net.sourceforge.jabm.learning
Interface ActionSelector

All Known Implementing Classes:

EpsilonGreedyActionSelector, SoftMaxActionSelector

public interface ActionSelector

An action selection policy for a reinforcement-learning algorithm. The job of the action selector is to select from the available actions in a particular state in such a way as to balance exploitation against exploration.

See:
Sutton, R. S., Barto, A. G., 1998. Reinforcement Learning: An Introduction. MIT Press.

Method Summary

int	act(int state, MDPLearner learner) Choose an action according to the current state and the current value estimates for each action.

Method Detail

act

int act(int state,
        MDPLearner learner)

Choose an action according to the current state and the current value estimates for each action.

Parameters:

state - The current state of the MDP.

learner - The algorithm used to update the value estimates.

Returns:

An integer representing the action chosen (indexed from 0).