"""
Independent Deep Deterministic Policy Gradient (IDDPG)
Implementation: TensorFlow 2.X
"""
from argparse import Namespace
from xuance.common import List
from xuance.tensorflow import tf, tk, Module
from xuance.tensorflow.learners import LearnerMAS
[docs]
class IDDPG_Learner(LearnerMAS):
def __init__(self,
config: Namespace,
model_keys: List[str],
agent_keys: List[str],
policy: Module,
callback):
super(IDDPG_Learner, self).__init__(config, model_keys, agent_keys, policy, callback)
self.build_optimizer()
self.gamma = self.config.gamma
self.tau = self.config.tau
[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.
self.optimizer = {
key: {'actor': tk.optimizers.legacy.Adam(self.config.learning_rate_actor),
'critic': tk.optimizers.legacy.Adam(self.config.learning_rate_critic)}
for key in self.model_keys}
else:
self.optimizer = {
key: {'actor': tk.optimizers.Adam(self.config.learning_rate_actor),
'critic': tk.optimizers.Adam(self.config.learning_rate_critic)}
for key in self.model_keys}
@tf.function
def forward_fn(self, bs, obs, actions, rewards, obs_next, terminals, IDs, agent_mask):
info_train = {}
for key in self.model_keys:
# update critic
with tf.GradientTape() as tape:
_, q_eval = self.policy.Qpolicy(observation=obs, actions=actions, agent_ids=IDs, agent_key=key)
_, next_actions = self.policy.Atarget(next_observation=obs_next, agent_ids=IDs, agent_key=key)
_, q_next = self.policy.Qtarget(next_observation=obs_next, next_actions=next_actions, agent_ids=IDs,
agent_key=key)
mask_values = agent_mask[key]
q_eval_a = tf.reshape(q_eval[key], [bs])
q_next_i = tf.reshape(q_next[key], [bs])
q_target = rewards[key] + (1 - terminals[key]) * self.gamma * q_next_i
td_error = (q_eval_a - tf.stop_gradient(q_target)) * mask_values
loss_c = tf.reduce_sum(td_error ** 2) / tf.reduce_sum(mask_values)
gradients = tape.gradient(loss_c, self.policy.critic_trainable_variables(key))
if self.use_grad_clip:
gradients, _ = tf.clip_by_global_norm(gradients, clip_norm=self.grad_clip_norm)
self.optimizer[key]['critic'].apply_gradients(zip(gradients,
self.policy.critic_trainable_variables(key)))
else:
self.optimizer[key]['critic'].apply_gradients(zip(gradients,
self.policy.critic_trainable_variables(key)))
info_train.update({f"{key}/loss_critic": loss_c,
f"{key}/predictQ": tf.math.reduce_mean(q_eval[key])})
# update actor
with tf.GradientTape() as tape:
_, actions_eval = self.policy(observation=obs, agent_ids=IDs, agent_key=key)
_, q_policy = self.policy.Qpolicy(observation=obs, actions=actions_eval, agent_ids=IDs, agent_key=key)
mask_values = agent_mask[key]
q_policy_i = tf.reshape(q_policy[key], [bs])
loss_a = -tf.reduce_sum(q_policy_i * mask_values) / tf.reduce_sum(mask_values)
gradients = tape.gradient(loss_a, self.policy.actor_trainable_variables(key))
if self.use_grad_clip:
gradients, _ = tf.clip_by_global_norm(gradients, clip_norm=self.grad_clip_norm)
self.optimizer[key]['actor'].apply_gradients(zip(gradients,
self.policy.actor_trainable_variables(key)))
else:
self.optimizer[key]['actor'].apply_gradients(zip(gradients,
self.policy.actor_trainable_variables(key)))
info_train.update({f"{key}/loss_actor": loss_a})
info_train.update(self.callback.on_update_agent_wise(self.iterations, key, info=info_train, method="update",
mask_values=mask_values, q_policy_i=q_policy_i,
q_eval_a=q_eval_a, q_next_i=q_next_i,
q_target=q_target, td_error=td_error))
self.policy.soft_update(self.tau)
info_train.update(self.callback.on_update_end(self.iterations,
method="update", policy=self.policy, info=info_train))
return info_train
@tf.function
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.
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
rewards[key] = tf.reshape(rewards[key], [batch_size * self.n_agents])
terminals[key] = tf.reshape(terminals[key], [batch_size * self.n_agents])
else:
bs = batch_size
info = self.callback.on_update_start(self.iterations, method="update",
policy=self.policy, sample_Tensor=sample_Tensor, bs=bs)
info_train = self.learn(bs, obs, actions, rewards, obs_next, terminals, IDs, agent_mask)
for k, v in info_train.items():
info_train[k] = v.numpy()
info.update(info_train)
return info