ESRLPolicy.hpp

Go to the documentation of this file.

#ifndef AI_TOOLBOX_BANDIT_ESRL_POLICY_HEADER_FILE
#define AI_TOOLBOX_BANDIT_ESRL_POLICY_HEADER_FILE
 
#include <AIToolbox/Bandit/Policies/LRPPolicy.hpp>
 
namespace AIToolbox::Bandit {
    class ESRLPolicy : public PolicyInterface {
        public:
            ESRLPolicy(size_t A, double a, unsigned timesteps, unsigned explorationPhases, unsigned window);
 
            void stepUpdateP(size_t a, bool result);
 
            bool isExploiting() const;
 
            virtual size_t sampleAction() const override;
 
            virtual double getActionProbability(const size_t & a) const override;
 
            void setAParam(double a);
 
            double getAParam() const;
 
            void setTimesteps(unsigned t);
 
            unsigned getTimesteps() const;
 
            void setExplorationPhases(unsigned p);
 
            unsigned getExplorationPhases() const;
 
            void setWindowSize(unsigned window);
 
            unsigned getWindowSize() const;
 
            virtual Vector getPolicy() const override;
 
        private:
            // Whether we have learned enough to start exploiting.
            bool exploit_;
            size_t bestAction_;
            // Timesteps in current exploration phase in overall exploration phases.
            size_t timestep_, N_, explorations_, explorationPhases_;
            // Average value obtained in last window in the last exploration phase.
            double average_;
            size_t window_;
 
            // Values obtained for all actions.
            Vector values_;
            // Allowed actions in the current exploration phase.
            std::vector<size_t> allowedActions_;
            // Exploration learning policy to learn Nash equilibria.
            LRPPolicy lri_;
    };
}
 
#endif