File size: 39,812 Bytes
079c32c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 |
from typing import List, Dict, Any, Tuple, Union
from collections import namedtuple
import torch
import copy
import numpy as np
from torch.distributions import Independent, Normal
from ding.torch_utils import Adam, to_device, to_dtype, unsqueeze, ContrastiveLoss
from ding.rl_utils import happo_data, happo_error, happo_policy_error, happo_policy_data, \
v_nstep_td_data, v_nstep_td_error, get_train_sample, gae, gae_data, happo_error_continuous, \
get_gae
from ding.model import model_wrap
from ding.utils import POLICY_REGISTRY, split_data_generator, RunningMeanStd
from ding.utils.data import default_collate, default_decollate
from .base_policy import Policy
from .common_utils import default_preprocess_learn
@POLICY_REGISTRY.register('happo')
class HAPPOPolicy(Policy):
"""
Overview:
Policy class of on policy version HAPPO algorithm. Paper link: https://arxiv.org/abs/2109.11251.
"""
config = dict(
# (str) RL policy register name (refer to function "POLICY_REGISTRY").
type='happo',
# (bool) Whether to use cuda for network.
cuda=False,
# (bool) Whether the RL algorithm is on-policy or off-policy. (Note: in practice PPO can be off-policy used)
on_policy=True,
# (bool) Whether to use priority(priority sample, IS weight, update priority)
priority=False,
# (bool) Whether to use Importance Sampling Weight to correct biased update due to priority.
# If True, priority must be True.
priority_IS_weight=False,
# (bool) Whether to recompurete advantages in each iteration of on-policy PPO
recompute_adv=True,
# (str) Which kind of action space used in PPOPolicy, ['discrete', 'continuous', 'hybrid']
action_space='discrete',
# (bool) Whether to use nstep return to calculate value target, otherwise, use return = adv + value
nstep_return=False,
# (bool) Whether to enable multi-agent training, i.e.: MAPPO
multi_agent=False,
# (bool) Whether to need policy data in process transition
transition_with_policy_data=True,
learn=dict(
epoch_per_collect=10,
batch_size=64,
learning_rate=3e-4,
# ==============================================================
# The following configs is algorithm-specific
# ==============================================================
# (float) The loss weight of value network, policy network weight is set to 1
value_weight=0.5,
# (float) The loss weight of entropy regularization, policy network weight is set to 1
entropy_weight=0.0,
# (float) PPO clip ratio, defaults to 0.2
clip_ratio=0.2,
# (bool) Whether to use advantage norm in a whole training batch
adv_norm=True,
value_norm=True,
ppo_param_init=True,
grad_clip_type='clip_norm',
grad_clip_value=0.5,
ignore_done=False,
),
collect=dict(
# (int) Only one of [n_sample, n_episode] shoule be set
# n_sample=64,
# (int) Cut trajectories into pieces with length "unroll_len".
unroll_len=1,
# ==============================================================
# The following configs is algorithm-specific
# ==============================================================
# (float) Reward's future discount factor, aka. gamma.
discount_factor=0.99,
# (float) GAE lambda factor for the balance of bias and variance(1-step td and mc)
gae_lambda=0.95,
),
eval=dict(),
)
def _init_learn(self) -> None:
"""
Overview:
Initialize the learn mode of policy, including related attributes and modules. For HAPPO, it mainly \
contains optimizer, algorithm-specific arguments such as loss weight, clip_ratio and recompute_adv. This \
method also executes some special network initializations and prepares running mean/std monitor for value.
This method will be called in ``__init__`` method if ``learn`` field is in ``enable_field``.
.. note::
For the member variables that need to be saved and loaded, please refer to the ``_state_dict_learn`` \
and ``_load_state_dict_learn`` methods.
.. note::
For the member variables that need to be monitored, please refer to the ``_monitor_vars_learn`` method.
.. note::
If you want to set some spacial member variables in ``_init_learn`` method, you'd better name them \
with prefix ``_learn_`` to avoid conflict with other modes, such as ``self._learn_attr1``.
"""
self._priority = self._cfg.priority
self._priority_IS_weight = self._cfg.priority_IS_weight
assert not self._priority and not self._priority_IS_weight, "Priority is not implemented in PPO"
assert self._cfg.action_space in ["continuous", "discrete"]
self._action_space = self._cfg.action_space
if self._cfg.learn.ppo_param_init:
for n, m in self._model.named_modules():
if isinstance(m, torch.nn.Linear):
torch.nn.init.orthogonal_(m.weight)
torch.nn.init.zeros_(m.bias)
if self._action_space in ['continuous']:
# init log sigma
for agent_id in range(self._cfg.agent_num):
# if hasattr(self._model.agent_models[agent_id].actor_head, 'log_sigma_param'):
# torch.nn.init.constant_(self._model.agent_models[agent_id].actor_head.log_sigma_param, 1)
# The above initialization step has been changed to reparameterizationHead.
for m in list(self._model.agent_models[agent_id].critic.modules()) + \
list(self._model.agent_models[agent_id].actor.modules()):
if isinstance(m, torch.nn.Linear):
# orthogonal initialization
torch.nn.init.orthogonal_(m.weight, gain=np.sqrt(2))
torch.nn.init.zeros_(m.bias)
# do last policy layer scaling, this will make initial actions have (close to)
# 0 mean and std, and will help boost performances,
# see https://arxiv.org/abs/2006.05990, Fig.24 for details
for m in self._model.agent_models[agent_id].actor.modules():
if isinstance(m, torch.nn.Linear):
torch.nn.init.zeros_(m.bias)
m.weight.data.copy_(0.01 * m.weight.data)
# Add the actor/critic parameters of each HAVACAgent in HAVAC to the parameter list of actor/critic_optimizer
actor_params = []
critic_params = []
for agent_idx in range(self._model.agent_num):
actor_params.append({'params': self._model.agent_models[agent_idx].actor.parameters()})
critic_params.append({'params': self._model.agent_models[agent_idx].critic.parameters()})
self._actor_optimizer = Adam(
actor_params,
lr=self._cfg.learn.learning_rate,
grad_clip_type=self._cfg.learn.grad_clip_type,
clip_value=self._cfg.learn.grad_clip_value,
# eps = 1e-5,
)
self._critic_optimizer = Adam(
critic_params,
lr=self._cfg.learn.critic_learning_rate,
grad_clip_type=self._cfg.learn.grad_clip_type,
clip_value=self._cfg.learn.grad_clip_value,
# eps = 1e-5,
)
self._learn_model = model_wrap(self._model, wrapper_name='base')
# self._learn_model = model_wrap(
# self._model,
# wrapper_name='hidden_state',
# state_num=self._cfg.learn.batch_size,
# init_fn=lambda: [None for _ in range(self._cfg.model.agent_num)]
# )
# Algorithm config
self._value_weight = self._cfg.learn.value_weight
self._entropy_weight = self._cfg.learn.entropy_weight
self._clip_ratio = self._cfg.learn.clip_ratio
self._adv_norm = self._cfg.learn.adv_norm
self._value_norm = self._cfg.learn.value_norm
if self._value_norm:
self._running_mean_std = RunningMeanStd(epsilon=1e-4, device=self._device)
self._gamma = self._cfg.collect.discount_factor
self._gae_lambda = self._cfg.collect.gae_lambda
self._recompute_adv = self._cfg.recompute_adv
# Main model
self._learn_model.reset()
def prepocess_data_agent(self, data: Dict[str, Any]):
"""
Overview:
Preprocess data for agent dim. This function is used in learn mode. \
It will be called recursively to process nested dict data. \
It will transpose the data with shape (B, agent_num, ...) to (agent_num, B, ...). \
Arguments:
- data (:obj:`dict`): Dict type data, where each element is the data of an agent of dict type.
Returns:
- ret (:obj:`dict`): Dict type data, where each element is the data of an agent of dict type.
"""
ret = {}
for key, value in data.items():
if isinstance(value, dict):
ret[key] = self.prepocess_data_agent(value)
elif isinstance(value, torch.Tensor) and len(value.shape) > 1:
ret[key] = value.transpose(0, 1)
else:
ret[key] = value
return ret
def _forward_learn(self, data: Dict[str, Any]) -> Dict[str, Any]:
"""
Overview:
Forward and backward function of learn mode.
Arguments:
- data (:obj:`dict`): List type data, where each element is the data of an agent of dict type.
Returns:
- info_dict (:obj:`Dict[str, Any]`):
Including current lr, total_loss, policy_loss, value_loss, entropy_loss, \
adv_abs_max, approx_kl, clipfrac
Overview:
Policy forward function of learn mode (training policy and updating parameters). Forward means \
that the policy inputs some training batch data from the replay buffer and then returns the output \
result, including various training information such as loss, clipfrac, approx_kl.
Arguments:
- data (:obj:`List[Dict[int, Any]]`): The input data used for policy forward, including the latest \
collected training samples for on-policy algorithms like HAPPO. For each element in list, the key of \
dict is the name of data items and the value is the corresponding data. Usually, the value is \
torch.Tensor or np.ndarray or there dict/list combinations. In the ``_forward_learn`` method, data \
often need to first be stacked in the batch dimension by some utility functions such as \
``default_preprocess_learn``. \
For HAPPO, each element in list is a dict containing at least the following keys: ``obs``, \
``action``, ``reward``, ``logit``, ``value``, ``done``. Sometimes, it also contains other keys \
such as ``weight``.
Returns:
- return_infos (:obj:`List[Dict[str, Any]]`): The information list that indicated training result, each \
training iteration contains append a information dict into the final list. The list will be precessed \
and recorded in text log and tensorboard. The value of the dict must be python scalar or a list of \
scalars. For the detailed definition of the dict, refer to the code of ``_monitor_vars_learn`` method.
.. tip::
The training procedure of HAPPO is three for loops. The outermost loop trains each agent separately. \
The middle loop trains all the collected training samples with ``epoch_per_collect`` epochs. The inner \
loop splits all the data into different mini-batch with the length of ``batch_size``.
.. note::
The input value can be torch.Tensor or dict/list combinations and current policy supports all of them. \
For the data type that not supported, the main reason is that the corresponding model does not support it. \
You can implement you own model rather than use the default model. For more information, please raise an \
issue in GitHub repo and we will continue to follow up.
.. note::
For more detailed examples, please refer to our unittest for HAPPOPolicy: ``ding.policy.tests.test_happo``.
"""
data = default_preprocess_learn(data, ignore_done=self._cfg.learn.ignore_done, use_nstep=False)
all_data_len = data['obs']['agent_state'].shape[0]
# fator is the ratio of the old and new strategies of the first m-1 agents, initialized to 1.
# Each transition has its own factor. ref: http://arxiv.org/abs/2109.11251
factor = torch.ones(all_data_len, 1) # (B, 1)
if self._cuda:
data = to_device(data, self._device)
factor = to_device(factor, self._device)
# process agent dim
data = self.prepocess_data_agent(data)
# ====================
# PPO forward
# ====================
return_infos = []
self._learn_model.train()
for agent_id in range(self._cfg.agent_num):
agent_data = {}
for key, value in data.items():
if value is not None:
if type(value) is dict:
agent_data[key] = {k: v[agent_id] for k, v in value.items()} # not feasible for rnn
elif len(value.shape) > 1:
agent_data[key] = data[key][agent_id]
else:
agent_data[key] = data[key]
else:
agent_data[key] = data[key]
# update factor
agent_data['factor'] = factor
# calculate old_logits of all data in buffer for later factor
inputs = {
'obs': agent_data['obs'],
# 'actor_prev_state': agent_data['actor_prev_state'],
# 'critic_prev_state': agent_data['critic_prev_state'],
}
old_logits = self._learn_model.forward(agent_id, inputs, mode='compute_actor')['logit']
for epoch in range(self._cfg.learn.epoch_per_collect):
if self._recompute_adv: # calculate new value using the new updated value network
with torch.no_grad():
inputs['obs'] = agent_data['obs']
# value = self._learn_model.forward(agent_id, agent_data['obs'], mode='compute_critic')['value']
value = self._learn_model.forward(agent_id, inputs, mode='compute_critic')['value']
inputs['obs'] = agent_data['next_obs']
next_value = self._learn_model.forward(agent_id, inputs, mode='compute_critic')['value']
if self._value_norm:
value *= self._running_mean_std.std
next_value *= self._running_mean_std.std
traj_flag = agent_data.get('traj_flag', None) # traj_flag indicates termination of trajectory
compute_adv_data = gae_data(
value, next_value, agent_data['reward'], agent_data['done'], traj_flag
)
agent_data['adv'] = gae(compute_adv_data, self._gamma, self._gae_lambda)
unnormalized_returns = value + agent_data['adv']
if self._value_norm:
agent_data['value'] = value / self._running_mean_std.std
agent_data['return'] = unnormalized_returns / self._running_mean_std.std
self._running_mean_std.update(unnormalized_returns.cpu().numpy())
else:
agent_data['value'] = value
agent_data['return'] = unnormalized_returns
else: # don't recompute adv
if self._value_norm:
unnormalized_return = agent_data['adv'] + agent_data['value'] * self._running_mean_std.std
agent_data['return'] = unnormalized_return / self._running_mean_std.std
self._running_mean_std.update(unnormalized_return.cpu().numpy())
else:
agent_data['return'] = agent_data['adv'] + agent_data['value']
for batch in split_data_generator(agent_data, self._cfg.learn.batch_size, shuffle=True):
inputs = {
'obs': batch['obs'],
# 'actor_prev_state': batch['actor_prev_state'],
# 'critic_prev_state': batch['critic_prev_state'],
}
output = self._learn_model.forward(agent_id, inputs, mode='compute_actor_critic')
adv = batch['adv']
if self._adv_norm:
# Normalize advantage in a train_batch
adv = (adv - adv.mean()) / (adv.std() + 1e-8)
# Calculate happo error
if self._action_space == 'continuous':
happo_batch = happo_data(
output['logit'], batch['logit'], batch['action'], output['value'], batch['value'], adv,
batch['return'], batch['weight'], batch['factor']
)
happo_loss, happo_info = happo_error_continuous(happo_batch, self._clip_ratio)
elif self._action_space == 'discrete':
happo_batch = happo_data(
output['logit'], batch['logit'], batch['action'], output['value'], batch['value'], adv,
batch['return'], batch['weight'], batch['factor']
)
happo_loss, happo_info = happo_error(happo_batch, self._clip_ratio)
wv, we = self._value_weight, self._entropy_weight
total_loss = happo_loss.policy_loss + wv * happo_loss.value_loss - we * happo_loss.entropy_loss
# actor update
# critic update
self._actor_optimizer.zero_grad()
self._critic_optimizer.zero_grad()
total_loss.backward()
self._actor_optimizer.step()
self._critic_optimizer.step()
return_info = {
'agent{}_cur_lr'.format(agent_id): self._actor_optimizer.defaults['lr'],
'agent{}_total_loss'.format(agent_id): total_loss.item(),
'agent{}_policy_loss'.format(agent_id): happo_loss.policy_loss.item(),
'agent{}_value_loss'.format(agent_id): happo_loss.value_loss.item(),
'agent{}_entropy_loss'.format(agent_id): happo_loss.entropy_loss.item(),
'agent{}_adv_max'.format(agent_id): adv.max().item(),
'agent{}_adv_mean'.format(agent_id): adv.mean().item(),
'agent{}_value_mean'.format(agent_id): output['value'].mean().item(),
'agent{}_value_max'.format(agent_id): output['value'].max().item(),
'agent{}_approx_kl'.format(agent_id): happo_info.approx_kl,
'agent{}_clipfrac'.format(agent_id): happo_info.clipfrac,
}
if self._action_space == 'continuous':
return_info.update(
{
'agent{}_act'.format(agent_id): batch['action'].float().mean().item(),
'agent{}_mu_mean'.format(agent_id): output['logit']['mu'].mean().item(),
'agent{}_sigma_mean'.format(agent_id): output['logit']['sigma'].mean().item(),
}
)
return_infos.append(return_info)
# calculate the factor
inputs = {
'obs': agent_data['obs'],
# 'actor_prev_state': agent_data['actor_prev_state'],
}
new_logits = self._learn_model.forward(agent_id, inputs, mode='compute_actor')['logit']
if self._cfg.action_space == 'discrete':
dist_new = torch.distributions.categorical.Categorical(logits=new_logits)
dist_old = torch.distributions.categorical.Categorical(logits=old_logits)
elif self._cfg.action_space == 'continuous':
dist_new = Normal(new_logits['mu'], new_logits['sigma'])
dist_old = Normal(old_logits['mu'], old_logits['sigma'])
logp_new = dist_new.log_prob(agent_data['action'])
logp_old = dist_old.log_prob(agent_data['action'])
if len(logp_new.shape) > 1:
# for logp with shape(B, action_shape), we need to calculate the product of all action dimensions.
factor = factor * torch.prod(
torch.exp(logp_new - logp_old), dim=-1
).reshape(all_data_len, 1).detach() # attention the shape
else:
# for logp with shape(B, ), directly calculate factor
factor = factor * torch.exp(logp_new - logp_old).reshape(all_data_len, 1).detach()
return return_infos
def _state_dict_learn(self) -> Dict[str, Any]:
"""
Overview:
Return the state_dict of learn mode optimizer and model.
Returns:
- state_dict (:obj:`Dict[str, Any]`): The dict of current policy learn mode. It contains the \
state_dict of current policy network and optimizer.
"""
return {
'model': self._learn_model.state_dict(),
'actor_optimizer': self._actor_optimizer.state_dict(),
'critic_optimizer': self._critic_optimizer.state_dict(),
}
def _load_state_dict_learn(self, state_dict: Dict[str, Any]) -> None:
"""
Overview:
Load the state_dict of learn mode optimizer and model.
Arguments:
- state_dict (:obj:`Dict[str, Any]`): The dict of policy learn mode. It contains the state_dict \
of current policy network and optimizer.
"""
self._learn_model.load_state_dict(state_dict['model'])
self._actor_optimizer.load_state_dict(state_dict['actor_optimizer'])
self._critic_optimizer.load_state_dict(state_dict['critic_optimizer'])
def _init_collect(self) -> None:
"""
Overview:
Initialize the collect mode of policy, including related attributes and modules. For HAPPO, it contains \
the collect_model to balance the exploration and exploitation (e.g. the multinomial sample mechanism in \
discrete action space), and other algorithm-specific arguments such as unroll_len and gae_lambda.
This method will be called in ``__init__`` method if ``collect`` field is in ``enable_field``.
.. note::
If you want to set some spacial member variables in ``_init_collect`` method, you'd better name them \
with prefix ``_collect_`` to avoid conflict with other modes, such as ``self._collect_attr1``.
.. tip::
Some variables need to initialize independently in different modes, such as gamma and gae_lambda in PPO. \
This design is for the convenience of parallel execution of different policy modes.
"""
self._unroll_len = self._cfg.collect.unroll_len
assert self._cfg.action_space in ["continuous", "discrete"]
self._action_space = self._cfg.action_space
if self._action_space == 'continuous':
self._collect_model = model_wrap(self._model, wrapper_name='reparam_sample')
elif self._action_space == 'discrete':
self._collect_model = model_wrap(self._model, wrapper_name='multinomial_sample')
self._collect_model.reset()
self._gamma = self._cfg.collect.discount_factor
self._gae_lambda = self._cfg.collect.gae_lambda
self._recompute_adv = self._cfg.recompute_adv
def _forward_collect(self, data: Dict[int, Any]) -> dict:
"""
Overview:
Policy forward function of collect mode (collecting training data by interacting with envs). Forward means \
that the policy gets some necessary data (mainly observation) from the envs and then returns the output \
data, such as the action to interact with the envs.
Arguments:
- data (:obj:`Dict[int, Any]`): The input data used for policy forward, including at least the obs. The \
key of the dict is environment id and the value is the corresponding data of the env.
Returns:
- output (:obj:`Dict[int, Any]`): The output data of policy forward, including at least the action and \
other necessary data (action logit and value) for learn mode defined in ``self._process_transition`` \
method. The key of the dict is the same as the input data, i.e. environment id.
.. tip::
If you want to add more tricks on this policy, like temperature factor in multinomial sample, you can pass \
related data as extra keyword arguments of this method.
.. note::
The input value can be torch.Tensor or dict/list combinations and current policy supports all of them. \
For the data type that not supported, the main reason is that the corresponding model does not support it. \
You can implement you own model rather than use the default model. For more information, please raise an \
issue in GitHub repo and we will continue to follow up.
.. note::
For more detailed examples, please refer to our unittest for HAPPOPolicy: ``ding.policy.tests.test_happo``.
"""
data_id = list(data.keys())
data = default_collate(list(data.values()))
if self._cuda:
data = to_device(data, self._device)
data = {k: v.transpose(0, 1) for k, v in data.items()} # not feasible for rnn
self._collect_model.eval()
with torch.no_grad():
outputs = []
for agent_id in range(self._cfg.agent_num):
# output = self._collect_model.forward(agent_id, data, mode='compute_actor_critic')
single_agent_obs = {k: v[agent_id] for k, v in data.items()}
input = {
'obs': single_agent_obs,
}
output = self._collect_model.forward(agent_id, input, mode='compute_actor_critic')
outputs.append(output)
# transfer data from (M, B, N)->(B, M, N)
result = {}
for key in outputs[0].keys():
if isinstance(outputs[0][key], dict):
subkeys = outputs[0][key].keys()
stacked_subvalues = {}
for subkey in subkeys:
stacked_subvalues[subkey] = \
torch.stack([output[key][subkey] for output in outputs], dim=0).transpose(0, 1)
result[key] = stacked_subvalues
else:
# If Value is tensor, stack it directly
if isinstance(outputs[0][key], torch.Tensor):
result[key] = torch.stack([output[key] for output in outputs], dim=0).transpose(0, 1)
else:
# If it is not tensor, assume that it is a non-stackable data type \
# (such as int, float, etc.), and directly retain the original value
result[key] = [output[key] for output in outputs]
output = result
if self._cuda:
output = to_device(output, 'cpu')
output = default_decollate(output)
return {i: d for i, d in zip(data_id, output)}
def _process_transition(self, obs: Any, model_output: dict, timestep: namedtuple) -> dict:
"""
Overview:
Process and pack one timestep transition data into a dict, which can be directly used for training and \
saved in replay buffer. For HAPPO, it contains obs, next_obs, action, reward, done, logit, value.
Arguments:
- obs (:obj:`torch.Tensor`): The env observation of current timestep.
- policy_output (:obj:`Dict[str, torch.Tensor]`): The output of the policy network with the observation \
as input. For PPO, it contains the state value, action and the logit of the action.
- timestep (:obj:`namedtuple`): The execution result namedtuple returned by the environment step method, \
except all the elements have been transformed into tensor data. Usually, it contains the next obs, \
reward, done, info, etc.
Returns:
- transition (:obj:`Dict[str, torch.Tensor]`): The processed transition data of the current timestep.
.. note::
``next_obs`` is used to calculate nstep return when necessary, so we place in into transition by default. \
You can delete this field to save memory occupancy if you do not need nstep return.
"""
transition = {
'obs': obs,
'next_obs': timestep.obs,
'action': model_output['action'],
'logit': model_output['logit'],
'value': model_output['value'],
'reward': timestep.reward,
'done': timestep.done,
}
return transition
def _get_train_sample(self, data: list) -> Union[None, List[Any]]:
"""
Overview:
For a given trajectory (transitions, a list of transition) data, process it into a list of sample that \
can be used for training directly. In HAPPO, a train sample is a processed transition with new computed \
``traj_flag`` and ``adv`` field. This method is usually used in collectors to execute necessary \
RL data preprocessing before training, which can help learner amortize revelant time consumption. \
In addition, you can also implement this method as an identity function and do the data processing \
in ``self._forward_learn`` method.
Arguments:
- transitions (:obj:`List[Dict[str, Any]`): The trajectory data (a list of transition), each element is \
the same format as the return value of ``self._process_transition`` method.
Returns:
- samples (:obj:`List[Dict[str, Any]]`): The processed train samples, each element is the similar format \
as input transitions, but may contain more data for training, such as GAE advantage.
"""
data = to_device(data, self._device)
for transition in data:
transition['traj_flag'] = copy.deepcopy(transition['done'])
data[-1]['traj_flag'] = True
if self._cfg.learn.ignore_done:
data[-1]['done'] = False
if data[-1]['done']:
last_value = torch.zeros_like(data[-1]['value'])
else:
with torch.no_grad():
last_values = []
for agent_id in range(self._cfg.agent_num):
inputs = {'obs': {k: unsqueeze(v[agent_id], 0) for k, v in data[-1]['next_obs'].items()}}
last_value = self._collect_model.forward(agent_id, inputs, mode='compute_actor_critic')['value']
last_values.append(last_value)
last_value = torch.cat(last_values)
if len(last_value.shape) == 2: # multi_agent case:
last_value = last_value.squeeze(0)
if self._value_norm:
last_value *= self._running_mean_std.std
for i in range(len(data)):
data[i]['value'] *= self._running_mean_std.std
data = get_gae(
data,
to_device(last_value, self._device),
gamma=self._gamma,
gae_lambda=self._gae_lambda,
cuda=False,
)
if self._value_norm:
for i in range(len(data)):
data[i]['value'] /= self._running_mean_std.std
# remove next_obs for save memory when not recompute adv
if not self._recompute_adv:
for i in range(len(data)):
data[i].pop('next_obs')
return get_train_sample(data, self._unroll_len)
def _init_eval(self) -> None:
"""
Overview:
Initialize the eval mode of policy, including related attributes and modules. For PPO, it contains the \
eval model to select optimial action (e.g. greedily select action with argmax mechanism in discrete action).
This method will be called in ``__init__`` method if ``eval`` field is in ``enable_field``.
.. note::
If you want to set some spacial member variables in ``_init_eval`` method, you'd better name them \
with prefix ``_eval_`` to avoid conflict with other modes, such as ``self._eval_attr1``.
"""
assert self._cfg.action_space in ["continuous", "discrete"]
self._action_space = self._cfg.action_space
if self._action_space == 'continuous':
self._eval_model = model_wrap(self._model, wrapper_name='deterministic_sample')
elif self._action_space == 'discrete':
self._eval_model = model_wrap(self._model, wrapper_name='argmax_sample')
self._eval_model.reset()
def _forward_eval(self, data: dict) -> dict:
"""
Overview:
Policy forward function of eval mode (evaluation policy performance by interacting with envs). Forward \
means that the policy gets some necessary data (mainly observation) from the envs and then returns the \
action to interact with the envs. ``_forward_eval`` in HAPPO often uses deterministic sample method to \
get actions while ``_forward_collect`` usually uses stochastic sample method for balance exploration and \
exploitation.
Arguments:
- data (:obj:`Dict[int, Any]`): The input data used for policy forward, including at least the obs. The \
key of the dict is environment id and the value is the corresponding data of the env.
Returns:
- output (:obj:`Dict[int, Any]`): The output data of policy forward, including at least the action. The \
key of the dict is the same as the input data, i.e. environment id.
.. note::
The input value can be torch.Tensor or dict/list combinations and current policy supports all of them. \
For the data type that not supported, the main reason is that the corresponding model does not support it. \
You can implement you own model rather than use the default model. For more information, please raise an \
issue in GitHub repo and we will continue to follow up.
.. note::
For more detailed examples, please refer to our unittest for HAPPOPolicy: ``ding.policy.tests.test_happo``.
"""
data_id = list(data.keys())
data = default_collate(list(data.values()))
if self._cuda:
data = to_device(data, self._device)
# transfer data from (B, M, N)->(M, B, N)
data = {k: v.transpose(0, 1) for k, v in data.items()} # not feasible for rnn
self._eval_model.eval()
with torch.no_grad():
outputs = []
for agent_id in range(self._cfg.agent_num):
single_agent_obs = {k: v[agent_id] for k, v in data.items()}
input = {
'obs': single_agent_obs,
}
output = self._eval_model.forward(agent_id, input, mode='compute_actor')
outputs.append(output)
output = self.revert_agent_data(outputs)
if self._cuda:
output = to_device(output, 'cpu')
output = default_decollate(output)
return {i: d for i, d in zip(data_id, output)}
def default_model(self) -> Tuple[str, List[str]]:
"""
Overview:
Return this algorithm default neural network model setting for demonstration. ``__init__`` method will \
automatically call this method to get the default model setting and create model.
Returns:
- model_info (:obj:`Tuple[str, List[str]]`): The registered model name and model's import_names.
.. note::
The user can define and use customized network model but must obey the same inferface definition indicated \
by import_names path. For example about HAPPO, its registered name is ``happo`` and the import_names is \
``ding.model.template.havac``.
"""
return 'havac', ['ding.model.template.havac']
def _monitor_vars_learn(self) -> List[str]:
"""
Overview:
Return the necessary keys for logging the return dict of ``self._forward_learn``. The logger module, such \
as text logger, tensorboard logger, will use these keys to save the corresponding data.
Returns:
- necessary_keys (:obj:`List[str]`): The list of the necessary keys to be logged.
"""
variables = super()._monitor_vars_learn() + [
'policy_loss',
'value_loss',
'entropy_loss',
'adv_max',
'adv_mean',
'approx_kl',
'clipfrac',
'value_max',
'value_mean',
]
if self._action_space == 'continuous':
variables += ['mu_mean', 'sigma_mean', 'sigma_grad', 'act']
prefixes = [f'agent{i}_' for i in range(self._cfg.agent_num)]
variables = [prefix + var for prefix in prefixes for var in variables]
return variables
def revert_agent_data(self, data: list):
"""
Overview:
Revert the data of each agent to the original data format.
Arguments:
- data (:obj:`list`): List type data, where each element is the data of an agent of dict type.
Returns:
- ret (:obj:`dict`): Dict type data, where each element is the data of an agent of dict type.
"""
ret = {}
# Traverse all keys of the first output
for key in data[0].keys():
if isinstance(data[0][key], torch.Tensor):
# If the value corresponding to the current key is tensor, stack N tensors
stacked_tensor = torch.stack([output[key] for output in data], dim=0)
ret[key] = stacked_tensor.transpose(0, 1)
elif isinstance(data[0][key], dict):
# If the value corresponding to the current key is a dictionary, recursively \
# call the function to process the contents inside the dictionary.
ret[key] = self.revert_agent_data([output[key] for output in data])
return ret
|