute_td_error
class TargetNetworkMixin:
"""Mixin class adding a method for (soft) target net(s) synchronizations.
- Adds the `update_target` method to the policy.
Calling `update_target` updates all target Q-networks' weights from their
respective "main" Q-metworks, based on tau (smooth, partial updating).
"""
def __init__(self):
# Hard initial update from Q-net(s) to target Q-net(s).
self.update_target(tau=1.0)
def update_target(self: TorchPolicyV2, tau=None):
# Update_target_fn will be called periodically to copy Q network to
# target Q network, using (soft) tau-synching.
tau = tau or self.config.get("tau")
model_state_dict = self.model.state_dict()
# Support partial (soft) synching.
# If tau == 1.0: Full sync from Q-model to target Q-model.
target_state_dict = next(iter(self.target_models.values())).state_dict()
model_state_dict = {
k: tau * model_state_dict[k] + (1 - tau) * v
for k, v in target_state_dict.items()
}
for target in self.target_models.values():
target.load_state_dict(model_state_dict)
@override(TorchPolicyV2)
def set_weights(self: TorchPolicyV2, weights):
# Makes sure that whenever we restore weights for this policy's
# model, we sync the target network (from the main model)
# at the same time.
TorchPolicyV2.set_weights(self, weights)
self.update_target()
class DDPGTorchPolicy(TargetNetworkMixin, ComputeTDErrorMixin, TorchPolicyV2):
def __init__(
self,
observation_space: gym.spaces.Space,
action_space: gym.spaces.Space,
config: AlgorithmConfigDict,
):
config = dict(ray.rllib.algorithms.ddpg.ddpg.DDPGConfig().to_dict(), **config)
# Create global step for counting the number of update operations.
self.global_step = 0
# Validate action space for DDPG
validate_spaces(self, observation_space, action_space)
TorchPolicyV2.__init__(
self,
observation_space,
action_space,
config,
max_seq_len=config["model"]["max_seq_len"],
)
ComputeTDErrorMixin.__init__(self)
# TODO: Don't require users to call this manually.
self._initialize_loss_from_dummy_batch()
TargetNetworkMixin.__init__(self)
@override(TorchPolicyV2)
def make_model_and_action_dist(
self,
) -> Tuple[ModelV2, Type[TorchDistributionWrapper]]:
model = make_ddpg_models(self)
if isinstance(self.action_space, Simplex):
distr_class = TorchDirichlet
else:
distr_class = TorchDeterministic
return model, distr_class
@override(TorchPolicyV2)
def optimizer(
self,
) -> List["torch.optim.Optimizer"]:
"""Create separate optimizers for actor & critic losses."""
# Set epsilons to match tf.keras.optimizers.Adam's epsilon default.
self._actor_optimizer = torch.optim.Adam(
params=self.model.policy_variables(), lr=self.config["actor_lr"], eps=1e-7
)
self._critic_optimizer = torch.optim.Adam(
params=self.model.q_variables(), lr=self.config["critic_lr"], eps=1e-7
)
# Return them in the same order as the respective loss terms are returned.
return [self._actor_optimizer, self._critic_optimizer]
@override(TorchPolicyV2)
def apply_gradients(self, gradients: ModelGradients) -> None:
# For policy gradient, update policy net one time v.s.
# update critic net `policy_delay` time(s).
if self.global_step % self.config["policy_delay"] == 0:
self._actor_optimizer.step()
self._critic_optimizer.step()
# Increment global step & apply ops.
self.global_step += 1
@override(TorchPolicyV2)
def action_distribution_fn(
self,
model: ModelV2,
*,
obs_batch: TensorType,
state_batches: TensorType,
is_training: bool = False,
**kwargs
) -> Tuple[TensorType, type, List[TensorType]]:
model_out, _ = model(
SampleBatch(obs=obs_batch[SampleBatch.CUR_OBS], _is_training=is_training)
)
dist_inputs = model.get_policy_output(model_out)
if isinstance(self.action_space, Simplex):
distr_class = TorchDirichlet
else:
distr_class = TorchDeterministic
return dist_inputs, distr_class, [] # []=state out
@override(TorchPolicyV2)
def postprocess_trajectory(
self,
sample_batch: SampleBatch,
other_agent_batches: Optional[Dict[Any, SampleBatch]] = None,
episode: Optional[Episode] = None,
) -> SampleBatch:
return postprocess_nstep_and_prio(
self, sample_batch, other_agent_batches, episode
)
@override(TorchPolicyV2)
def loss(
self,
model: ModelV2,
dist_class: Type[TorchDistributionWrapper],
train_batch: SampleBatch,
) -> List[TensorType]:
target_model = self.target_models[model]
twin_q = self.config["twin_q"]
gamma = self.config["gamma"]
n_step = self.config["n_step"]
use_huber = self.config["use_huber"]
huber_threshold = self.config["huber_threshold"]
l2_reg = self.config["l2_reg"]
input_dict = SampleBatch(
obs=train_batch[SampleBatch.CUR_OBS], _is_training=True
)
input_dict_next = SampleBatch(
obs=train_batch[SampleBatch.NEXT_OBS], _is_training=True
)
model_out_t, _ = model(input_dict, [], None)
model_out_tp1, _ = model(input_dict_next, [], None)
target_model_out_tp1, _ = target_model(input_dict_next, [], None)
# Policy network evaluation.
# prev_update_ops = set(tf1.get_collection(tf.GraphKeys.UPDATE_OPS))
policy_t = model.get_policy_output(model_out_t)
# policy_batchnorm_update_ops = list(
# set(tf1.get_collection(tf.GraphKeys.UPDATE_OPS)) - prev_update_ops)
policy_tp1 = target_model.get_policy_output(target_model_out_tp1)
# Action outputs.
if self.config["smooth_target_policy"]:
target_noise_clip = self.config["target_noise_clip"]
clipped_normal_sample = torch.clamp(
torch.normal(
mean=torch.zeros(policy_tp1.size()), std=self.config["target_noise"]
).to(policy_tp1.device),
-target_noise_clip,
target_noise_clip,
)
policy_tp1_smoothed = torch.min(
torch.max(
policy_tp1 + clipped_normal_sample,
torch.tensor(
self.action_space.low,
dtype=torch.float32,
device=policy_tp1.device,
),
),
torch.tensor(
self.action_space.high,
dtype=torch.float32,
device=policy_tp1.device,
),
)
else:
# No smoothing, just use deterministic actions.
policy_tp1_smoothed = policy_tp1
# Q-net(s) evaluation.
# prev_update_ops = set(tf1.get_collection(tf.GraphKeys.UPDATE_OPS))
# Q-values for given actions & observations in given current
q_t = model.get_q_values(model_out_t, train_batch[SampleBatch.ACTIONS])
# Q-values for current policy (no noise) in given current state
q_t_det_policy = model.get_q_values(model_out_t, policy_t)
actor_loss = -torch.mean(q_t_det_policy)
if twin_q:
twin_q_t = model.get_twin_q_values(
model_out_t, train_batch[SampleBatch.ACTIONS]
)
# q_batchnorm_update_ops = list(
# set(tf1.get_collection(tf.GraphKeys.UPDATE_OPS)) - prev_update_ops)
# Target q-net(s) evaluation.
q_tp1 = target_model.get_q_values(target_model_out_tp1, policy_tp1_smoothed)
if twin_q:
twin_q_tp1 = target_model.get_twin_q_values(
target_model_out_tp1, policy_tp1_smoothed
)
q_t_selected = torch.squeeze(q_t, axis=len(q_t.shape) - 1)
if twin_q:
twin_q_t_selected = torch.squeeze(twin_q_t, axis=len(q_t.shape) - 1)
q_tp1 = torch.min(q_tp1, twin_q_tp1)
q_tp1_best = torch.squeeze(input=q_tp1, axis=len(q_tp1.shape) - 1)
q_tp1_best_masked = (1.0 - train_batch[SampleBatch.DONES].float()) * q_tp1_best
# Compute RHS of bellman equation.
q_t_selected_target = (
train_batch[SampleBatch.REWARDS] + gamma ** n_step * q_tp1_best_masked
).detach()
# Compute the error (potentially clipped).
if twin_q:
td_error = q_t_selected - q_t_selected_target
twin_td_error = twin_q_t_selected - q_t_selected_target
if use_huber:
errors = huber_loss(td_error, huber_threshold) + huber_loss(
twin_td_error, huber_threshold
)
else:
errors = 0.5 * (
torch.pow(td_error, 2.0) + torch.pow(twin_td_error, 2.0)
)
else:
td_error = q_t_selected - q_t_selected_target
if use_huber:
errors = huber_loss(td_error, huber_threshold)
else:
errors = 0.5 * torch.pow(td_error, 2.0)
critic_loss = torch.mean(train_batch[PRIO_WEIGHTS] * errors)
# Add l2-regularization if required.
if l2_reg is not None:
for name, var in model.policy_variables(as_dict=True).items():
if "bias" not in name:
actor_loss += l2_reg * l2_loss(var)
for name, var in model.q_variables(as_dict=True).items():
if "bias" not in name:
critic_loss += l2_reg * l2_loss(var)
# Model self-supervised losses.
if self.config["use_state_preprocessor"]:
# Expand input_dict in case custom_loss' need them.
input_dict[SampleBatch.ACTIONS] = train_batch[SampleBatch.ACTIONS]
input_dict[SampleBatch.REWARDS] = train_batch[SampleBatch.REWARDS]
input_dict[SampleBatch.DONES] = train_batch[SampleBatch.DONES]
input_dict[SampleBatch.NEXT_OBS] = train_batch[SampleBatch.NEXT_OBS]
[actor_loss, critic_loss] = model.custom_loss(
[actor_loss, critic_loss], input_dict
)
# Store values for stats function in model (tower), such that for
# multi-GPU, we do not override them during the parallel loss phase.
model.tower_stats["q_t"] = q_t
model.tower_stats["actor_loss"] = actor_loss
model.tower_stats["critic_loss"] = critic_loss
# TD-error tensor in final stats
# will be concatenated and retrieved for each individual batch item.
model.tower_stats["td_error"] = td_error
# Return two loss terms (corresponding to the two optimizers, we create).
return [actor_loss, critic_loss]
@override(TorchPolicyV2)
def extra_grad_process(
self, optimizer: torch.optim.Optimizer, loss: TensorType
) -> Dict[str, TensorType]:
# Clip grads if configured.
return apply_grad_clipping(self, optimizer, loss)
@override(TorchPolicyV2)
def extra_compute_grad_fetches(self) -> Dict[str, Any]:
fetches = convert_to_numpy(concat_multi_gpu_td_errors(self))
return dict({LEARNER_STATS_KEY: {}}, **fetches)
@override(TorchPolicyV2)
def stats_fn(self, train_batch: SampleBatch) -> Dict[str, TensorType]:
q_t = torch.stack(self.get_tower_stats("q_t"))
stats = {
"actor_loss": torch.mean(torch.stack(self.get_tower_stats("actor_loss"))),
"critic_loss": torch.mean(torch.stack(self.get_tower_stats("critic_loss"))),
"mean_q": torch.mean(q_t),
"max_q": torch.max(q_t),
"min_q": torch.min(q_t),
}
return convert_to_numpy(stats)