Policy server with Tune

Ray Libraries (Data, Train, Tune, Serve) Ray Tune
1

We have set up client server similar to this example:

ray/cartpole_server.py at master · ray-project/ray · GitHub

We would like to use Tune for hyperparameter tuning.

What will be the recommended way to use it in this setting?

Is it possible to subscribe to Tune event when starting a new trial so we can start additional simulators?

2

Hi @antmask, does this Ray Tune example help? ray/pbt_ppo_example.py at master · ray-project/ray · GitHub
You probably don’t want to use PBT here, but the rest of the setup (the call to tune.run()) should be helpful.

What do you mean with subscribing to Tune events? Do you mean event callbacks? Like this: Tune Internals — Ray v2.0.0.dev0

3

Thanks Kai, we’ll have a look on the callbacks. Looks helpful thanks.
For the first part, the ray tune example doesn’t use the client server architecture, are you aware of an example that uses this architecture?

Thank you.