zjowowen's picture
init space
079c32c
from typing import List, Dict, Any, Tuple, Union
from collections import namedtuple
import copy
import torch
from ding.torch_utils import Adam, to_device
from ding.rl_utils import v_1step_td_data, v_1step_td_error, get_train_sample
from ding.model import model_wrap
from ding.utils import POLICY_REGISTRY
from ding.utils.data import default_collate, default_decollate
from .base_policy import Policy
from .common_utils import default_preprocess_learn
@POLICY_REGISTRY.register('atoc')
class ATOCPolicy(Policy):
r"""
Overview:
Policy class of ATOC algorithm.
Interface:
__init__, set_setting, __repr__, state_dict_handle
Property:
learn_mode, collect_mode, eval_mode
"""
config = dict(
# (str) RL policy register name (refer to function "POLICY_REGISTRY").
type='atoc',
# (bool) Whether to use cuda for network.
cuda=False,
# (bool) whether use on-policy training pipeline(behaviour policy and training policy are the same)
on_policy=False,
# (bool) Whether use priority(priority sample, IS weight, update priority)
priority=False,
# (bool) Whether use Importance Sampling Weight to correct biased update. If True, priority must be True.
priority_IS_weight=False,
model=dict(
# (bool) Whether to use communication module in ATOC, if not, it is a multi-agent DDPG
communication=True,
# (int) The number of thought size
thought_size=8,
# (int) The number of agent for each communication group
agent_per_group=2,
),
learn=dict(
# (int) Collect n_sample data, update model n_iteration time
update_per_collect=5,
# (int) The number of data for a train iteration
batch_size=64,
# (float) Gradient-descent step size of actor
learning_rate_actor=0.001,
# (float) Gradient-descent step size of critic
learning_rate_critic=0.001,
# ==============================================================
# The following configs is algorithm-specific
# ==============================================================
# (float) Target network update weight, theta * new_w + (1 - theta) * old_w, defaults in [0, 0.1]
target_theta=0.005,
# (float) Discount factor for future reward, defaults int [0, 1]
discount_factor=0.99,
# (bool) Whether to use communication module in ATOC, if not, it is a multi-agent DDPG
communication=True,
# (int) The frequency of actor update, each critic update
actor_update_freq=1,
# (bool) Whether use noise in action output when learning
noise=True,
# (float) The std of noise distribution for target policy smooth
noise_sigma=0.15,
# (float, float) The minimum and maximum value of noise
noise_range=dict(
min=-0.5,
max=0.5,
),
# (bool) Whether to use reward batch norm in the total batch
reward_batch_norm=False,
ignore_done=False,
),
collect=dict(
# (int) Collect n_sample data, update model n_iteration time
# n_sample=64,
# (int) Unroll length of a train iteration(gradient update step)
unroll_len=1,
# ==============================================================
# The following configs is algorithm-specific
# ==============================================================
# (float) The std of noise distribution for exploration
noise_sigma=0.4,
),
eval=dict(),
other=dict(
replay_buffer=dict(
# (int) The max size of replay buffer
replay_buffer_size=100000,
# (int) The max use count of data, if count is bigger than this value, the data will be removed
max_use=10,
),
),
)
def default_model(self) -> Tuple[str, List[str]]:
return 'atoc', ['ding.model.template.atoc']
def _init_learn(self) -> None:
r"""
Overview:
Learn mode init method. Called by ``self.__init__``.
Init actor and critic optimizers, algorithm config, main and target models.
"""
self._priority = self._cfg.priority
self._priority_IS_weight = self._cfg.priority_IS_weight
assert not self._priority and not self._priority_IS_weight
# algorithm config
self._communication = self._cfg.learn.communication
self._gamma = self._cfg.learn.discount_factor
self._actor_update_freq = self._cfg.learn.actor_update_freq
# actor and critic optimizer
self._optimizer_actor = Adam(
self._model.actor.parameters(),
lr=self._cfg.learn.learning_rate_actor,
)
self._optimizer_critic = Adam(
self._model.critic.parameters(),
lr=self._cfg.learn.learning_rate_critic,
)
if self._communication:
self._optimizer_actor_attention = Adam(
self._model.actor.attention.parameters(),
lr=self._cfg.learn.learning_rate_actor,
)
self._reward_batch_norm = self._cfg.learn.reward_batch_norm
# main and target models
self._target_model = copy.deepcopy(self._model)
self._target_model = model_wrap(
self._target_model,
wrapper_name='target',
update_type='momentum',
update_kwargs={'theta': self._cfg.learn.target_theta}
)
if self._cfg.learn.noise:
self._target_model = model_wrap(
self._target_model,
wrapper_name='action_noise',
noise_type='gauss',
noise_kwargs={
'mu': 0.0,
'sigma': self._cfg.learn.noise_sigma
},
noise_range=self._cfg.learn.noise_range
)
self._learn_model = model_wrap(self._model, wrapper_name='base')
self._learn_model.reset()
self._target_model.reset()
self._forward_learn_cnt = 0 # count iterations
def _forward_learn(self, data: dict) -> Dict[str, Any]:
r"""
Overview:
Forward and backward function of learn mode.
Arguments:
- data (:obj:`dict`): Dict type data, including at least ['obs', 'action', 'reward', 'next_obs']
Returns:
- info_dict (:obj:`Dict[str, Any]`): Including at least actor and critic lr, different losses.
"""
loss_dict = {}
data = default_preprocess_learn(data, ignore_done=self._cfg.learn.ignore_done, use_nstep=False)
if self._cuda:
data = to_device(data, self._device)
# ====================
# critic learn forward
# ====================
self._learn_model.train()
self._target_model.train()
next_obs = data['next_obs']
reward = data['reward']
if self._reward_batch_norm:
reward = (reward - reward.mean()) / (reward.std() + 1e-8)
# current q value
q_value = self._learn_model.forward(data, mode='compute_critic')['q_value']
# target q value.
with torch.no_grad():
next_action = self._target_model.forward(next_obs, mode='compute_actor')['action']
next_data = {'obs': next_obs, 'action': next_action}
target_q_value = self._target_model.forward(next_data, mode='compute_critic')['q_value']
td_data = v_1step_td_data(q_value.mean(-1), target_q_value.mean(-1), reward, data['done'], data['weight'])
critic_loss, td_error_per_sample = v_1step_td_error(td_data, self._gamma)
loss_dict['critic_loss'] = critic_loss
# ================
# critic update
# ================
self._optimizer_critic.zero_grad()
critic_loss.backward()
self._optimizer_critic.step()
# ===============================
# actor learn forward and update
# ===============================
# actor updates every ``self._actor_update_freq`` iters
if (self._forward_learn_cnt + 1) % self._actor_update_freq == 0:
if self._communication:
output = self._learn_model.forward(data['obs'], mode='compute_actor', get_delta_q=False)
output['delta_q'] = data['delta_q']
attention_loss = self._learn_model.forward(output, mode='optimize_actor_attention')['loss']
loss_dict['attention_loss'] = attention_loss
self._optimizer_actor_attention.zero_grad()
attention_loss.backward()
self._optimizer_actor_attention.step()
output = self._learn_model.forward(data['obs'], mode='compute_actor', get_delta_q=False)
critic_input = {'obs': data['obs'], 'action': output['action']}
actor_loss = -self._learn_model.forward(critic_input, mode='compute_critic')['q_value'].mean()
loss_dict['actor_loss'] = actor_loss
# actor update
self._optimizer_actor.zero_grad()
actor_loss.backward()
self._optimizer_actor.step()
# =============
# after update
# =============
loss_dict['total_loss'] = sum(loss_dict.values())
self._forward_learn_cnt += 1
self._target_model.update(self._learn_model.state_dict())
return {
'cur_lr_actor': self._optimizer_actor.defaults['lr'],
'cur_lr_critic': self._optimizer_critic.defaults['lr'],
'priority': td_error_per_sample.abs().tolist(),
'q_value': q_value.mean().item(),
**loss_dict,
}
def _state_dict_learn(self) -> Dict[str, Any]:
return {
'model': self._learn_model.state_dict(),
'target_model': self._target_model.state_dict(),
'optimizer_actor': self._optimizer_actor.state_dict(),
'optimizer_critic': self._optimizer_critic.state_dict(),
'optimize_actor_attention': self._optimizer_actor_attention.state_dict(),
}
def _load_state_dict_learn(self, state_dict: Dict[str, Any]) -> None:
self._learn_model.load_state_dict(state_dict['model'])
self._target_model.load_state_dict(state_dict['target_model'])
self._optimizer_actor.load_state_dict(state_dict['optimizer_actor'])
self._optimizer_critic.load_state_dict(state_dict['optimizer_critic'])
self._optimizer_actor_attention.load_state_dict(state_dict['optimize_actor_attention'])
def _init_collect(self) -> None:
r"""
Overview:
Collect mode init method. Called by ``self.__init__``.
Init traj and unroll length, collect model.
"""
self._unroll_len = self._cfg.collect.unroll_len
# collect model
self._collect_model = model_wrap(
self._model,
wrapper_name='action_noise',
noise_type='gauss',
noise_kwargs={
'mu': 0.0,
'sigma': self._cfg.collect.noise_sigma
},
noise_range=None, # no noise clip in actor
)
self._collect_model.reset()
def _forward_collect(self, data: dict) -> dict:
r"""
Overview:
Forward function of collect mode.
Arguments:
- data (:obj:`Dict[str, Any]`): Dict type data, stacked env data for predicting policy_output(action), \
values are torch.Tensor or np.ndarray or dict/list combinations, keys are env_id indicated by integer.
Returns:
- output (:obj:`Dict[int, Any]`): Dict type data, including at least inferred action according to input obs.
ReturnsKeys
- necessary: ``action``
"""
data_id = list(data.keys())
data = default_collate(list(data.values()))
if self._cuda:
data = to_device(data, self._device)
self._collect_model.eval()
with torch.no_grad():
output = self._collect_model.forward(data, mode='compute_actor', get_delta_q=True)
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[str, Any]:
r"""
Overview:
Generate dict type transition data from inputs.
Arguments:
- obs (:obj:`Any`): Env observation
- model_output (:obj:`dict`): Output of collect model, including at least ['action']
- timestep (:obj:`namedtuple`): Output after env step, including at least ['obs', 'reward', 'done'] \
(here 'obs' indicates obs after env step, i.e. next_obs).
Return:
- transition (:obj:`Dict[str, Any]`): Dict type transition data.
"""
if self._communication:
transition = {
'obs': obs,
'next_obs': timestep.obs,
'action': model_output['action'],
'delta_q': model_output['delta_q'],
'reward': timestep.reward,
'done': timestep.done,
}
else:
transition = {
'obs': obs,
'next_obs': timestep.obs,
'action': model_output['action'],
'reward': timestep.reward,
'done': timestep.done,
}
return transition
def _get_train_sample(self, data: list) -> Union[None, List[Any]]:
if self._communication:
delta_q_batch = [d['delta_q'] for d in data]
delta_min = torch.stack(delta_q_batch).min()
delta_max = torch.stack(delta_q_batch).max()
for i in range(len(data)):
data[i]['delta_q'] = (data[i]['delta_q'] - delta_min) / (delta_max - delta_min + 1e-8)
return get_train_sample(data, self._unroll_len)
def _init_eval(self) -> None:
r"""
Overview:
Evaluate mode init method. Called by ``self.__init__``.
Init eval model. Unlike learn and collect model, eval model does not need noise.
"""
self._eval_model = model_wrap(self._model, wrapper_name='base')
self._eval_model.reset()
def _forward_eval(self, data: dict) -> dict:
r"""
Overview:
Forward function of eval mode, similar to ``self._forward_collect``.
Arguments:
- data (:obj:`Dict[str, Any]`): Dict type data, stacked env data for predicting policy_output(action), \
values are torch.Tensor or np.ndarray or dict/list combinations, keys are env_id indicated by integer.
Returns:
- output (:obj:`Dict[int, Any]`): The dict of predicting action for the interaction with env.
ReturnsKeys
- necessary: ``action``
"""
data_id = list(data.keys())
data = default_collate(list(data.values()))
if self._cuda:
data = to_device(data, self._device)
self._eval_model.eval()
with torch.no_grad():
output = self._eval_model.forward(data, mode='compute_actor')
if self._cuda:
output = to_device(output, 'cpu')
output = default_decollate(output)
return {i: d for i, d in zip(data_id, output)}
def _monitor_vars_learn(self) -> List[str]:
r"""
Overview:
Return variables' name if variables are to used in monitor.
Returns:
- vars (:obj:`List[str]`): Variables' name list.
"""
return [
'cur_lr_actor',
'cur_lr_critic',
'critic_loss',
'actor_loss',
'attention_loss',
'total_loss',
'q_value',
]