"""
MFAC: Mean Field Actor-Critic
Paper link:
http://proceedings.mlr.press/v80/yang18d/yang18d.pdf
Implementation: TensorFlow 2.X
"""
from argparse import Namespace
from operator import itemgetter
from xuance.common import Optional, List
from xuance.tensorflow import tf, tk, Module
from xuance.tensorflow.learners.multi_agent_rl.ippo_learner import IPPO_Learner
[docs]
class MFAC_Learner(IPPO_Learner):
def __init__(self,
config: Namespace,
model_keys: List[str],
agent_keys: List[str],
policy: Module,
callback):
super(MFAC_Learner, self).__init__(config, model_keys, agent_keys, policy, callback)
# @tf.function
[docs]
def forward_fn(self, *args):
bs, obs, actions, act_mean, values, returns, advantages, log_pi_old, agent_mask, avail_actions, IDs = args
info_train, gradients = {}, {}
with tf.GradientTape() as tape:
# feedforward
_, pi_logits_dict = self.policy(observation=obs, agent_ids=IDs, avail_actions=avail_actions)
_, values_pred_dict = self.policy.get_values(observation=obs, actions_mean=act_mean, agent_ids=IDs)
loss_a, loss_e, loss_c = [], [], []
for key in self.model_keys:
mask_values = agent_mask[key]
mask_values_sum = tf.reduce_sum(mask_values)
# actor loss
pi_logits = pi_logits_dict[key] / self.policy.temperature
log_pi = tf.reshape(tf.gather(tf.nn.log_softmax(pi_logits, axis=-1), actions[key],
axis=-1, batch_dims=-1), [bs])
ratio = tf.exp(log_pi - log_pi_old[key])
advantages_mask = tf.stop_gradient(advantages[key] * mask_values)
surrogate1 = ratio * advantages_mask
surrogate2 = tf.clip_by_value(ratio, 1 - self.clip_range, 1 + self.clip_range) * advantages_mask
pg_loss = -tf.reduce_sum(tf.minimum(surrogate1, surrogate2)) / mask_values_sum
loss_a.append(pg_loss)
# entropy loss
probs = tf.exp(log_pi)
entropy = -tf.reduce_sum(probs * log_pi, axis=-1, keepdims=False)
entropy_loss = tf.reduce_sum(entropy * mask_values) / mask_values_sum
loss_e.append(entropy_loss)
# critic 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)
info_train.update({
f"{key}/actor_loss": loss_a[-1],
f"{key}/critic_loss": loss_c[-1],
f"{key}/entropy": loss_e[-1],
f"{key}/predict_value": tf.reduce_mean(value_pred_i),
})
loss = sum(loss_a) + self.vf_coef * sum(loss_c) - self.ent_coef * sum(loss_e)
gradients[key] = tape.gradient(loss, self.policy.trainable_variables)
if self.use_grad_clip:
gradients[key], _ = tf.clip_by_global_norm(gradients[key], clip_norm=self.grad_clip_norm)
self.optimizer.apply_gradients(zip(gradients[key], self.policy.trainable_variables))
else:
self.optimizer.apply_gradients(zip(gradients[key], self.policy.trainable_variables))
info_train.update({
"loss": loss,
})
return info_train
# @tf.function
[docs]
def learn(self, *inputs):
if self.distributed_training:
info_train = self.policy.mirrored_strategy.run(self.forward_fn, args=inputs)
return info_train[0]
else:
return self.forward_fn(*inputs)
[docs]
def update(self, sample):
self.iterations += 1
# prepare training data
act_mean = self.build_actions_mean_input(sample=sample,
use_parameter_sharing=self.use_parameter_sharing)
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']
log_pi_old = sample_Tensor['log_pi_old']
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", actions_mean=act_mean,
policy=self.policy, sample_Tensor=sample_Tensor, bs=bs)
info_train = self.learn(bs, obs, actions, act_mean, values, returns, advantages, log_pi_old,
agent_mask, avail_actions, IDs)
for k, v in info_train.items():
info_train[k] = v.numpy()
info.update(info_train)
info.update(self.callback.on_update_end(self.iterations, method="update", policy=self.policy, info=info))
return info