// Copyright (C) 2004, International Business Machines // Corporation and others. All Rights Reserved. #ifndef CbcTreeLocal_H #define CbcTreeLocal_H //############################################################################# /* This implements (approximately) local branching as in the 2002 paper by Matteo Fischetti and Andrea Lodi. The very simple version of the algorithm for problems with 0-1 variables and continuous is as follows: Obtain a feasible solution (one can be passed in). Add a cut which limits search to a k neighborhood of this solution. (At most k 0-1 variables may change value) Do branch and bound on this problem. If finished search and proven optimal then we can reverse cut so any solutions must be at least k+1 away from solution and we can add a new cut limiting search to a k neighborhood of new solution repeat. If finished search and no new solution then the simplest version would reverse last cut and complete search. The version implemented here can use time and node limits and can widen search (increase effective k) .... and more */ #include "CbcTree.hpp" #include "CbcNode.hpp" #include "OsiRowCut.hpp" class CbcModel; class CbcTreeLocal : public CbcTree { public: // Default Constructor CbcTreeLocal (); /* Constructor with solution. If solution NULL no solution, otherwise must be integer range is initial upper bound (k) on difference from given solution. typeCuts - 0 means just 0-1 cuts and will need to refine 0-1 solution 1 uses weaker cuts on all integer variables maxDiversification is maximum number of range widenings to try timeLimit is seconds in subTree nodeLimit is nodes in subTree refine is whether to see if we can prove current solution is optimal when we fix all 0-1 (in case typeCuts==0 and there are general integer variables) if false then no refinement but reverse cuts weaker */ CbcTreeLocal (CbcModel * model,const double * solution ,int range=10, int typeCuts=0,int maxDiversification=0, int timeLimit=1000000, int nodeLimit=1000000,bool refine=true); // Copy constructor CbcTreeLocal ( const CbcTreeLocal & rhs); // = operator CbcTreeLocal & operator=(const CbcTreeLocal & rhs); virtual ~CbcTreeLocal(); /// Clone virtual CbcTree * clone() const; /// Create C++ lines to get to current state virtual void generateCpp( FILE * fp) ; /*! \name Heap access and maintenance methods */ //@{ /// Return the top node of the heap virtual CbcNode * top() const; /// Add a node to the heap virtual void push(CbcNode * x); /// Remove the top node from the heap virtual void pop() ; //@} /*! \name Other stuff */ //@{ /// Create cut - return -1 if bad, 0 if okay and 1 if cut is everything int createCut(const double * solution, OsiRowCut & cut); /// Test if empty *** note may be overridden virtual bool empty() ; /// We may have got an intelligent tree so give it one more chance virtual void endSearch() ; /// Other side of last cut branch (if bias==rhs_ will be weakest possible) void reverseCut(int state, double bias=0.0); /// Delete last cut branch void deleteCut(OsiRowCut & cut); /// Pass in solution (so can be used after heuristic) void passInSolution(const double * solution, double solutionValue); // range i.e. k inline int range() const { return range_;} // setrange i.e. k inline void setRange(int value) { range_ = value;} // Type of cuts - 0=just 0-1, 1=all inline int typeCuts() const { return typeCuts_;} // Type of cuts - 0=just 0-1, 1=all inline void setTypeCuts(int value) { typeCuts_ = value;} // maximum number of diversifications inline int maxDiversification() const { return maxDiversification_;} // maximum number of diversifications inline void setMaxDiversification(int value) { maxDiversification_ = value;} // time limit per subtree inline int timeLimit() const { return timeLimit_;} // time limit per subtree inline void setTimeLimit(int value) { timeLimit_ = value;} // node limit for subtree inline int nodeLimit() const { return nodeLimit_;} // node limit for subtree inline void setNodeLimit(int value) { nodeLimit_ = value;} // Whether to do refinement step inline bool refine() const { return refine_;} // Whether to do refinement step inline void setRefine(bool yesNo) { refine_ = yesNo;} //@} private: // Node for local cuts CbcNode * localNode_; // best solution double * bestSolution_; // saved solution double * savedSolution_; // solution number at start of pass int saveNumberSolutions_; /* Cut. If zero size then no solution yet. Otherwise is left hand branch */ OsiRowCut cut_; // This cut fixes all 0-1 variables OsiRowCut fixedCut_; // Model CbcModel * model_; // Original lower bounds double * originalLower_; // Original upper bounds double * originalUpper_; // range i.e. k int range_; // Type of cuts - 0=just 0-1, 1=all int typeCuts_; // maximum number of diversifications int maxDiversification_; // current diversification int diversification_; // Whether next will be strong diversification bool nextStrong_; // Current rhs double rhs_; // Save allowable gap double savedGap_; // Best solution double bestCutoff_; // time limit per subtree int timeLimit_; // time when subtree started int startTime_; // node limit for subtree int nodeLimit_; // node count when subtree started int startNode_; // -1 not started, 0 == stop on first solution, 1 don't stop on first, 2 refinement step int searchType_; // Whether to do refinement step bool refine_; }; #endif