"""
Multi-agent Soft Actor-critic (MASAC)
Implementation: Pytorch
"""
import torch
from torch import nn
from xuance.common import List
from argparse import Namespace
from xuance.torch.learners.multi_agent_rl.isac_learner import ISAC_Learner
[docs]
class MASAC_Learner(ISAC_Learner):
def __init__(self,
config: Namespace,
model_keys: List[str],
agent_keys: List[str],
policy: nn.Module,
callback):
super(MASAC_Learner, self).__init__(config, model_keys, agent_keys, policy, callback)
[docs]
def update(self, sample):
self.iterations += 1
# Prepare training data.
sample_Tensor = self.build_training_data(sample,
use_parameter_sharing=self.use_parameter_sharing,
use_actions_mask=False)
batch_size = sample_Tensor['batch_size']
obs = sample_Tensor['obs']
actions = sample_Tensor['actions']
obs_next = sample_Tensor['obs_next']
rewards = sample_Tensor['rewards']
terminals = sample_Tensor['terminals']
agent_mask = sample_Tensor['agent_mask']
IDs = sample_Tensor['agent_ids']
if self.use_parameter_sharing:
key = self.model_keys[0]
bs = batch_size * self.n_agents
obs_joint = obs[key].reshape(batch_size, -1)
next_obs_joint = obs_next[key].reshape(batch_size, -1)
actions_joint = actions[key].reshape(batch_size, -1)
rewards[key] = rewards[key].reshape(batch_size * self.n_agents)
terminals[key] = terminals[key].reshape(batch_size * self.n_agents)
else:
bs = batch_size
obs_joint = self.get_joint_input(obs, (batch_size, -1))
next_obs_joint = self.get_joint_input(obs_next, (batch_size, -1))
actions_joint = self.get_joint_input(actions, (batch_size, -1))
info = self.callback.on_update_start(self.iterations, method="update", policy=self.policy,
sample_Tensor=sample_Tensor, bs=bs, obs_joint=obs_joint,
next_obs_joint=next_obs_joint, actions_joint=actions_joint)
# train the model
_, actions_eval, log_pi_eval = self.policy(observation=obs, agent_ids=IDs)
_, actions_next, log_pi_next = self.policy(observation=obs_next, agent_ids=IDs)
if self.use_parameter_sharing:
key = self.model_keys[0]
actions_next_joint = actions_next[key].reshape(batch_size, self.n_agents, -1).reshape(batch_size, -1)
else:
actions_next_joint = self.get_joint_input(actions_next, (batch_size, -1))
_, _, action_q_1, action_q_2 = self.policy.Qpolicy(joint_observation=obs_joint, joint_actions=actions_joint,
agent_ids=IDs)
_, _, target_q = self.policy.Qtarget(joint_observation=next_obs_joint, joint_actions=actions_next_joint,
agent_ids=IDs)
for key in self.model_keys:
mask_values = agent_mask[key]
# critic update
action_q_1_i = action_q_1[key].reshape(bs)
action_q_2_i = action_q_2[key].reshape(bs)
log_pi_next_eval = log_pi_next[key].reshape(bs)
target_value = target_q[key].reshape(bs) - self.alpha[key] * log_pi_next_eval
backup = rewards[key] + (1 - terminals[key]) * self.gamma * target_value
td_error_1, td_error_2 = action_q_1_i - backup.detach(), action_q_2_i - backup.detach()
td_error_1 *= mask_values
td_error_2 *= mask_values
loss_c = ((td_error_1 ** 2).sum() + (td_error_2 ** 2).sum()) / mask_values.sum()
self.optimizer[key]['critic'].zero_grad()
loss_c.backward()
if self.use_grad_clip:
torch.nn.utils.clip_grad_norm_(self.policy.parameters_critic[key], self.grad_clip_norm)
self.optimizer[key]['critic'].step()
if self.scheduler[key]['critic'] is not None:
self.scheduler[key]['critic'].step()
# actor update
if self.use_parameter_sharing:
actions_eval_joint = actions_eval[key].reshape(batch_size, self.n_agents, -1).reshape(batch_size, -1)
else:
actions_eval_detach_others = {k: actions_eval[k] if k == key else actions_eval[k].detach()
for k in self.model_keys}
actions_eval_joint = self.get_joint_input(actions_eval_detach_others, (batch_size, -1))
_, _, policy_q_1, policy_q_2 = self.policy.Qpolicy(joint_observation=obs_joint,
joint_actions=actions_eval_joint,
agent_ids=IDs, agent_key=key)
log_pi_eval_i = log_pi_eval[key].reshape(bs)
policy_q = torch.min(policy_q_1[key], policy_q_2[key]).reshape(bs)
loss_a = ((self.alpha[key] * log_pi_eval_i - policy_q) * mask_values).sum() / mask_values.sum()
self.optimizer[key]['actor'].zero_grad()
loss_a.backward()
if self.use_grad_clip:
torch.nn.utils.clip_grad_norm_(self.policy.parameters_actor[key], self.grad_clip_norm)
self.optimizer[key]['actor'].step()
if self.scheduler[key]['actor'] is not None:
self.scheduler[key]['actor'].step()
# automatic entropy tuning
if self.use_automatic_entropy_tuning:
alpha_loss = -(self.log_alpha[key] * (log_pi_eval_i + self.target_entropy[key]).detach()).mean()
self.alpha_optimizer[key].zero_grad()
alpha_loss.backward()
self.alpha_optimizer[key].step()
self.alpha[key] = self.log_alpha[key].exp()
else:
alpha_loss = 0
learning_rate_actor = self.optimizer[key]['actor'].state_dict()['param_groups'][0]['lr']
learning_rate_critic = self.optimizer[key]['critic'].state_dict()['param_groups'][0]['lr']
info.update({
f"{key}/learning_rate_actor": learning_rate_actor,
f"{key}/learning_rate_critic": learning_rate_critic,
f"{key}/loss_actor": loss_a.item(),
f"{key}/loss_critic": loss_c.item(),
f"{key}/predictQ": policy_q.mean().item(),
})
if self.use_automatic_entropy_tuning:
info.update({f"{key}/alpha_loss": alpha_loss.item(),
f"{key}/alpha": self.alpha[key].item()})
info.update(self.callback.on_update_agent_wise(self.iterations, key, info=info, method="update",
mask_values=mask_values,
action_q_1_i=action_q_1_i, action_q_2_i=action_q_2_i,
log_pi_next_eval=log_pi_next_eval,
target_value=target_value, backup=backup,
td_error_1=td_error_1, td_error_2=td_error_2,
policy_q_1=policy_q_1, policy_q_2=policy_q_2,
log_pi_eval_i=log_pi_eval_i, policy_q=policy_q))
self.policy.soft_update(self.tau)
info.update(self.callback.on_update_end(self.iterations, method="update", policy=self.policy, info=info))
return info
[docs]
def update_rnn(self, sample):
self.iterations += 1
# prepare training data
sample_Tensor = self.build_training_data(sample=sample,
use_parameter_sharing=self.use_parameter_sharing,
use_actions_mask=self.use_actions_mask)
batch_size = sample_Tensor['batch_size']
seq_len = sample_Tensor['seq_length']
obs = sample_Tensor['obs']
actions = sample_Tensor['actions']
rewards = sample_Tensor['rewards']
terminals = sample_Tensor['terminals']
agent_mask = sample_Tensor['agent_mask']
filled = sample_Tensor['filled']
IDs = sample_Tensor['agent_ids']
if self.use_parameter_sharing:
key = self.model_keys[0]
bs_rnn = batch_size * self.n_agents
filled = filled.unsqueeze(1).expand(-1, self.n_agents, -1).reshape(bs_rnn, seq_len)
obs_joint = obs[key].reshape(batch_size, self.n_agents, seq_len + 1, -1).transpose(
1, 2).reshape(batch_size, seq_len + 1, -1)
actions_joint = actions[key].reshape(batch_size, self.n_agents, seq_len, -1).transpose(
1, 2).reshape(batch_size, seq_len, -1)
rewards[key] = rewards[key].reshape(bs_rnn, seq_len)
terminals[key] = terminals[key].reshape(bs_rnn, seq_len)
IDs_t = IDs[:, :-1]
else:
bs_rnn, IDs_t = batch_size, None
obs_joint = self.get_joint_input(obs, (batch_size, seq_len + 1, -1))
actions_joint = self.get_joint_input(actions, (batch_size, seq_len, -1))
info = self.callback.on_update_start(self.iterations, method="update_rnn", policy=self.policy,
sample_Tensor=sample_Tensor, bs_rnn=bs_rnn,
obs_joint=obs_joint, actions_joint=actions_joint)
# initial hidden states for rnn
rnn_hidden_actor = {k: self.policy.actor_representation[k].init_hidden(bs_rnn) for k in self.model_keys}
rnn_hidden_critic = {k: self.policy.critic_1_representation[k].init_hidden(batch_size) for k in self.model_keys}
_, actions_eval, log_pi_eval = self.policy(observation=obs, agent_ids=IDs, rnn_hidden=rnn_hidden_actor)
if self.use_parameter_sharing:
key = self.model_keys[0]
actions_eval_joint = actions_eval[key].reshape(batch_size, self.n_agents, seq_len + 1, -1).transpose(
1, 2).reshape(batch_size, seq_len + 1, -1)
else:
actions_eval_joint = self.get_joint_input(actions_eval, (batch_size, seq_len + 1, -1))
_, _, action_q_1, action_q_2 = self.policy.Qpolicy(joint_observation=obs_joint[:, :-1],
joint_actions=actions_joint,
agent_ids=IDs_t,
rnn_hidden_critic_1=rnn_hidden_critic,
rnn_hidden_critic_2=rnn_hidden_critic)
_, _, target_q = self.policy.Qtarget(joint_observation=obs_joint, joint_actions=actions_eval_joint,
agent_ids=IDs,
rnn_hidden_critic_1=rnn_hidden_critic,
rnn_hidden_critic_2=rnn_hidden_critic)
for key in self.model_keys:
mask_values = agent_mask[key] * filled
# critic update
action_q_1_i = action_q_1[key].reshape(bs_rnn, seq_len)
action_q_2_i = action_q_2[key].reshape(bs_rnn, seq_len)
log_pi_next_eval = log_pi_eval[key][:, 1:].reshape(bs_rnn, seq_len)
target_value = target_q[key][:, 1:].reshape(bs_rnn, seq_len) - self.alpha[key] * log_pi_next_eval
backup = rewards[key] + (1 - terminals[key]) * self.gamma * target_value
td_error_1, td_error_2 = action_q_1_i - backup.detach(), action_q_2_i - backup.detach()
td_error_1 *= mask_values
td_error_2 *= mask_values
loss_c = ((td_error_1 ** 2).sum() + (td_error_2 ** 2).sum()) / mask_values.sum()
self.optimizer[key]['critic'].zero_grad()
loss_c.backward()
if self.use_grad_clip:
torch.nn.utils.clip_grad_norm_(self.policy.parameters_critic[key], self.grad_clip_norm)
self.optimizer[key]['critic'].step()
if self.scheduler[key]['critic'] is not None:
self.scheduler[key]['critic'].step()
# actor update
if self.use_parameter_sharing:
actions_eval_joint = actions_eval_joint[:, :-1]
else:
actions_eval_detach_others = {k: actions_eval[k] if k == key else actions_eval[k].detach()
for k in self.model_keys}
actions_eval_joint = self.get_joint_input(actions_eval_detach_others,
(batch_size, seq_len + 1, -1))[:, :-1]
_, _, policy_q_1, policy_q_2 = self.policy.Qpolicy(joint_observation=obs_joint[:, :-1],
joint_actions=actions_eval_joint,
agent_ids=IDs_t, agent_key=key,
rnn_hidden_critic_1=rnn_hidden_critic,
rnn_hidden_critic_2=rnn_hidden_critic)
log_pi_eval_i = log_pi_eval[key][:, :-1].reshape(bs_rnn, seq_len)
policy_q = torch.min(policy_q_1[key], policy_q_2[key]).reshape(bs_rnn, seq_len)
loss_a = ((self.alpha[key] * log_pi_eval_i - policy_q) * mask_values).sum() / mask_values.sum()
self.optimizer[key]['actor'].zero_grad()
loss_a.backward()
if self.use_grad_clip:
torch.nn.utils.clip_grad_norm_(self.policy.parameters_actor[key], self.grad_clip_norm)
self.optimizer[key]['actor'].step()
if self.scheduler[key]['actor'] is not None:
self.scheduler[key]['actor'].step()
# automatic entropy tuning
if self.use_automatic_entropy_tuning:
alpha_loss = -(self.log_alpha[key] * (log_pi_eval_i + self.target_entropy[key]).detach()).mean()
self.alpha_optimizer[key].zero_grad()
alpha_loss.backward()
self.alpha_optimizer[key].step()
self.alpha[key] = self.log_alpha[key].exp()
else:
alpha_loss = 0
learning_rate_actor = self.optimizer[key]['actor'].state_dict()['param_groups'][0]['lr']
learning_rate_critic = self.optimizer[key]['critic'].state_dict()['param_groups'][0]['lr']
info.update({
f"{key}/learning_rate_actor": learning_rate_actor,
f"{key}/learning_rate_critic": learning_rate_critic,
f"{key}/loss_actor": loss_a.item(),
f"{key}/loss_critic": loss_c.item(),
f"{key}/predictQ": policy_q.mean().item(),
})
if self.use_automatic_entropy_tuning:
info.update({f"{key}/alpha_loss": alpha_loss.item(),
f"{key}/alpha": self.alpha[key].item()})
info.update(self.callback.on_update_agent_wise(self.iterations, key, info=info, method="update_rnn",
mask_values=mask_values,
action_q_1_i=action_q_1_i, action_q_2_i=action_q_2_i,
log_pi_next_eval=log_pi_next_eval,
target_value=target_value, backup=backup,
td_error_1=td_error_1, td_error_2=td_error_2,
policy_q_1=policy_q_1, policy_q_2=policy_q_2,
log_pi_eval_i=log_pi_eval_i, policy_q=policy_q))
self.policy.soft_update(self.tau)
info.update(self.callback.on_update_end(self.iterations, method="update_rnn", policy=self.policy, info=info))
return info