def step(self) -> None:
# Run the learning step.
fetches = self._step()
if self._require_loggers:
self._create_loggers(list(fetches.keys()))
self._require_loggers = False
# compute statistics
self._compute_statistics(fetches)
# Compute elapsed time.
timestamp = time.time()
elapsed_time = timestamp - self._timestamp if self._timestamp else 0
self._timestamp: float = timestamp
# Update our counts and record it.
counts = self._counter.increment(steps=1, walltime=elapsed_time)
fetches.update(counts)
if self._system_checkpointer:
train_utils.checkpoint_networks(self._system_checkpointer)
if self._logger:
self._logger.write(fetches)
def step(self) -> None:
"""trainer step to update the parameters of the agents in the system"""
# Run the learning step.
fetches = self._step()
# Compute elapsed time.
timestamp = time.time()
if self._timestamp:
elapsed_time = timestamp - self._timestamp
else:
elapsed_time = 0
self._timestamp = timestamp # type: ignore
# Update our counts and record it.
counts = self._counter.increment(steps=1, walltime=elapsed_time)
fetches.update(counts)
# Checkpoint and attempt to write the logs.
if self._checkpoint:
train_utils.checkpoint_networks(self._system_checkpointer)
# Log and decrement epsilon
epsilon = self.get_epsilon()
fetches["epsilon"] = epsilon
self._decrement_epsilon()
if self._logger:
self._logger.write(fetches)
def step(self) -> None:
fetches = self._step()
# Compute elapsed time.
timestamp = time.time()
elapsed_time = timestamp - self._timestamp if self._timestamp else 0
self._timestamp: float = timestamp
# Update our counts and record it.
counts = self._counter.increment(steps=1, walltime=elapsed_time)
fetches.update(counts)
if self._system_checkpointer:
train_utils.checkpoint_networks(self._system_checkpointer)
if self._logger:
self._logger.write(fetches)