"""
Deep Deterministic Policy Gradient (DDPG)
Paper link: https://arxiv.org/pdf/1509.02971.pdf
Implementation: TensorFlow2
"""
from argparse import Namespace
from xuance.tensorflow import tf, tk, Module
from xuance.tensorflow.learners import Learner
[docs]
class DDPG_Learner(Learner):
def __init__(self,
config: Namespace,
policy: Module,
callback):
super(DDPG_Learner, self).__init__(config, policy, callback)
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 = {'actor': tk.optimizers.legacy.Adam(config.learning_rate_actor),
'critic': tk.optimizers.legacy.Adam(config.learning_rate_critic)}
else:
self.optimizer = {'actor': tk.optimizers.legacy.Adam(config.learning_rate_actor),
'critic': tk.optimizers.legacy.Adam(config.learning_rate_critic)}
else:
if self.distributed_training:
with self.policy.mirrored_strategy.scope():
self.optimizer = {'actor': tk.optimizers.Adam(config.learning_rate_actor),
'critic': tk.optimizers.Adam(config.learning_rate_critic)}
else:
self.optimizer = {'actor': tk.optimizers.Adam(config.learning_rate_actor),
'critic': tk.optimizers.Adam(config.learning_rate_critic)}
self.tau = config.tau
self.gamma = config.gamma
self.mse_loss = tk.losses.MeanSquaredError()
@tf.function
def actor_forward_fn(self, obs_batch):
with tf.GradientTape() as tape:
policy_q = self.policy.Qpolicy(obs_batch)
p_loss = -tf.reduce_mean(policy_q)
gradients = tape.gradient(p_loss, self.policy.actor_trainable_variables)
if self.use_grad_clip:
gradients, _ = tf.clip_by_global_norm(gradients, clip_norm=self.grad_clip_norm)
self.optimizer['actor'].apply_gradients(zip(gradients, self.policy.actor_trainable_variables))
else:
self.optimizer['actor'].apply_gradients(zip(gradients, self.policy.actor_trainable_variables))
return p_loss
@tf.function
def critic_forward_fn(self, obs_batch, act_batch, next_batch, rew_batch, ter_batch):
with tf.GradientTape() as tape:
action_q = self.policy.Qaction(obs_batch, act_batch)
next_q = self.policy.Qtarget(next_batch)
backup = rew_batch + (1 - ter_batch) * self.gamma * next_q
y_true = tf.reshape(tf.stop_gradient(backup), [-1])
y_pred = tf.reshape(action_q, [-1])
q_loss = self.mse_loss(y_true, y_pred)
gradients = tape.gradient(q_loss, self.policy.critic_trainable_variables)
if self.use_grad_clip:
gradients, _ = tf.clip_by_global_norm(gradients, clip_norm=self.grad_clip_norm)
self.optimizer['critic'].apply_gradients(zip(gradients, self.policy.critic_trainable_variables))
else:
self.optimizer['critic'].apply_gradients(zip(gradients, self.policy.critic_trainable_variables))
return q_loss, action_q
@tf.function
def learn_actor(self, *inputs):
if self.distributed_training:
p_loss = self.policy.mirrored_strategy.run(self.actor_forward_fn, args=inputs)
return self.policy.mirrored_strategy.reduce(tf.distribute.ReduceOp.SUM, p_loss, axis=None)
else:
return self.actor_forward_fn(*inputs)
@tf.function
def learn_critic(self, *inputs):
if self.distributed_training:
q_loss, action_q = self.policy.mirrored_strategy.run(self.critic_forward_fn, args=inputs)
return (self.policy.mirrored_strategy.reduce(tf.distribute.ReduceOp.SUM, q_loss, axis=None),
self.policy.mirrored_strategy.reduce(tf.distribute.ReduceOp.SUM, action_q, axis=None))
else:
return self.critic_forward_fn(*inputs)
[docs]
def update(self, **samples):
self.iterations += 1
obs_batch = samples['obs']
act_batch = samples['actions']
next_batch = samples['obs_next']
rew_batch = samples['rewards']
ter_batch = samples['terminals']
info = self.callback.on_update_start(self.iterations,
policy=self.policy, obs=obs_batch, act=act_batch,
next_obs=next_batch, rew=rew_batch, termination=ter_batch)
# critic update
q_loss, action_q = self.learn_critic(obs_batch, act_batch, next_batch, rew_batch, ter_batch)
# actor update
p_loss = self.learn_actor(obs_batch)
self.policy.soft_update(self.tau)
info.update({
"Qloss": q_loss.numpy(),
"Ploss": p_loss.numpy(),
"Qvalue": tf.reduce_mean(action_q).numpy(),
})
info.update(self.callback.on_update_end(self.iterations, policy=self.policy, info=info,
action_q=action_q, q_loss=q_loss, p_loss=p_loss))
return info