How to change policies to train during a training run?

How severe does this issue affect your experience of using Ray?

  • High: It blocks me to complete my task.

Hi, I am currently trying to have a self-play implementation and I am taking inspiration from the open-spiel example. I am trying to customise the algorithm to change the policies to train. I have tried two methods. The first is using algorithm.setstate() to change the variable “policies_to_train”. Secondly is to use add_algorithm and use the algorithm.add_policy function and define the policies_to_train variable. Both methods do not change the policies_to_train variable for some reason. I verified this by printing the algorithm.getstate()[‘config’][‘multiagent’] in the on_train_result method of the DefaultCallback class. Can anyone guide me on how to resolve this issue, specifically changing policies to train during a training run? Thanks