I want to use exponential decay for the entropy coefficient in PPO. The exponential decay is defined here ray.rllib.utils.schedules.exponential_schedule — Ray v1.10.0. How can I pass it to the algorithm?
The default schedule is “entropy_coeff_schedule”: None.
It will be really helpful if someone can answer this question. If the default is None, it would be helpful to know how to set its values. If a nonzero value is passed, an error messages is shown saying use entropy_coeff to setup a schedule. It would be really helpful if someone can tell how to do this.