BanditPolicyAdaptor.hpp

Go to the documentation of this file.

#ifndef AI_TOOLBOX_MDP_BANDIT_POLICY_ADAPTOR_HEADER_FILE
#define AI_TOOLBOX_MDP_BANDIT_POLICY_ADAPTOR_HEADER_FILE
 
#include <AIToolbox/Bandit/Policies/PolicyInterface.hpp>
#include <AIToolbox/MDP/Policies/PolicyInterface.hpp>
 
namespace AIToolbox::MDP {
    template <typename BanditPolicy>
    class BanditPolicyAdaptor : public PolicyInterface {
        public:
            template <typename... Args>
            BanditPolicyAdaptor(size_t s, Args&&... params);
 
            virtual size_t sampleAction(const size_t & s) const override;
 
            virtual double getActionProbability(const size_t & s, const size_t & a) const override;
 
            virtual Matrix2D getPolicy() const override;
 
            BanditPolicy & getBanditPolicy();
 
            const BanditPolicy & getBanditPolicy() const;
 
        private:
            BanditPolicy policy_;
    };
 
    template <typename BP>
    template <typename... Args>
    BanditPolicyAdaptor<BP>::BanditPolicyAdaptor(const size_t s, Args&&... args) :
            Base(s, 1), policy_(std::forward<Args>(args)...)
    {
        // We need to fix this later since we can't initialize the policy
        // before Base.
        A = policy_.getA();
    }
 
    template <typename BP>
    size_t BanditPolicyAdaptor<BP>::sampleAction(const size_t &) const {
        return policy_.sampleAction();
    }
 
    template <typename BP>
    double BanditPolicyAdaptor<BP>::getActionProbability(const size_t &, const size_t & a) const {
        return policy_.getActionProbability(a);
    }
 
    template <typename BP>
    Matrix2D BanditPolicyAdaptor<BP>::getPolicy() const {
        return policy_.getPolicy().transpose().replicate(getS(), 1);
    }
 
    template <typename BP>
    BP & BanditPolicyAdaptor<BP>::getBanditPolicy() { return policy_; }
 
    template <typename BP>
    const BP & BanditPolicyAdaptor<BP>::getBanditPolicy() const { return policy_; }
}
 
#endif