wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-08-17 11:25:34 +08:00
Code Issues Packages Projects Releases Wiki Activity

[rllib] Better document which methods are abstract and which ones are overrides (#3480)

Browse Source
This commit is contained in:
Eric Liang
2018-12-08 16:28:58 -08:00
committed by GitHub
parent 462e6ef066
commit 8b5827b9da
40 changed files with 1385 additions and 1167 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/rllib/optimizers/sync_replay_optimizer.py
+18 -14
View File
@@ -11,6 +11,7 @@ from ray.rllib.optimizers.replay_buffer import ReplayBuffer, \
from ray.rllib.optimizers.policy_optimizer import PolicyOptimizer
from ray.rllib.evaluation.sample_batch import SampleBatch, DEFAULT_POLICY_ID, \
MultiAgentBatch
from ray.rllib.utils.annotations import override
from ray.rllib.utils.compression import pack_if_needed
from ray.rllib.utils.filter import RunningStat
from ray.rllib.utils.timer import TimerStat
@@ -24,6 +25,7 @@ class SyncReplayOptimizer(PolicyOptimizer):
"td_error" array in the info return of compute_gradients(). This error
term will be used for sample prioritization."""
@override(PolicyOptimizer)
def _init(self,
learning_starts=1000,
buffer_size=10000,
@@ -70,6 +72,7 @@ class SyncReplayOptimizer(PolicyOptimizer):
assert buffer_size >= self.replay_starts
@override(PolicyOptimizer)
def step(self):
with self.update_weights_timer:
if self.remote_evaluators:
@@ -106,6 +109,21 @@ class SyncReplayOptimizer(PolicyOptimizer):
self.num_steps_sampled += batch.count
@override(PolicyOptimizer)
def stats(self):
return dict(
PolicyOptimizer.stats(self), **{
"sample_time_ms": round(1000 * self.sample_timer.mean, 3),
"replay_time_ms": round(1000 * self.replay_timer.mean, 3),
"grad_time_ms": round(1000 * self.grad_timer.mean, 3),
"update_time_ms": round(1000 * self.update_weights_timer.mean,
3),
"opt_peak_throughput": round(self.grad_timer.mean_throughput,
3),
"opt_samples": round(self.grad_timer.mean_units_processed, 3),
"learner": self.learner_stats,
})
def _optimize(self):
samples = self._replay()
@@ -151,17 +169,3 @@ class SyncReplayOptimizer(PolicyOptimizer):
"batch_indexes": batch_indexes
})
return MultiAgentBatch(samples, self.train_batch_size)
def stats(self):
return dict(
PolicyOptimizer.stats(self), **{
"sample_time_ms": round(1000 * self.sample_timer.mean, 3),
"replay_time_ms": round(1000 * self.replay_timer.mean, 3),
"grad_time_ms": round(1000 * self.grad_timer.mean, 3),
"update_time_ms": round(1000 * self.update_weights_timer.mean,
3),
"opt_peak_throughput": round(self.grad_timer.mean_throughput,
3),
"opt_samples": round(self.grad_timer.mean_units_processed, 3),
"learner": self.learner_stats,
})