* Add base for Soft Actor-Critic
* Pick changes from old SAC branch
* Update sac.py
* First implementation of sac model
* Remove unnecessary SAC imports
* Prune unnecessary noise and exploration code
* Implement SAC model and use that in SAC policy
* runs but doesn't learn
* clear state
* fix batch size
* Add missing alpha grads and vars
* -200 by 2k timesteps
* doc
* lazy squash
* one file
* ignore tfp
* revert done
auto wrap multi-agent dict and tuple spaces by keeping a policy -> preprocessor in the sampler
add some Q-learning debug stats
report min, max of custom metrics
better errors
This includes most of the TF code used for the OSDI experiment. Perf sanity check on p3.16xl instances: Overall scaling looks ok, with the multi-node results within 5% of OSDI final numbers. This seems reasonable given that hugepages are not enabled here, and the param server shards are placed randomly.
$ RAY_USE_XRAY=1 ./test_sgd.py --gpu --batch-size=64 --num-workers=N \
--devices-per-worker=M --strategy=<simple|ps> \
--warmup --object-store-memory=10000000000
Images per second total
gpus total | simple | ps
========================================
1 | 218
2 (1 worker) | 388
4 (1 worker) | 759
4 (2 workers) | 176 | 623
8 (1 worker) | 985
8 (2 workers) | 349 | 1031
16 (2 nodes, 2 workers) | 600 | 1661
16 (2 nodes, 4 workers) | 468 | 1712 <--- OSDI perf was 1817
It's possible to configure PPO in a way that ends up discarding most of the samples (they are treated as "stragglers"). Add a warning when this happens, and raise an exception if the waste is particularly egregious.
A bunch of minor rllib fixes:
pull in latest baselines atari wrapper changes (and use deepmind wrapper by default)
move reward clipping to policy evaluator
add a2c variant of a3c
reduce vision network fc layer size to 256 units
switch to 84x84 images
doc tweaks
print timesteps in tune status
This PR introduces the following changes:
* Ray Tune -> Tune
* [breaking] Creation of `schedulers/`, moving PBT, HyperBand into a submodule
* [breaking] Search Algorithms now must take in experiment configurations via `add_configurations` rather through initialization
* Support `"run": (function | class | str)` with automatic registering of trainable
* Documentation Changes
Rename AsyncSamplesOptimizer -> AsyncReplayOptimizer
Add AsyncSamplesOptimizer that implements the IMPALA architecture
integrate V-trace with a3c policy graph
audit V-trace integration
benchmark compare vs A3C and with V-trace on/off
PongNoFrameskip-v4 on IMPALA scaling from 16 to 128 workers, solving Pong in <10 min. For reference, solving this env takes ~40 minutes for Ape-X and several hours for A3C.
This also removes the async resetting code in VectorEnv. While that improves benchmark performance slightly, it substantially complicates env configuration and probably isn't worth it for most envs.
This makes it easy to efficiently support setups like Joint PPO: https://s3-us-west-2.amazonaws.com/openai-assets/research-covers/retro-contest/gotta_learn_fast_report.pdf
For example, for 188 envs, you could do something like num_envs: 10, num_envs_per_worker: 19.
