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
Add new search algorithm (genetic) along with the base framework of the searcher (which performs some basic jobs such as logging, recording and organizing in our project).
Note that this is the initial commit. In the following days, we will add example, UT, and other refinements.
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
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.
The dict merge prevents crashes when tune is trying to get resource requests for agents and you override a config subkey. The min iter time prevents iterations from getting too small, incurring high overhead. This is easy to run into on Ape-X since throughput can get very high.
This adds a simple DQN+PPO example for multi-agent. We don't do anything fancy here, just syncing weights between two separate trainers. This potentially is wasting some compute, but is very simple to set up.
It might be nice to share experience collection between the top-level trainers in the future.
Cleanup: TFPolicyGraph now automatically adds loss input entries for state_in_*, so that graph sub-classes don't need to worry about it.
Multi-GPU support:
Allow setting up model tower replicas with existing state input tensors
Truncate the per-device minibatch slices so that they are always a multiple of max_seq_len.
* Add profile table and store profiling information there.
* Code for dumping timeline.
* Improve color scheme.
* Push timeline events on driver only for raylet.
* Improvements to profiling and timeline visualization
* Some linting
* Small fix.
* Linting
* Propagate node IP address through profiling events.
* Fix test.
* object_id.hex() should return byte string in python 2.
* Include gcs.fbs in node_manager.fbs.
* Remove flatbuffer definition duplication.
* Decode to unicode in Python 3 and bytes in Python 2.
* Minor
* Submit profile events in a batch. Revert some CMake changes.
* Fix
* Workaround test failure.
* Fix linting
* Linting
* Don't return anything from chrome_tracing_dump when filename is provided.
* Remove some redundancy from profile table.
* Linting
* Move TODOs out of docstring.
* Minor
## What do these changes do?
**Vectorized envs**: Users can either implement `VectorEnv`, or alternatively set `num_envs=N` to auto-vectorize gym envs (this vectorizes just the action computation part).
```
# CartPole-v0 on single core with 64x64 MLP:
# vector_width=1:
Actions per second 2720.1284458322966
# vector_width=8:
Actions per second 13773.035334888269
# vector_width=64:
Actions per second 37903.20472563333
```
**Async envs**: The more general form of `VectorEnv` is `AsyncVectorEnv`, which allows agents to execute out of lockstep. We use this as an adapter to support `ServingEnv`. Since we can convert any other form of env to `AsyncVectorEnv`, utils.sampler has been rewritten to run against this interface.
**Policy serving**: This provides an env which is not stepped. Rather, the env executes in its own thread, querying the policy for actions via `self.get_action(obs)`, and reporting results via `self.log_returns(rewards)`. We also support logging of off-policy actions via `self.log_action(obs, action)`. This is a more convenient API for some use cases, and also provides parallelizable support for policy serving (for example, if you start a HTTP server in the env) and ingest of offline logs (if the env reads from serving logs).
Any of these types of envs can be passed to RLlib agents. RLlib handles conversions internally in CommonPolicyEvaluator, for example:
```
gym.Env => rllib.VectorEnv => rllib.AsyncVectorEnv
rllib.ServingEnv => rllib.AsyncVectorEnv
```
* Use F.softmax instead of a pointless network layer
Stateless functions should not be network layers.
* Use correct pytorch functions
* Rename argument name to out_size
Matches in_size and makes more sense.
* Fix shapes of tensors
Advantages and rewards both should be scalars, and therefore a list of them
should be 1D.
* Fmt
* replace deprecated function
* rm unnecessary Variable wrapper
* rm all use of torch Variables
Torch does this for us now.
* Ensure that values are flat list
* Fix shape error in conv nets
* fmt
* Fix shape errors
Reshaping the action before stepping in the env fixes a few errors.
* Add TODO
* Use correct filter size
Works when `self.config['model']['channel_major'] = True`.
* Add missing channel major
* Revert reshape of action
This should be handled by the agent or at least in a cleaner way that doesn't
break existing envs.
* Squeeze action
* Squeeze actions along first dimension
This should deal with some cases such as cartpole where actions are scalars
while leaving alone cases where actions are arrays (some robotics tasks).
* try adding pytorch tests
* typo
* fixup docker messages
* Fix A3C for some envs
Pendulum doesn't work since it's an edge case (expects singleton arrays, which
`.squeeze()` collapses to scalars).
* fmt
* nit flake
* small lint
* Treat actor creation like a regular task.
* Small cleanups.
* Change semantics of actor resource handling.
* Bug fix.
* Minor linting
* Bug fix
* Fix jenkins test.
* Fix actor tests
* Some cleanups
* Bug fix
* Fix bug.
* Remove cached actor tasks when a driver is removed.
* Add more info to taskspec in global state API.
* Fix cyclic import bug in tune.
* Fix
* Fix linting.
* Fix linting.
* Don't schedule any tasks (especially actor creaiton tasks) on local schedulers with 0 CPUs.
* Bug fix.
* Add test for 0 CPU case
* Fix linting
* Address comments.
* Fix typos and add comment.
* Add assertion and fix test.
