"""
Independent Advantage Actor Critic (IAC)
Paper link: https://ojs.aaai.org/index.php/AAAI/article/view/11794
Implementation: TensorFlow2
"""
import numpy as np
from argparse import Namespace
from operator import itemgetter
from xuance.common import Optional, List
from xuance.tensorflow import tf, tk, Module
from xuance.tensorflow.utils import ValueNorm
from xuance.tensorflow.learners import LearnerMAS
[docs]
class IAC_Learner(LearnerMAS):
def __init__(self,
config: Namespace,
model_keys: List[str],
agent_keys: List[str],
policy: Module,
callback):
super(IAC_Learner, self).__init__(config, model_keys, agent_keys, policy, callback)
self.build_optimizer()
self.use_value_clip, self.value_clip_range = config.use_value_clip, config.value_clip_range
self.use_huber_loss, self.huber_delta = config.use_huber_loss, config.huber_delta
self.use_value_norm = config.use_value_norm
self.vf_coef, self.ent_coef = config.vf_coef, config.ent_coef
if self.use_value_norm:
self.value_normalizer = {key: ValueNorm(1) for key in self.model_keys}
else:
self.value_normalizer = None
self.is_continuous = self.policy.is_continuous
[docs]
def build_optimizer(self):
if ("macOS" in self.os_name) and ("arm" in self.os_name): # For macOS with Apple's M-series chips.
if self.distributed_training:
with self.policy.mirrored_strategy.scope():
self.optimizer = tk.optimizers.legacy.Adam(self.config.learning_rate)
else:
self.optimizer = tk.optimizers.legacy.Adam(self.config.learning_rate)
else:
if self.distributed_training:
with self.policy.mirrored_strategy.scope():
self.optimizer = tk.optimizers.Adam(self.config.learning_rate)
else:
self.optimizer = tk.optimizers.Adam(self.config.learning_rate)
[docs]
def build_training_data(self, sample: Optional[dict],
use_parameter_sharing: Optional[bool] = False,
use_actions_mask: Optional[bool] = False,
use_global_state: Optional[bool] = False):
"""
Prepare the training data.
Parameters:
sample (dict): The raw sampled data.
use_parameter_sharing (bool): Whether to use parameter sharing for individual agent models.
use_actions_mask (bool): Whether to use actions mask for unavailable actions.
use_global_state (bool): Whether to use global state.
Returns:
sample_Tensor (dict): The formatted sampled data.
"""
batch_size = sample['batch_size']
seq_length = sample['sequence_length'] if self.use_rnn else 1
state, avail_actions, filled, IDs = None, None, None, None
if use_parameter_sharing:
k = self.model_keys[0]
bs = batch_size * self.n_agents
obs_tensor = tf.stack(itemgetter(*self.agent_keys)(sample['obs']), axis=1)
actions_tensor = tf.stack(itemgetter(*self.agent_keys)(sample['actions']), axis=1)
values_tensor = tf.stack(itemgetter(*self.agent_keys)(sample['values']), axis=1)
returns_tensor = tf.stack(itemgetter(*self.agent_keys)(sample['returns']), axis=1)
advantages_tensor = tf.stack(itemgetter(*self.agent_keys)(sample['advantages']), axis=1)
log_pi_old_tensor = tf.stack(itemgetter(*self.agent_keys)(sample['log_pi_old']), axis=1)
ter_tensor = tf.cast(tf.stack(itemgetter(*self.agent_keys)(sample['terminals']), axis=1), dtype=tf.float32)
msk_tensor = tf.cast(tf.stack(itemgetter(*self.agent_keys)(sample['agent_mask']), axis=1), dtype=tf.float32)
if self.use_rnn:
obs = {k: tf.reshape(obs_tensor, [bs, seq_length, -1])}
if len(actions_tensor.shape) == 3:
actions = {k: tf.reshape(actions_tensor, [bs, seq_length])}
elif len(actions_tensor.shape) == 4:
actions = {k: tf.reshape(actions_tensor, [bs, seq_length, -1])}
else:
raise AttributeError("Wrong actions shape.")
values = {k: tf.reshape(values_tensor, [bs, seq_length])}
returns = {k: tf.reshape(returns_tensor, [bs, seq_length])}
advantages = {k: tf.reshape(advantages_tensor, [bs, seq_length])}
log_pi_old = {k: tf.reshape(log_pi_old_tensor, [bs, seq_length])}
terminals = {k: tf.reshape(ter_tensor, [bs, seq_length])}
agent_mask = {k: tf.reshape(msk_tensor, [bs, seq_length])}
IDs = tf.reshape(tf.tile(tf.eye(self.n_agents, dtype=np.float32)[None, :, None, :],
[batch_size, 1, seq_length + 1, 1]), [bs, seq_length + 1, self.n_agents])
else:
obs = {k: tf.reshape(obs_tensor, [bs, -1])}
if self.is_continuous:
actions = {k: tf.reshape(tf.cast(actions_tensor, dtype=tf.float32), [bs, -1])}
else:
actions = {k: tf.reshape(tf.cast(actions_tensor, dtype=tf.int32), [bs, 1])}
values = {k: tf.reshape(values_tensor, [bs])}
returns = {k: tf.reshape(returns_tensor, [bs])}
advantages = {k: tf.reshape(advantages_tensor, [bs])}
log_pi_old = {k: tf.reshape(log_pi_old_tensor, [bs])}
terminals = {k: tf.reshape(ter_tensor, [bs])}
agent_mask = {k: tf.reshape(msk_tensor, [bs])}
IDs = tf.reshape(tf.tile(tf.eye(self.n_agents, dtype=np.float32)[None],
[batch_size, 1, 1]), [bs, self.n_agents])
if use_actions_mask:
avail_a = tf.stack(itemgetter(*self.agent_keys)(sample['avail_actions']), axis=1)
if self.use_rnn:
avail_actions = {k: tf.reshape(avail_a, [bs, seq_length, -1])}
else:
avail_actions = {k: tf.reshape(avail_a, [bs, -1])}
else:
obs = {k: tf.convert_to_tensor(sample['obs'][k], dtype=tf.float32) for k in self.agent_keys}
if self.is_continuous:
actions = {k: tf.convert_to_tensor(sample['actions'][k], dtype=tf.float32) for k in self.agent_keys}
else:
actions = {k: tf.expand_dims(tf.convert_to_tensor(sample['actions'][k], dtype=tf.int32), axis=-1)
for k in self.agent_keys}
values = {k: tf.convert_to_tensor(sample['values'][k], dtype=tf.float32) for k in self.agent_keys}
returns = {k: tf.convert_to_tensor(sample['returns'][k], dtype=tf.float32) for k in self.agent_keys}
advantages = {k: tf.convert_to_tensor(sample['advantages'][k], dtype=tf.float32) for k in self.agent_keys}
log_pi_old = {k: tf.convert_to_tensor(sample['log_pi_old'][k], dtype=tf.float32) for k in self.agent_keys}
terminals = {k: tf.convert_to_tensor(sample['terminals'][k], dtype=tf.float32) for k in self.agent_keys}
agent_mask = {k: tf.convert_to_tensor(sample['agent_mask'][k], dtype=tf.float32) for k in self.agent_keys}
if use_actions_mask:
avail_actions = {k: tf.convert_to_tensor(sample['avail_actions'][k], dtype=tf.float32)
for k in self.agent_keys}
if use_global_state:
state = tf.convert_to_tensor(sample['state'], dtype=tf.float32)
if self.use_rnn:
filled = tf.convert_to_tensor(sample['filled'], dtype=tf.float32)
sample_Tensor = {
'batch_size': batch_size,
'state': state,
'obs': obs,
'actions': actions,
'values': values,
'returns': returns,
'advantages': advantages,
'log_pi_old': log_pi_old,
'terminals': terminals,
'agent_mask': agent_mask,
'avail_actions': avail_actions,
'agent_ids': IDs,
'filled': filled,
'seq_length': seq_length,
}
return sample_Tensor
# @tf.function
[docs]
def forward_fn(self, *args):
bs, obs, actions, agent_mask, avail_actions, values, returns, advantages, IDs = args
with tf.GradientTape() as tape:
loss_a, loss_e, loss_c = [], [], []
if self.is_continuous:
_, pi_mu, pi_std = self.policy(observation=obs, agent_ids=IDs, avail_actions=avail_actions)
for key in self.model_keys:
mask_values = agent_mask[key]
mask_values_sum = tf.reduce_sum(mask_values)
log_2pi = tf.math.log(2.0 * np.pi)
# policy gradient loss
log_std = tf.math.log(pi_std[key] + 1e-8)
log_prob = -0.5 * (((actions[key] - pi_mu[key]) / (pi_std[key] + 1e-8)) ** 2 + 2.0 * log_std + log_2pi)
log_pi = tf.reduce_sum(log_prob, axis=-1, keepdims=False)
pg_loss = -tf.reduce_sum((advantages[key] * log_pi) * mask_values) / mask_values_sum
loss_a.append(pg_loss)
# entropy loss
entropy = tf.reduce_sum(0.5 + 0.5 * log_2pi + log_std, axis=-1, keepdims=True)
entropy_loss = tf.reduce_sum(entropy * mask_values) / mask_values_sum
loss_e.append(entropy_loss)
else:
_, pi_logits = self.policy(observation=obs, agent_ids=IDs, avail_actions=avail_actions)
for key in self.model_keys:
mask_values = agent_mask[key]
mask_values_sum = tf.reduce_sum(mask_values)
# policy gradient loss
log_prob = tf.nn.log_softmax(pi_logits[key], axis=-1)
log_pi = tf.gather(log_prob, actions[key], axis=-1, batch_dims=-1)
log_pi = tf.squeeze(log_pi, axis=-1)
pg_loss = -tf.reduce_sum((advantages[key] * log_pi) * mask_values) / mask_values_sum
loss_a.append(pg_loss)
# entropy loss
probs = tf.exp(log_prob)
entropy = -tf.reduce_sum(probs * log_prob, axis=-1, keepdims=False)
entropy_loss = tf.reduce_sum(entropy * mask_values) / mask_values_sum
loss_e.append(entropy_loss)
_, values_pred_dict = self.policy.get_values(observation=obs, agent_ids=IDs)
for key in self.model_keys:
# value loss
value_pred_i = tf.reshape(values_pred_dict[key], [bs])
value_target = tf.reshape(returns[key], [bs])
values_i = tf.reshape(values[key], [bs])
if self.use_value_clip:
value_clipped = values_i + tf.clip_by_value(value_pred_i - values_i,
-self.value_clip_range, self.value_clip_range)
if self.use_value_norm:
self.value_normalizer[key].update(tf.reshape(value_target, [bs, 1]))
value_target = tf.reshape(self.value_normalizer[key].normalize(tf.reshape(value_target,
[bs, 1])), [bs])
if self.use_huber_loss:
loss_v = tk.losses.huber(value_target, value_pred_i, self.huber_delta)
loss_v_clipped = tk.losses.huber(value_target, value_clipped, self.huber_delta)
else:
loss_v = (value_pred_i - value_target) ** 2
loss_v_clipped = (value_clipped - value_target) ** 2
loss_c_ = tf.maximum(loss_v, loss_v_clipped) * mask_values
loss_c.append(tf.reduce_sum(loss_c_) / mask_values_sum)
else:
if self.use_value_norm:
self.value_normalizer[key].update(value_target)
value_target = self.value_normalizer[key].normalize(value_target)
if self.use_huber_loss:
loss_v = tk.losses.huber(value_target, value_pred_i, self.huber_delta) * mask_values
else:
loss_v = ((value_pred_i - value_target) ** 2) * mask_values
loss_c.append(tf.reduce_sum(loss_v) / mask_values_sum)
# Total loss
loss = sum(loss_a) + self.vf_coef * sum(loss_c) - self.ent_coef * sum(loss_e)
gradients = tape.gradient(loss, self.policy.trainable_variables)
if self.use_grad_clip:
gradients, _ = tf.clip_by_global_norm(gradients, clip_norm=self.grad_clip_norm)
self.optimizer.apply_gradients(zip(gradients, self.policy.trainable_variables))
else:
self.optimizer.apply_gradients(zip(gradients, self.policy.trainable_variables))
return loss, loss_a, loss_c, loss_e, values_pred_dict
# @tf.function
[docs]
def learn(self, *inputs):
if self.distributed_training:
loss, a_loss, c_loss, e_loss, v_pred = self.policy.mirrored_strategy.run(self.forward_fn, args=inputs)
return (self.policy.mirrored_strategy.reduce(tf.distribute.ReduceOp.SUM, loss, axis=None),
self.policy.mirrored_strategy.reduce(tf.distribute.ReduceOp.SUM, a_loss, axis=None),
self.policy.mirrored_strategy.reduce(tf.distribute.ReduceOp.SUM, c_loss, axis=None),
self.policy.mirrored_strategy.reduce(tf.distribute.ReduceOp.SUM, e_loss, axis=None),
self.policy.mirrored_strategy.reduce(tf.distribute.ReduceOp.SUM, v_pred, axis=None))
else:
return self.forward_fn(*inputs)
[docs]
def update(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']
obs = sample_Tensor['obs']
actions = sample_Tensor['actions']
agent_mask = sample_Tensor['agent_mask']
avail_actions = sample_Tensor['avail_actions']
values = sample_Tensor['values']
returns = sample_Tensor['returns']
advantages = sample_Tensor['advantages']
IDs = sample_Tensor['agent_ids']
bs = batch_size * self.n_agents if self.use_parameter_sharing else batch_size
info = self.callback.on_update_start(self.iterations, method="update",
policy=self.policy, sample_Tensor=sample_Tensor, bs=bs)
loss, a_loss, c_loss, e_loss, v_pred = self.learn(bs, obs, actions, agent_mask, avail_actions,
values, returns, advantages, IDs)
info.update({f"predict_value/{key}": tf.reduce_mean(v_pred[key]).numpy() for key in self.model_keys})
info.update({
# "learning_rate": lr,
"pg_loss": sum(a_loss).numpy(),
"vf_loss": sum(c_loss).numpy(),
"entropy_loss": sum(e_loss).numpy(),
"loss": loss.numpy(),
})
info.update(self.callback.on_update_end(self.iterations, method="update", policy=self.policy, info=info))
return info