def _compute_loss(self, observations, actions, rewards, observations_n,
dones):
'''Compute the loss of the agents.'''
feed_dict = self.placeholders._make(
[observations, actions, rewards, observations_n, dones])
feed_dict = flatten_map(feed_dict._asdict())
return unflatten_map(self._compute_loss_h(feed_dict))
def _train_step(self,
observations,
actions,
rewards,
observations_n,
dones,
step=None):
'''Train the agents.'''
feed_dict = self.placeholders._make(
[observations, actions, rewards, observations_n, dones])
feed_dict = flatten_map(feed_dict._asdict())
if step and step % 2 == 0:
results = unflatten_map(self._train(feed_dict))
else:
results = unflatten_map(self._train_critic(feed_dict))
return results
def _train_step(self,
observations,
actions,
rewards,
observations_n,
dones,
step=None):
'''Train the agents.'''
feed_dict = self.placeholders._make(
[observations, actions, rewards, observations_n, dones])
feed_dict = flatten_map(feed_dict._asdict())
return unflatten_map(self._train(feed_dict))