6.4.1. Custom Policies

• Policy Creation

  

  Creating a GT policy that satisfies the MLPro interface is pretty straightforward. You just require to assure that the GT policy consists of at least the following 3 main functions:

  from mlpro.rl.models import *
  
  class MyPolicy(Policy):
      """
      Creates a policy that satisfies the mlpro interface.
      """
      C_NAME          = 'MyPolicy'
  
      def _init_hyperparam(self, **p_par):
          """
          Implementation-specific hyperparameters can be added here. Please follow these steps:
          a) Add each hyperparameter as an object of type HyperParam to the internal hyperparameter
             space object self._hyperparam_space
          b) Create a hyperparameter tuple and bind it to self._hyperparam_tuple
          c) Set the default value for each hyperparameter
  
          Parameters
          ----------
          p_par : Dict
              Further model-specific parameters, that are passed through the constructor.
  
          """
          ....
  
      def compute_action(self, p_obs: State) -> Action:
          """
          Specific action computation method to be redefined.
  
          Parameters
          ----------
          p_obs : State
              Observation data.
  
          Returns
          -------
          action : Action
              Action object.
  
          """
          ....
  
      def _adapt(self, *p_args) -> bool:
          """
          Adapts the policy based on State-Action-Reward-State (SARS) data.
  
          Parameters
          ----------
          p_arg[0] : SARSElement
              Object of type SARSElement.
  
          Returns
          -------
          adapted : bool
              True, if something has been adapted. False otherwise.
  
          """
          ....
  

  To set up a hyperparameter space, please refer to Howto BF-ML-010: Hyperparameters.

• Algorithm Checker

  A test script using the unit test to check the developed policies will be available soon!