How to run tuned example

RLlib Configure Algorithm, Training, Evaluation, Scaling
1

Is there any way to run the tuned example from Ray? I did not find any instructions or documents regarding this.

2

These files have a special syntax.

Today, you can do something like. python rllib/tests/run_regression_tests.py --dir tuned_examples/ppo/cartpole-ppo.yaml. This is mainly for our CI.

But you can copy these files and modify them a little to get just the AlgorithmConfig itself. Then use rllib.utils.from_config.from_file to generate a config from it.

Be aware that our configs are currently changing and we will soon provide these examples as actual Python config files that hold define AlgorithmConfig objects, rather than yamls.

3

Adding to this: Our new CLI makes the following possible:

rllib train file tuned_examples/ppo/cartpole-ppo.yaml

You can also pass a URI.