ClpMatrixBase.hpp

/* $Id$ */
// Copyright (C) 2002, International Business Machines
// Corporation and others.  All Rights Reserved.
// This code is licensed under the terms of the Eclipse Public License (EPL).

#ifndef ClpMatrixBase_H
#define ClpMatrixBase_H

#include "CoinPragma.hpp"
#include "CoinTypes.hpp"

#include "CoinPackedMatrix.hpp"
class CoinIndexedVector;
class ClpSimplex;
class ClpModel;
// Compilers can produce better code if they know about __restrict
#ifndef COIN_RESTRICT
#ifdef COIN_USE_RESTRICT
#define COIN_RESTRICT __restrict
#else
#define COIN_RESTRICT
#endif
#endif

/** Abstract base class for Clp Matrices

Since this class is abstract, no object of this type can be created.

If a derived class provides all methods then all Clp algorithms
should work.  Some can be very inefficient e.g. getElements etc is
only used for tightening bounds for dual and the copies are
deleted.  Many methods can just be dummy i.e. abort(); if not
all features are being used.  So if column generation was being done
then it makes no sense to do steepest edge so there would be
no point providing subsetTransposeTimes.
*/

class ClpMatrixBase {

public:
  /**@name Virtual methods that the derived classes must provide */
  //@{
  /// Return a complete CoinPackedMatrix
  virtual CoinPackedMatrix *getPackedMatrix() const = 0;
  /** Whether the packed matrix is column major ordered or not. */
  virtual bool isColOrdered() const = 0;
  /** Number of entries in the packed matrix. */
  virtual CoinBigIndex getNumElements() const = 0;
  /** Number of columns. */
  virtual int getNumCols() const = 0;
  /** Number of rows. */
  virtual int getNumRows() const = 0;

  /** A vector containing the elements in the packed matrix. Note that there
         might be gaps in this list, entries that do not belong to any
         major-dimension vector. To get the actual elements one should look at
         this vector together with vectorStarts and vectorLengths. */
  virtual const double *getElements() const = 0;
  /** A vector containing the minor indices of the elements in the packed
         matrix. Note that there might be gaps in this list, entries that do not
         belong to any major-dimension vector. To get the actual elements one
         should look at this vector together with vectorStarts and
         vectorLengths. */
  virtual const int *getIndices() const = 0;

  virtual const CoinBigIndex *getVectorStarts() const = 0;
  /** The lengths of the major-dimension vectors. */
  virtual const int *getVectorLengths() const = 0;
  /** The length of a single major-dimension vector. */
  virtual int getVectorLength(int index) const;
  /** Delete the columns whose indices are listed in <code>indDel</code>. */
  virtual void deleteCols(const int numDel, const int *indDel) = 0;
  /** Delete the rows whose indices are listed in <code>indDel</code>. */
  virtual void deleteRows(const int numDel, const int *indDel) = 0;
#ifndef CLP_NO_VECTOR
  /// Append Columns
  virtual void appendCols(int number, const CoinPackedVectorBase *const *columns);
  /// Append Rows
  virtual void appendRows(int number, const CoinPackedVectorBase *const *rows);
#endif
  /** Modify one element of packed matrix.  An element may be added.
         This works for either ordering If the new element is zero it will be
         deleted unless keepZero true */
  virtual void modifyCoefficient(int row, int column, double newElement,
    bool keepZero = false);
  /** Append a set of rows/columns to the end of the matrix. Returns number of errors
         i.e. if any of the new rows/columns contain an index that's larger than the
         number of columns-1/rows-1 (if numberOther>0) or duplicates
         If 0 then rows, 1 if columns */
  virtual int appendMatrix(int number, int type,
    const CoinBigIndex *starts, const int *index,
    const double *element, int numberOther = -1);

  /** Returns a new matrix in reverse order without gaps
         Is allowed to return NULL if doesn't want to have row copy */
  virtual ClpMatrixBase *reverseOrderedCopy() const
  {
    return NULL;
  }

  /// Returns number of elements in column part of basis
  virtual int countBasis(const int *whichColumn,
    int &numberColumnBasic)
    = 0;
  /// Fills in column part of basis
  virtual void fillBasis(ClpSimplex *model,
    const int *whichColumn,
    int &numberColumnBasic,
    int *row, int *start,
    int *rowCount, int *columnCount,
    CoinFactorizationDouble *element)
    = 0;
  /** Creates scales for column copy (rowCopy in model may be modified)
         default does not allow scaling
         returns non-zero if no scaling done */
  virtual int scale(ClpModel *, ClpSimplex * = NULL) const
  {
    return 1;
  }
  /** Scales rowCopy if column copy scaled
         Only called if scales already exist */
  virtual void scaleRowCopy(ClpModel *) const {}
  /// Returns true if can create row copy
  virtual bool canGetRowCopy() const
  {
    return true;
  }
  /** Realy really scales column copy
         Only called if scales already exist.
         Up to user to delete */
  inline virtual ClpMatrixBase *scaledColumnCopy(ClpModel *) const
  {
    return this->clone();
  }

  /** Checks if all elements are in valid range.  Can just
         return true if you are not paranoid.  For Clp I will
         probably expect no zeros.  Code can modify matrix to get rid of
         small elements.
         check bits (can be turned off to save time) :
         1 - check if matrix has gaps
         2 - check if zero elements
         4 - check and compress duplicates
         8 - report on large and small
     */
  virtual bool allElementsInRange(ClpModel *,
    double, double,
    int = 15)
  {
    return true;
  }
  /** Set the dimensions of the matrix. In effect, append new empty
         columns/rows to the matrix. A negative number for either dimension
         means that that dimension doesn't change. Otherwise the new dimensions
         MUST be at least as large as the current ones otherwise an exception
         is thrown. */
  virtual void setDimensions(int numrows, int numcols);
  /** Returns largest and smallest elements of both signs.
         Largest refers to largest absolute value.
         If returns zeros then can't tell anything */
  virtual void rangeOfElements(double &smallestNegative, double &largestNegative,
    double &smallestPositive, double &largestPositive);

  /** Unpacks a column into an CoinIndexedvector
      */
  virtual void unpack(const ClpSimplex *model, CoinIndexedVector *rowArray,
    int column) const = 0;
  /** Unpacks a column into an CoinIndexedvector
      ** in packed format
      Note that model is NOT const.  Bounds and objective could
      be modified if doing column generation (just for this variable) */
  virtual void unpackPacked(ClpSimplex *model,
    CoinIndexedVector *rowArray,
    int column) const = 0;
  /** Purely for column generation and similar ideas.  Allows
         matrix and any bounds or costs to be updated (sensibly).
         Returns non-zero if any changes.
     */
  virtual int refresh(ClpSimplex *)
  {
    return 0;
  }

  // Really scale matrix
  virtual void reallyScale(const double *rowScale, const double *columnScale);
  /** Given positive integer weights for each row fills in sum of weights
         for each column (and slack).
         Returns weights vector
         Default returns vector of ones
     */
  virtual CoinBigIndex *dubiousWeights(const ClpSimplex *model, int *inputWeights) const;
  /** Adds multiple of a column into an CoinIndexedvector
         You can use quickAdd to add to vector */
  virtual void add(const ClpSimplex *model, CoinIndexedVector *rowArray,
    int column, double multiplier) const = 0;
  /** Adds multiple of a column into an array */
  virtual void add(const ClpSimplex *model, double *array,
    int column, double multiplier) const = 0;
  /// Allow any parts of a created CoinPackedMatrix to be deleted
  virtual void releasePackedMatrix() const = 0;
  /// Says whether it can do partial pricing
  virtual bool canDoPartialPricing() const;
  /// Returns number of hidden rows e.g. gub
  virtual int hiddenRows() const;
  /// Partial pricing
  virtual void partialPricing(ClpSimplex *model, double start, double end,
    int &bestSequence, int &numberWanted);
  /** expands an updated column to allow for extra rows which the main
         solver does not know about and returns number added.

         This will normally be a no-op - it is in for GUB but may get extended to
         general non-overlapping and embedded networks.

         mode 0 - extend
         mode 1 - delete etc
     */
  virtual int extendUpdated(ClpSimplex *model, CoinIndexedVector *update, int mode);
  /**
        utility primal function for dealing with dynamic constraints
        mode=0  - Set up before "update" and "times" for primal solution using extended rows
        mode=1  - Cleanup primal solution after "times" using extended rows.
        mode=2  - Check (or report on) primal infeasibilities
     */
  virtual void primalExpanded(ClpSimplex *model, int mode);
  /**
         utility dual function for dealing with dynamic constraints
         mode=0  - Set up before "updateTranspose" and "transposeTimes" for duals using extended
                   updates array (and may use other if dual values pass)
         mode=1  - Update dual solution after "transposeTimes" using extended rows.
         mode=2  - Compute all djs and compute key dual infeasibilities
         mode=3  - Report on key dual infeasibilities
         mode=4  - Modify before updateTranspose in partial pricing
     */
  virtual void dualExpanded(ClpSimplex *model, CoinIndexedVector *array,
    double *other, int mode);
  /**
         general utility function for dealing with dynamic constraints
         mode=0  - Create list of non-key basics in pivotVariable_ using
                   number as numberBasic in and out
         mode=1  - Set all key variables as basic
         mode=2  - return number extra rows needed, number gives maximum number basic
         mode=3  - before replaceColumn
         mode=4  - return 1 if can do primal, 2 if dual, 3 if both
         mode=5  - save any status stuff (when in good state)
         mode=6  - restore status stuff
         mode=7  - flag given variable (normally sequenceIn)
         mode=8  - unflag all variables
         mode=9  - synchronize costs and bounds
         mode=10  - return 1 if there may be changing bounds on variable (column generation)
         mode=11  - make sure set is clean (used when a variable rejected - but not flagged)
         mode=12  - after factorize but before permute stuff
         mode=13  - at end of simplex to delete stuff

     */
  virtual int generalExpanded(ClpSimplex *model, int mode, int &number);
  /**
        update information for a pivot (and effective rhs)
     */
  virtual int updatePivot(ClpSimplex *model, double oldInValue, double oldOutValue);
  /** Creates a variable.  This is called after partial pricing and may modify matrix.
         May update bestSequence.
     */
  virtual void createVariable(ClpSimplex *model, int &bestSequence);
  /** Just for debug if odd type matrix.
         Returns number of primal infeasibilities. */
  virtual int checkFeasible(ClpSimplex *model, double &sum) const;
  /// Returns reduced cost of a variable
  double reducedCost(ClpSimplex *model, int sequence) const;
  /// Correct sequence in and out to give true value (if both -1 maybe do whole matrix)
  virtual void correctSequence(const ClpSimplex *model, int &sequenceIn, int &sequenceOut);
  //@}

  //---------------------------------------------------------------------------
  /**@name Matrix times vector methods
        They can be faster if scalar is +- 1
        Also for simplex I am not using basic/non-basic split */
  //@{
  /** Return <code>y + A * x * scalar</code> in <code>y</code>.
         @pre <code>x</code> must be of size <code>numColumns()</code>
         @pre <code>y</code> must be of size <code>numRows()</code> */
  virtual void times(double scalar,
    const double *COIN_RESTRICT x, double *COIN_RESTRICT y) const = 0;
  /** And for scaling - default aborts for when scaling not supported
         (unless pointers NULL when as normal)
     */
  virtual void times(double scalar,
    const double *COIN_RESTRICT x, double *COIN_RESTRICT y,
    const double *COIN_RESTRICT rowScale,
    const double *COIN_RESTRICT columnScale) const;
  /** Return <code>y + x * scalar * A</code> in <code>y</code>.
         @pre <code>x</code> must be of size <code>numRows()</code>
         @pre <code>y</code> must be of size <code>numColumns()</code> */
  virtual void transposeTimes(double scalar,
    const double *COIN_RESTRICT x, double *COIN_RESTRICT y) const = 0;
  /** And for scaling - default aborts for when scaling not supported
         (unless pointers NULL when as normal)
     */
  virtual void transposeTimes(double scalar,
    const double *COIN_RESTRICT x, double *COIN_RESTRICT y,
    const double *COIN_RESTRICT rowScale,
    const double *COIN_RESTRICT columnScale,
    double *COIN_RESTRICT spare = NULL) const;
#if COIN_LONG_WORK
  // For long double versions (aborts if not supported)
  virtual void times(CoinWorkDouble scalar,
    const CoinWorkDouble *COIN_RESTRICT x, CoinWorkDouble *COIN_RESTRICT y) const;
  virtual void transposeTimes(CoinWorkDouble scalar,
    const CoinWorkDouble *COIN_RESTRICT x, CoinWorkDouble *COIN_RESTRICT y) const;
#endif
  /** Return <code>x * scalar *A + y</code> in <code>z</code>.
         Can use y as temporary array (will be empty at end)
         Note - If x packed mode - then z packed mode
         Squashes small elements and knows about ClpSimplex */
  virtual void transposeTimes(const ClpSimplex *model, double scalar,
    const CoinIndexedVector *x,
    CoinIndexedVector *y,
    CoinIndexedVector *z) const = 0;
  /** Return <code>x *A</code> in <code>z</code> but
         just for indices in y.
         This is only needed for primal steepest edge.
         Note - z always packed mode */
  virtual void subsetTransposeTimes(const ClpSimplex *model,
    const CoinIndexedVector *x,
    const CoinIndexedVector *y,
    CoinIndexedVector *z) const = 0;
  /** Returns true if can combine transposeTimes and subsetTransposeTimes
         and if it would be faster */
  virtual bool canCombine(const ClpSimplex *,
    const CoinIndexedVector *) const
  {
    return false;
  }
  /** Updates two arrays for steepest and does devex weights 
	 (need not be coded)
	 Returns nonzero if updates reduced cost and infeas -
	 new infeas in dj1 */
  virtual int transposeTimes2(const ClpSimplex *model,
    const CoinIndexedVector *pi1, CoinIndexedVector *dj1,
    const CoinIndexedVector *pi2,
    CoinIndexedVector *spare,
    double *infeas, double *reducedCost,
    double referenceIn, double devex,
    // Array for exact devex to say what is in reference framework
    unsigned int *reference,
    double *weights, double scaleFactor);
  /// Updates second array for steepest and does devex weights (need not be coded)
  virtual void subsetTimes2(const ClpSimplex *model,
    CoinIndexedVector *dj1,
    const CoinIndexedVector *pi2, CoinIndexedVector *dj2,
    double referenceIn, double devex,
    // Array for exact devex to say what is in reference framework
    unsigned int *reference,
    double *weights, double scaleFactor);
  /** Return <code>x *A</code> in <code>z</code> but
         just for number indices in y.
         Default cheats with fake CoinIndexedVector and
         then calls subsetTransposeTimes */
  virtual void listTransposeTimes(const ClpSimplex *model,
    double *x,
    int *y,
    int number,
    double *z) const;
  //@}
  //@{
  ///@name Other
  /// Clone
  virtual ClpMatrixBase *clone() const = 0;
  /** Subset clone (without gaps).  Duplicates are allowed
         and order is as given.
         Derived classes need not provide this as it may not always make
         sense */
  virtual ClpMatrixBase *subsetClone(
    int numberRows, const int *whichRows,
    int numberColumns, const int *whichColumns) const;
  /// Gets rid of any mutable by products
  virtual void backToBasics() {}
  /** Returns type.
         The types which code may need to know about are:
         1  - ClpPackedMatrix
         11 - ClpNetworkMatrix
         12 - ClpPlusMinusOneMatrix
     */
  inline int type() const
  {
    return type_;
  }
  /// Sets type
  void setType(int newtype)
  {
    type_ = newtype;
  }
  /// Sets up an effective RHS
  void useEffectiveRhs(ClpSimplex *model);
  /** Returns effective RHS offset if it is being used.  This is used for long problems
         or big gub or anywhere where going through full columns is
         expensive.  This may re-compute */
  virtual double *rhsOffset(ClpSimplex *model, bool forceRefresh = false,
    bool check = false);
  /// If rhsOffset used this is iteration last refreshed
  inline int lastRefresh() const
  {
    return lastRefresh_;
  }
  /// If rhsOffset used this is refresh frequency (0==off)
  inline int refreshFrequency() const
  {
    return refreshFrequency_;
  }
  inline void setRefreshFrequency(int value)
  {
    refreshFrequency_ = value;
  }
  /// whether to skip dual checks most of time
  inline bool skipDualCheck() const
  {
    return skipDualCheck_;
  }
  inline void setSkipDualCheck(bool yes)
  {
    skipDualCheck_ = yes;
  }
  /** Partial pricing tuning parameter - minimum number of "objects" to scan.
         e.g. number of Gub sets but could be number of variables */
  inline int minimumObjectsScan() const
  {
    return minimumObjectsScan_;
  }
  inline void setMinimumObjectsScan(int value)
  {
    minimumObjectsScan_ = value;
  }
  /// Partial pricing tuning parameter - minimum number of negative reduced costs to get
  inline int minimumGoodReducedCosts() const
  {
    return minimumGoodReducedCosts_;
  }
  inline void setMinimumGoodReducedCosts(int value)
  {
    minimumGoodReducedCosts_ = value;
  }
  /// Current start of search space in matrix (as fraction)
  inline double startFraction() const
  {
    return startFraction_;
  }
  inline void setStartFraction(double value)
  {
    startFraction_ = value;
  }
  /// Current end of search space in matrix (as fraction)
  inline double endFraction() const
  {
    return endFraction_;
  }
  inline void setEndFraction(double value)
  {
    endFraction_ = value;
  }
  /// Current best reduced cost
  inline double savedBestDj() const
  {
    return savedBestDj_;
  }
  inline void setSavedBestDj(double value)
  {
    savedBestDj_ = value;
  }
  /// Initial number of negative reduced costs wanted
  inline int originalWanted() const
  {
    return originalWanted_;
  }
  inline void setOriginalWanted(int value)
  {
    originalWanted_ = value;
  }
  /// Current number of negative reduced costs which we still need
  inline int currentWanted() const
  {
    return currentWanted_;
  }
  inline void setCurrentWanted(int value)
  {
    currentWanted_ = value;
  }
  /// Current best sequence
  inline int savedBestSequence() const
  {
    return savedBestSequence_;
  }
  inline void setSavedBestSequence(int value)
  {
    savedBestSequence_ = value;
  }
  //@}

protected:
  /**@name Constructors, destructor<br>
        <strong>NOTE</strong>: All constructors are protected. There's no need
        to expose them, after all, this is an abstract class. */
  //@{
  /** Default constructor. */
  ClpMatrixBase();
  /** Destructor (has to be public) */
public:
  virtual ~ClpMatrixBase();

protected:
  // Copy
  ClpMatrixBase(const ClpMatrixBase &);
  // Assignment
  ClpMatrixBase &operator=(const ClpMatrixBase &);
  //@}

protected:
  /**@name Data members
        The data members are protected to allow access for derived classes. */
  //@{
  /** Effective RHS offset if it is being used.  This is used for long problems
         or big gub or anywhere where going through full columns is
         expensive */
  double *rhsOffset_;
  /// Current start of search space in matrix (as fraction)
  double startFraction_;
  /// Current end of search space in matrix (as fraction)
  double endFraction_;
  /// Best reduced cost so far
  double savedBestDj_;
  /// Initial number of negative reduced costs wanted
  int originalWanted_;
  /// Current number of negative reduced costs which we still need
  int currentWanted_;
  /// Saved best sequence in pricing
  int savedBestSequence_;
  /// type (may be useful)
  int type_;
  /// If rhsOffset used this is iteration last refreshed
  int lastRefresh_;
  /// If rhsOffset used this is refresh frequency (0==off)
  int refreshFrequency_;
  /// Partial pricing tuning parameter - minimum number of "objects" to scan
  int minimumObjectsScan_;
  /// Partial pricing tuning parameter - minimum number of negative reduced costs to get
  int minimumGoodReducedCosts_;
  /// True sequence in (i.e. from larger problem)
  int trueSequenceIn_;
  /// True sequence out (i.e. from larger problem)
  int trueSequenceOut_;
  /// whether to skip dual checks most of time
  bool skipDualCheck_;
  //@}
};
// bias for free variables
#define FREE_BIAS 1.0e1
// Acceptance criteria for free variables
#define FREE_ACCEPT 1.0e2

#endif

/* vi: softtabstop=2 shiftwidth=2 expandtab tabstop=2
*/