Commit graph

17 commits

Author SHA1 Message Date
Kai Fricke
a5ea99cf95
[rfc] [tune/rllib] Fetch _progress_metrics from trainable for verbose=2 display (#26967)
RLLibs trainables produce a large number of metrics which makethe log output with verbose=2 illegible. This PR introduces a private `_progress_metrics` property for trainables. If set, the trial progress callback will only print these metrics per default, unless overridden e.g. with a custom `TrialProgressCallback`.
2022-07-27 16:04:23 +01:00
Jun Gong
c7ae787cc8
[RLlib] Beef up worker failure test. (#26953) 2022-07-27 00:10:45 -07:00
Jun Gong
ca5e0dcaf4
[RLLib] Record framework and algorithm used by an RLlib run. (#26956)
Automatically record framework and algorithm used by RLlib jobs.
For better planning.
2022-07-25 16:16:36 -07:00
Rohan Potdar
38c9e1d52a
[RLlib]: Fix OPE trainables (#26279)
Co-authored-by: Kourosh Hakhamaneshi <kourosh@anyscale.com>
2022-07-17 14:25:53 -07:00
Sven Mika
4aea24c8a8
[RLlib] restart_failed_sub_environments now works for MA cases and crashes during reset(); +more tests and logging; add eval worker sub-env fault tolerance test. (#26276) 2022-07-15 08:55:14 +02:00
Rohan Potdar
09ce4711fd
[RLlib]: Move OPE to evaluation config (#25911) 2022-07-12 11:04:34 -07:00
Sven Mika
ca913ff6d6
[RLlib] Eval WorkerSet crashes when trying to re-add a failed worker (eval set does not have local worker). (#26134) 2022-06-30 13:25:22 +02:00
Jun Gong
d83bbda281
[RLlib] Save serialized PolicySpec. Extract num_gpus related logics into a util function. (#25954) 2022-06-30 11:38:21 +02:00
Sven Mika
2b43713785
[RLlib] Move IMPALA and APPO back to exec plan (for now; due to unresolved learning/performance issues). (#25851) 2022-06-29 08:41:47 +02:00
Sven Mika
762cfbdff1
[RLlib] IMPALA and APPO metrics fixes; remove deprecated async_parallel_requests utility. (#26117) 2022-06-28 15:14:37 +02:00
Artur Niederfahrenhorst
bed9083f35
[RLlib] Add timeout to filter synchronization. (#25959) 2022-06-24 14:37:43 +02:00
Sven Mika
59a967a3a0
[RLlib] Cleanup some deprecated metric keys and classes. (#26036) 2022-06-23 21:30:01 +02:00
Kai Fricke
8a2f6bda62
[tune/structure] Introduce experiment package (#26033)
Experiment, Trial, and config parsing moves into an `experiment` package.

Notably, the new public facing APIs will be

```
from ray.tune.experiment import Experiment
from ray.tune.experiment import Trial
```
2022-06-23 14:52:46 +01:00
Kai Fricke
0959f44b6f
[tune/structure] Introduce execution package (#26015)
Execution-specific packages are moved to tune.execution.

Co-authored-by: Xiaowei Jiang <xwjiang2010@gmail.com>
2022-06-23 11:13:19 +01:00
Sven Mika
1499af945b
[RLlib] Algorithm step() fixes: evaluation should NOT be part of timed training_step loop. (#25924) 2022-06-20 19:53:47 +02:00
Kai Fricke
6313ddc47c
[tune] Refactor Syncer / deprecate Sync client (#25655)
This PR includes / depends on #25709

The two concepts of Syncer and SyncClient are confusing, as is the current API for passing custom sync functions.

This PR refactors Tune's syncing behavior. The Sync client concept is hard deprecated. Instead, we offer a well defined Syncer API that can be extended to provide own syncing functionality. However, the default will be to use Ray AIRs file transfer utilities.

New API:
- Users can pass `syncer=CustomSyncer` which implements the `Syncer` API
- Otherwise our off-the-shelf syncing is used
- As before, syncing to cloud disables syncing to driver

Changes:
- Sync client is removed
- Syncer interface introduced
- _DefaultSyncer is a wrapper around the URI upload/download API from Ray AIR
- SyncerCallback only uses remote tasks to synchronize data
- Rsync syncing is fully depracated and removed
- Docker and kubernetes-specific syncing is fully deprecated and removed
- Testing is improved to use `file://` URIs instead of mock sync clients
2022-06-14 14:46:30 +02:00
Sven Mika
130b7eeaba
[RLlib] Trainer to Algorithm renaming. (#25539) 2022-06-11 15:10:39 +02:00