gjoliver
d81885c1f1
[RLlib] Fix all the CI tests that were broken by is_training and replay buffer changes; re-comment-in the failing RLlib tests ( #19809 )
...
* Fix DDPG, since it is based on GenericOffPolicyTrainer.
* Fix QMix, SAC, and MADDPA too.
* Undo QMix change.
* Fix DQN input batch type. Always use SampleBatch.
* apex ddpg should not use replay_buffer_config yet.
* Make eager tf policy to use SampleBatch.
* lint
* LINT.
* Re-enable RLlib broken tests to make sure things work ok now.
* fixes.
Co-authored-by: sven1977 <svenmika1977@gmail.com>
2021-10-28 18:06:47 +02:00
Sven Mika
4a82d3ea6c
Revert "[RLlib; Docs overhaul] Docstring cleanup: Trainer, trainer_template, Callbacks. ( #19758 )" ( #19806 )
...
This reverts commit 80eeb13175
2021-10-27 23:30:07 +02:00
Sven Mika
80eeb13175
[RLlib; Docs overhaul] Docstring cleanup: Trainer, trainer_template, Callbacks. ( #19758 )
2021-10-27 19:15:35 +02:00
gjoliver
99a0088233
[RLlib] Unify the way we create local replay buffer for all agents ( #19627 )
...
* [RLlib] Unify the way we create and use LocalReplayBuffer for all the agents.
This change
1. Get rid of the try...except clause when we call execution_plan(),
and get rid of the Deprecation warning as a result.
2. Fix the execution_plan() call in Trainer._try_recover() too.
3. Most importantly, makes it much easier to create and use different types
of local replay buffers for all our agents.
E.g., allow us to easily create a reservoir sampling replay buffer for
APPO agent for Riot in the near future.
* Introduce explicit configuration for replay buffer types.
* Fix is_training key error.
* actually deprecate buffer_size field.
2021-10-26 20:56:02 +02:00
Sven Mika
b213565783
[RLlib] Fix failing test cases: Soft-deprecate ModelV2.from_batch (in favor of ModelV2.__call__). ( #19693 )
2021-10-25 15:00:00 +02:00
gjoliver
89fbfc00f8
[RLlib] Some minor cleanups (buffer buffer_size -> capacity and others). ( #19623 )
2021-10-25 09:42:39 +02:00
roireshef
9b0352f363
[RLlib] Added LearningRateSchedule and EntropyCoeffSchedule to TF and Torch versions of A3C and PPO ( #19276 )
2021-10-25 09:39:35 +02:00
gjoliver
c3c42278e4
[RLlib] clean up all the SampleBatch['is_training'] deprecation warnings ( #19652 )
...
* [RLlib] clean up all the SampleBatch['is_training'] deprecation warnings.
* wip
2021-10-25 09:38:56 +02:00
gjoliver
44a4e42172
[rllib] Add entropy_coeff_schedule support for APPO. ( #19544 )
...
* Add entropy_coeff_schedule support for APPO.
* lint
2021-10-20 14:18:01 -07:00
gjoliver
9226f9bddc
[RLlib] Report timesteps_this_iter to Tune, so it can track/checkpoint/restore total timesteps trained. ( #19264 )
...
* Report timesteps_this_iter to Tune, so it can track/checkpoint/restore
total timesteps trained.
* Trigger Build
* lint
2021-10-12 16:03:41 +02:00
Sven Mika
bd2d2079d2
[RLlib] Support >1 loss terms and optimizers for framework=tf2 (already supported for framework=[tf|torch]) ( #19269 )
2021-10-10 12:19:47 +02:00
Sven Mika
d439fd7f17
[RLlib] TF2/eager memory leak fixes. ( #19198 )
2021-10-09 00:11:53 +02:00
Sven Mika
fd438d5630
[RLlib] Issue 18104: Cannot set remote_worker_envs=True for non local-mode and MultiAgentEnv. ( #19133 )
2021-10-07 22:39:21 +02:00
Sven Mika
1f0646f658
[RLlib] Issue 18418: SAC w/ dict space not working. ( #19101 )
2021-10-06 09:05:50 +02:00
Sven Mika
b4300dd532
[RLlib] Issue 18812: Torch multi-GPU stats not protected against race conditions. ( #18937 )
2021-10-04 13:29:00 +02:00
Sven Mika
73f5c4039b
[RLlib] Fix flakey test_a3c, test_maml, test_apex_dqn. ( #19035 )
2021-10-04 13:23:51 +02:00
Jiajun Yao
7588bfd315
[Lint] Add flake8-bugbear ( #19053 )
...
* Add flake8-bugbear
* Add flake8-bugbear
2021-10-03 23:24:11 -07:00
Sven Mika
ed85f59194
[RLlib] Unify all RLlib Trainer.train() -> results[info][learner][policy ID][learner_stats] and add structure tests. ( #18879 )
2021-09-30 16:39:05 +02:00
Sven Mika
828f5d26b7
[RLlib] Custom view requirements (e.g. for prev-n-obs) work with compute_single_action and compute_actions_from_input_dict. ( #18921 )
2021-09-30 15:03:37 +02:00
Avnish Narayan
6dc1a6b72f
[RLlib] Raise error for kl penalty ddpo ( #18959 )
...
* [RLlib] Raise error for kl penalty ddpo
DDPPO doesn't support KL penalties like PPO-1.
In order to support KL penalties, DDPPO would need to
become undecentralized, which defeats the purpose of the
algorithm. Users can still tune the entropy coefficient to
control the policy entropy (similar to controlling the KL
penalty.)
* Update rllib/agents/ppo/ddppo.py
Co-authored-by: avnishn <avnishnarayan@gmail.com>
Co-authored-by: Sven Mika <sven@anyscale.io>
2021-09-30 10:56:22 +02:00
Sven Mika
9c9b482661
[RLlib] Allow n-step > 1 and prio. replay for R2D2 and RNNSAC. ( #18939 )
2021-09-29 21:31:34 +02:00
Sven Mika
b99943806e
[RLlib] Add support for IMPALA to handle more than one loss/optimizer (analogous to recent enhancement for APPO). ( #18971 )
2021-09-29 21:30:04 +02:00
Sven Mika
61a1274619
[RLlib] No Preprocessors (part 2). ( #18468 )
2021-09-23 12:56:45 +02:00
Sven Mika
a2a077b874
[RLlib] Faster remote worker space inference (don't infer if not required). ( #18805 )
2021-09-23 10:54:37 +02:00
Sven Mika
698b4eeed3
[RLlib] POC: Separate losses for APPO/IMPALA. Enable TFPolicy to handle multiple optimizers/losses (like TorchPolicy). ( #18669 )
2021-09-21 22:00:14 +02:00
Sven Mika
fd13bac9b3
[RLlib] Add worker arg (optional) to policy_mapping_fn. ( #18184 )
2021-09-17 12:07:11 +02:00
Sven Mika
ba1c489b79
[RLlib Testing] Lower --smoke-test "time_total_s" to make sure it doesn't time out. ( #18670 )
2021-09-16 18:22:23 +02:00
Sven Mika
8a00154038
[RLlib] Bump tf version in ML docker to tf==2.5.0; add tfp to ML-docker. ( #18544 )
2021-09-15 08:46:37 +02:00
Sven Mika
08c09737fa
[RLlib] Fix R2D2 (torch) multi-GPU issue. ( #18550 )
2021-09-14 19:58:10 +02:00
Sven Mika
3803e796ff
[RLlib] Multi-GPU learner thread (IMPALA) error messages/comments/code-cleanup. ( #18540 )
2021-09-13 19:27:53 +02:00
Sven Mika
ea4a22249c
[RLlib] Add simple action-masking example script/env/model (tf and torch). ( #18494 )
2021-09-11 23:08:09 +02:00
Sven Mika
3f89f35e52
[RLlib] Better error messages and hints; + failure-mode tests; ( #18466 )
2021-09-10 16:52:47 +02:00
Sven Mika
8a066474d4
[RLlib] No Preprocessors; preparatory PR #1 ( #18367 )
2021-09-09 08:10:42 +02:00
Sven Mika
1520c3d147
[RLlib] Deepcopy env_ctx for vectorized sub-envs AND add eval-worker-option to Trainer.add_policy() ( #18428 )
2021-09-09 07:10:06 +02:00
gjoliver
808b683f81
[RLlib] Add a unittest for learning rate schedule used with APEX agent. ( #18389 )
2021-09-08 23:29:40 +02:00
Sven Mika
45f60e51a9
[RLlib] DDPPO fixes and benchmarks. ( #18390 )
2021-09-08 19:39:01 +02:00
Sven Mika
56f142cac1
[RLlib] Add support for evaluation_num_episodes=auto (run eval for as long as the parallel train step takes). ( #18380 )
2021-09-07 08:08:37 +02:00
Sven Mika
e3e6ed7aaa
[RLlib] Issues 17844, 18034: Fix n-step > 1 bug. ( #18358 )
2021-09-06 12:14:20 +02:00
Sven Mika
ba58f5edb1
[RLlib] Strictly run evaluation_num_episodes episodes each evaluation run (no matter the other eval config settings). ( #18335 )
2021-09-05 15:37:05 +02:00
Sven Mika
a772c775cd
[RLlib] Set random seed (if provided) to Trainer process as well. ( #18307 )
2021-09-04 11:02:30 +02:00
Sven Mika
9a8ca6a69d
[RLlib] Fix Atari learning test regressions (2 bugs) and 1 minor attention net bug. ( #18306 )
2021-09-03 13:29:57 +02:00
Sven Mika
82465f9342
[RLlib] Better PolicyServer example (w/ or w/o tune) and add printing out actual listen port address in log-level=INFO. ( #18254 )
2021-08-31 22:03:23 +02:00
Sven Mika
599e589481
[RLlib] Move existing fake multi-GPU learning tests into separate buildkite job. ( #18065 )
2021-08-31 14:56:53 +02:00
Sven Mika
4888d7c9af
[RLlib] Replay buffers: Add config option to store contents in checkpoints. ( #17999 )
2021-08-31 12:21:49 +02:00
Joseph Suarez
8136d2912b
[RLlib] Add policies arg to callback: on_episode_step (already exists in all other episode-related callbacks) ( #18119 )
2021-08-27 16:12:19 +02:00
Sven Mika
b6aa8223bc
[RLlib] Fix final_scale's default value to 0.02 (see OrnsteinUhlenbeck exploration). ( #18070 )
2021-08-25 14:22:09 +02:00
Sven Mika
9883505e84
[RLlib] Add [LSTM=True + multi-GPU]-tests to nightly RLlib testing suite (for all algos supporting RNNs, except R2D2, RNNSAC, and DDPPO). ( #18017 )
2021-08-24 21:55:27 +02:00
Sven Mika
494ddd98c1
[RLlib] Replace "seq_lens" w/ SampleBatch.SEQ_LENS. ( #17928 )
2021-08-21 17:05:48 +02:00
Sven Mika
8248ba531b
[RLlib] Redo #17410 : Example script: Remote worker envs with inference done on main node. ( #17960 )
2021-08-20 08:02:18 +02:00
Alex Wu
318ba6fae0
Revert "[RLlib] Add example script for how to have n remote (parallel) envs with inference happening on "main" (possibly GPU) node. ( #17410 )" ( #17951 )
...
This reverts commit 8fc16b9a18
2021-08-19 07:55:10 -07:00
