Tuning hyperparameters of RLLIB PPO on AWS Sagemakerr

I have build a docker container of ppo algorithm of RLLIB. I am using amazon sagemaker to tune hyperparameters. Sagemaker requires metric_definition to optimize. What should I use as in metric_defination of sagemaker. I am not using RL estimator of sagemaker. Thanking in advance for your response.

The RLlib version for hyperparameter tuning uses Ray Tune under the hood.

A good example of how you can pass it in RLlib config is here: ray/halfcheetah-ppo.yaml at master · ray-project/ray · GitHub.

In this file, grid_search allows for searching over different discrete hyperparameters. The “metric definition” of Sagemaker is equivalent to a Stop condition (e.g. when an agent hits 9k reward in halfcheetah) also defined in this file.