8000 Commits · tensorflow/agents · GitHub

More Web Proxy on the site http://driver.im/

Commits

Commits on Dec 3, 2021

Add entropy metric in PPO.

summer-yue
authored and
copybara-github
committed

Commits on Nov 19, 2021

Update state checking logic in PPO policy to work with autograph for stateful networks.

summer-yue
authored and
copybara-github
committed

Commits on Oct 13, 2021

Fix Reverb signature for distributed SAC nightly.

summer-yue
authored and
copybara-github
committed

Commits on Oct 7, 2021

Stop reopening the trajectory writer in observer.reset as it's slow.

summer-yue
authored and
copybara-github
committed
Make aggregation with `reduce_mean` optional in PPOAgent instead of aggregating across cores.

summer-yue
authored and
copybara-github
committed

Commits on Oct 6, 2021

Fix TFMetricsTest.

summer-yue
authored and
copybara-github
committed

Commits on Sep 20, 2021

Add testing for Reverb observers using the real reverb client and writer.

summer-yue
authored and
copybara-github
committed

Commits on Sep 17, 2021

Convert history to trajectory column when writing in episodic replay buffer.

summer-yue
authored and
copybara-github
committed

Commits on Sep 1, 2021

Fix bug in PPO advantage normalization.

summer-yue
authored and
copybara-github
committed

Commits on Aug 27, 2021

Make env_name a FLAG for distributed examples instead of gin configured.

summer-yue
authored and
copybara-github
committed

Commits on Aug 25, 2021

Add metric aggregator to TFA example for combining distribution evaluation metrics.

summer-yue
authored and
copybara-github
committed

Commits on Aug 10, 2021

Update README for using the new checkpoint based evaluator (as opposed to Reverb VC based).

summer-yue
authored and
copybara-github
committed
Set up checkpoint based eval job (instead of relying on VC)

summer-yue
authored and
copybara-github
committed

Commits on Aug 2, 2021

Skip interleaving dataset optimizations when cycle_length==1 in reverb RB.

summer-yue
authored and
copybara-github
committed

Commits on Jul 16, 2021

Add note on caveat of episodic observer.

summer-yue
authored and
copybara-github
committed

Commits on Jun 3, 2021

Explain how the number of training steps is calculated in PPO README.

summer-yue
authored and
copybara-github
committed

Commits on May 25, 2021

Better error handling in PPOLearner.

summer-yue
authored and
copybara-github
committed

Commits on May 24, 2021

Rename PPOLearner.num_batches to num_samples to de-confuse with the number of mini batches use with training.

summer-yue
authored and
copybara-github
committed

Commits on May 21, 2021

Adjust reverb signature in SAC example to be compatible with new Reverb writer.

summer-yue
authored and
copybara-github
committed

Commits on May 18, 2021

[TF-Agents] Avoid setting tf.config.set_device_placement if it's already set.

summer-yue
authored and
copybara-github
committed
Add option to flush the writer from the observers.

summer-yue
authored and
copybara-github
committed

Commits on May 13, 2021

Only write items in Reverb when there's data appended in the writer.

summer-yue
authored and
copybara-github
committed
Adjust reverb signature in SAC example to be compatible with new Reverb writer.

summer-yue
authored and
copybara-github
committed

Commits on Apr 28, 2021

Fully align PPO sequential network setup with the legacy actor distribution setup.

summer-yue
authored and
copybara-github
committed

Commits on Apr 22, 2021

Use the current user as the group owner for reverb checkpointing.

summer-yue
authored and
copybara-github
committed

Commits on Apr 21, 2021

Always use batch-wise advantage normalization for PPO.

summer-yue
authored and
copybara-github
committed

Commits on Apr 20, 2021

Use the current user as the group owner for reverb checkpointing.

summer-yue
authored and
copybara-github
committed

Commits on Apr 1, 2021

Add ReverbCheckpointTrigger and enable checkpointing the replay buffer periodically in the SAC example.

summer-yue
authored and
copybara-github
committed

Commits on Feb 10, 2021

Use sequential network in PPO examples and the learner unit test.

summer-yue
authored and
copybara-github
committed

Commits on Jan 26, 2021

Add an optional callback in the distributed eval job to enable optional reporting.

summer-yue
authored and
copybara-github
committed

Commits on Jan 14, 2021

Use trajectory observer with padding for DQN RNN example.

summer-yue
authored and
copybara-github
committed
Support padding for ReverbAddTrajectoryObserver and better align observer interface.

summer-yue
authored and
copybara-github
committed

Commits on Jan 13, 2021

Internal change.

summer-yue
authored and
copybara-github
committed
Reinforce agent bug fix on conversion from dict to ReinforceAgentLossInfo.

summer-yue
authored and
copybara-github
committed

Commits on Dec 29, 2020

API changes for ReverbAddEpisodeObserver to better align observer interface.

summer-yue
authored and
copybara-github
committed

0