SampleBP.h

Go to the documentation of this file.

/***************************************************************************
File: SampleBP.h
Data: 20/03/2007
Author: rn
Brief: Sample subclass for BP sampling
 ***************************************************************************/
#ifndef SAMPLEBP_H
#define SAMPLEBP_H

#define BP_INIT_MAX_DEPTH (10)
#define BP_MAX_DEPTH_ADJUST_RATIO (1.1)
//#define BP_IMPROVEMENT_CONSTANT (0.95)  // now is a variable in SolverParams

#define USE_HSVI_ADAPTIVE_DEPTH 1

#include "Sample.h"
#include "BeliefCache.h"
#include "BinManager.h"
#include "BinManagerSet.h"
#include "PointBasedAlgorithm.h"
#include "BeliefTreeNode.h"


using namespace std;
using namespace momdp;

/****************************************
  DECLARATION OF SampleBP
 ****************************************/

namespace momdp{


    class SARSOP;

    struct BPUpdateResult {
        int maxUBAction;
        double maxUBVal;
        double ubResidual;
        int maxExcessUncStateOutcome;  // chosen value of observed state variable       
        int maxExcessUncOutcome;        // chosen value of observation variable
        double maxExcessUnc;
    };

    class SampleBP: public Sample{
        private:
            //attributes
#if USE_HSVI_ADAPTIVE_DEPTH
            std::vector<int> depthArr; //int depth;             // SYLMOD17102008 for parallel trials
            double logOcc;
            double maxDepth,oldMaxDepth;
            double oldQualitySum, newQualitySum;
            int oldNumUpdates, newNumUpdates;
#endif
            //  std::vector<double> trialTargetPrecisionArr;  // double trialTargetPrecision;           // SYLMOD17102008 for parallel trials



            //BeliefCache* beliefCache;  // removed for factored, now accessed through Bounds class
            std::vector<list<cacherow_stval> > priorityQueueArr; //list<cacherow_stval> priorityQueue; // modified for factored, prevly, list <int> priorityQueue;
            // SYLMOD17102008 for parallel trials

        public:
            int numTrials;
            int numSubOptimal; //xan-edit to count no of suboptimal paths found
            //double targetPrecision; //xan-edit                // SYLMOD17102008 doesnt seem to be used
            BinManagerSet * bm;//for Bin Heuristic
            int numBinProceed;//for Bin Heuristic
            std::vector<double> nextNodeTargetUbArr;  //double nextNodeTargetUb;  // SYLMOD17102008 for parallel trials //for BP (revised)
            bool isRoot;//for BP (revised)
            std::vector<double> nextNodeTargetLbArr; //double nextNodeTargetLb;// SYLMOD17102008 for parallel trials //new heuristic

            std::vector<double> trialTargetPrecisionArr;

            std::vector<int> newTrialFlagArr; //int newTrialFlag; //ADD SYLTAG : 1 indicates to the chooser of roots that a new trial should start 
            // SYLMOD17102008 for parallel trials
            //constructor
            SampleBP();
            virtual void setup(SharedPointer<MOMDP> _problem, SARSOP* _solver);

            SARSOP* solver;
            //methods
            list<cacherow_stval> sample(cacherow_stval currentBeliefIndex, unsigned int currentRoot);
            // SYLMOD17102008 for parallel trials
            //list<cacherow_stval> sample(cacherow_stval currentBeliefIndex); // prevly, list<int> sample(int currentBeliefIndex);
            void getMaxExcessUncOutcome(BeliefTreeNode& cn, BPUpdateResult& r, unsigned int currentRoot) const;
            // SYLMOD17102008 for parallel trials       
            //void getMaxExcessUncOutcome(BeliefTreeNode& cn, BPUpdateResult& r) const;

            void setTargetPrecision(double _targetPrecision);
            void setBinManager(BinManagerSet* _bm);
            //bool CompareIfLowerBoundImprovesAction(int action, int observation, BeliefTreeNode & currentNode);
            bool CompareIfLowerBoundImprovesAction(int action, int observation, int xstate, BeliefTreeNode & currentNode, unsigned int currentRoot);
            // SYLMOD17102008 for parallel trials               
            //bool CompareIfLowerBoundImprovesAction(int action, int observation, int xstate, BeliefTreeNode & currentNode);
            double calculateLowerBoundValue(int action, BeliefTreeNode & currentNode);
            int chooseAction(BeliefTreeNode & currentNode);

            bool doRandomization;
            void setRandomization(bool newFlag);
            ~SampleBP(){}
    };//end class SampleBP

};//end namespace momdp
#endif