public MDPPolicy <STATE_TYPE, ACTION_TYPE> policyIteration(double gamma)
{
MDPUtilityFunction <STATE_TYPE> U = initialUtilityFunction();
MDPPolicy <STATE_TYPE, ACTION_TYPE> pi = randomPolicy();
bool unchanged = false;
do
{
unchanged = true;
U = policyEvaluation(pi, U, gamma, 3);
foreach (STATE_TYPE s in nonFinalstates)
{
Pair <ACTION_TYPE, Double> maxTransit = transitionModel
.getTransitionWithMaximumExpectedUtility(s, U);
Pair <ACTION_TYPE, Double> maxPolicyTransit = transitionModel
.getTransitionWithMaximumExpectedUtilityUsingPolicy(pi,
s, U);
if (maxTransit.getSecond() > maxPolicyTransit.getSecond())
{
pi.setAction(s, maxTransit.getFirst());
unchanged = false;
}
}
} while (unchanged == false);
return(pi);
}
public MDPPolicy <STATE_TYPE, ACTION_TYPE> randomPolicy()
{
MDPPolicy <STATE_TYPE, ACTION_TYPE> policy = new MDPPolicy <STATE_TYPE, ACTION_TYPE>();
foreach (STATE_TYPE s in nonFinalstates)
{
policy.setAction(s, transitionModel.randomActionFor(s));
}
return(policy);
}