def postprocess_trajectory(self, sample_batch, other_agent_batches=None):
completed = sample_batch["dones"][-1]
if completed:
last_r = 0.0
else:
last_r = self._value(sample_batch["new_obs"][-1])
return compute_advantages(
sample_batch, last_r, self.config["gamma"], self.config["lambda"])
def postprocess_trajectory(self, sample_batch, other_agent_batches=None):
last_r = 0.0
batch = compute_advantages(sample_batch,
last_r,
self.config["gamma"],
self.config["lambda"],
use_gae=self.config["use_gae"])
return batch
def postprocess_trajectory(self, sample_batch, other_agent_batches=None):
completed = sample_batch["dones"][-1]
if completed:
last_r = 0.0
else:
next_state = []
for i in range(len(self.state_in)):
next_state.append([sample_batch["state_out_{}".format(i)][-1]])
last_r = self.value(sample_batch["new_obs"][-1], *next_state)
return compute_advantages(sample_batch, last_r, self.config["gamma"],
self.config["lambda"])
def postprocess_trajectory(self, sample_batch, other_agent_batches=None):
return compute_advantages(sample_batch,
0.0,
self.config["gamma"],
use_gae=False)
def postprocess_trajectory(self, batch, other_agent_batches=None):
return compute_advantages(batch, 100.0, 0.9, use_gae=False)