PolicyIteration.hpp

Go to the documentation of this file.

#ifndef AI_TOOLBOX_MDP_POLICY_ITERATION_HEADER_FILE
#define AI_TOOLBOX_MDP_POLICY_ITERATION_HEADER_FILE
 
#include <AIToolbox/MDP/Types.hpp>
#include <AIToolbox/MDP/TypeTraits.hpp>
#include <AIToolbox/MDP/Utils.hpp>
#include <AIToolbox/MDP/Policies/QGreedyPolicy.hpp>
#include <AIToolbox/MDP/Algorithms/Utils/PolicyEvaluation.hpp>
 
namespace AIToolbox::MDP {
    class PolicyIteration {
        public:
            PolicyIteration(unsigned horizon, double tolerance = 0.001);
 
            template <IsModel M>
            QFunction operator()(const M & m);
 
            void setTolerance(double t);
 
            void setHorizon(unsigned h);
 
            double getTolerance() const;
 
            unsigned getHorizon() const;
 
        private:
            unsigned horizon_;
            double tolerance_;
    };
 
    template <IsModel M>
    QFunction PolicyIteration::operator()(const M & m) {
        const auto S = m.getS();
        const auto A = m.getA();
 
        PolicyEvaluation<M> eval(m, horizon_, tolerance_);
 
        auto qfun = makeQFunction(m.getS(), m.getA());
        QGreedyPolicy p(qfun);
        auto matrix = p.getPolicy();
 
        {
nextLoop:
            auto [bound, v, q] = eval(p);
            (void)bound;
 
            eval.setValues(std::move(v));
            qfun = std::move(q);
 
            auto newMatrix = p.getPolicy();
            for (size_t s = 0; s < S; ++s) {
                for (size_t a = 0; a < A; ++a) {
                    if (checkDifferentSmall(matrix(s,a), newMatrix(s,a))) {
                        matrix = std::move(newMatrix);
                        goto nextLoop;
                    }
                }
            }
        }
        return qfun;
    }
}
 
#endif