| 1 | <?xml version="1.0" encoding="ISO-8859-1" standalone="no"?> |
|---|
| 2 | <html xmlns="http://www.w3.org/1999/xhtml"><head><title>Chapter 4. |
|---|
| 3 | Getting Good Bounds in CBC |
|---|
| 4 | </title><meta name="generator" content="DocBook XSL Stylesheets V1.61.2"/><link rel="home" href="index.html" title="CBC User Guide"/><link rel="up" href="index.html" title="CBC User Guide"/><link rel="previous" href="ch03.html" title="Chapter 3. Selecting the Next Node in the Search Tree "/><link rel="next" href="ch05.html" title="Chapter 5. Branching "/></head><body><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center">Chapter 4. |
|---|
| 5 | Getting Good Bounds in CBC |
|---|
| 6 | </th></tr><tr><td width="20%" align="left"><a accesskey="p" href="ch03.html">Prev</a> </td><th width="60%" align="center"> </th><td width="20%" align="right"> <a accesskey="n" href="ch05.html">Next</a></td></tr></table><hr/></div><div class="chapter" lang="en"><div class="titlepage"><div><div><h2 class="title"><a id="hueristicChap"/>Chapter 4. |
|---|
| 7 | Getting Good Bounds in CBC |
|---|
| 8 | </h2></div></div><div/></div><div class="toc"><p><b>Table of Contents</b></p><dl><dt><a href="ch04.html#heuristics">CbcHeuristic - Heuristic Methods</a></dt></dl></div><div class="section" lang="en"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="heuristics"/>CbcHeuristic - Heuristic Methods</h2></div></div><div/></div><p> |
|---|
| 9 | In practice, it is very useful to get a good solution reasonably fast. Any MIP-feasible solution produces an upper bound, and a good bound will greatly reduce the run time. Good solutions can satisfy the user |
|---|
| 10 | on very large problems where a complete search is impossible. Obviously, heuristics are |
|---|
| 11 | problem dependent, although some do have more general use. |
|---|
| 12 | At present there is only one heuristic in CBC itself, <tt class="classname">CbcRounding</tt>. Hopefully, the number will grow. Other heuristics are in the <tt class="filename">COIN/Cbc/Samples</tt> |
|---|
| 13 | directory. A heuristic tries to obtain a solution to the original |
|---|
| 14 | problem so it only needs to consider the original rows and does not have to use the |
|---|
| 15 | current bounds. CBC provides an abstract base class <tt class="classname">CbcHeuristic</tt> and a rounding heuristic in CBC. |
|---|
| 16 | </p><p> |
|---|
| 17 | This chapter describes how to build a greedy heuristic for a set covering problem, e.g., the miplib problem fast0507. A more general (and efficient) version of the heuristic is in <tt class="filename">CbcHeuristicGreedy.hpp</tt> and <tt class="filename">CbcHeuristicGreedy.cpp</tt> located in the <tt class="filename">COIN/Cbc/Samples</tt> directory, see <a href="ch08.html" title="Chapter 8. More Samples ">Chapter 8, <i> |
|---|
| 18 | More Samples |
|---|
| 19 | </i></a>. |
|---|
| 20 | </p><p> |
|---|
| 21 | The greedy heuristic will leave all variables taking value one at this node of the |
|---|
| 22 | tree at value one, and will initially set all other variables to value zero. |
|---|
| 23 | All variables are then sorted in order of their cost |
|---|
| 24 | divided by the number of entries in rows which are not yet covered. (We may randomize that |
|---|
| 25 | value a bit so that ties will be broken in different ways on different runs of the heuristic.) |
|---|
| 26 | The best one is choosen, and set to one. The process is repeated. Because this is |
|---|
| 27 | a set covering problem (i.e., all constraints are ≥), the heuristic is guaranteed to find a solution (but not necessarily an improved solution). The speed of the heuristic could be improved by just redoing those affected, but for illustrative purposes we will keep it simple. (The speed could also be improved if all elements are 1.0). |
|---|
| 28 | </p><p> |
|---|
| 29 | The key <tt class="classname">CbcHeuristic</tt> method is <tt class="function">int solution(double & solutionValue, |
|---|
| 30 | double * betterSolution)</tt>. |
|---|
| 31 | The <tt class="function">solution()</tt> method returns 0 if no solution found, and returns 1 if a solution is found, in which case it fills in the objective value and primal solution. The code in <tt class="filename">CbcHeuristicGreedy.cpp</tt> is a little more complicated than this following example. For instance, the code here assumes all variables are integer. The important bit of data is a copy of the matrix (stored by column) before any cuts have been made. The data used are bounds, objective and the matrix plus two work arrays. |
|---|
| 32 | </p><div class="example"><a id="id3000043"/><p class="title"><b>Example 4.1. Data</b></p><pre class="programlisting"> |
|---|
| 33 | |
|---|
| 34 | OsiSolverInterface * solver = model_->solver(); // Get solver from CbcModel |
|---|
| 35 | const double * columnLower = solver->getColLower(); // Column Bounds |
|---|
| 36 | const double * columnUpper = solver->getColUpper(); |
|---|
| 37 | const double * rowLower = solver->getRowLower(); // We know we only need lower bounds |
|---|
| 38 | const double * solution = solver->getColSolution(); |
|---|
| 39 | const double * objective = solver->getObjCoefficients(); // In code we also use min/max |
|---|
| 40 | double integerTolerance = model_->getDblParam(CbcModel::CbcIntegerTolerance); |
|---|
| 41 | double primalTolerance; |
|---|
| 42 | solver->getDblParam(OsiPrimalTolerance,primalTolerance); |
|---|
| 43 | int numberRows = originalNumberRows_; // This is number of rows when matrix was passed in |
|---|
| 44 | // Column copy of matrix (before cuts) |
|---|
| 45 | const double * element = matrix_.getElements(); |
|---|
| 46 | const int * row = matrix_.getIndices(); |
|---|
| 47 | const CoinBigIndex * columnStart = matrix_.getVectorStarts(); |
|---|
| 48 | const int * columnLength = matrix_.getVectorLengths(); |
|---|
| 49 | |
|---|
| 50 | // Get solution array for heuristic solution |
|---|
| 51 | int numberColumns = solver->getNumCols(); |
|---|
| 52 | double * newSolution = new double [numberColumns]; |
|---|
| 53 | // And to sum row activities |
|---|
| 54 | double * rowActivity = new double[numberRows]; |
|---|
| 55 | |
|---|
| 56 | </pre></div><p> |
|---|
| 57 | The <tt class="varname">newSolution</tt> is then initialized to the rounded down solution. |
|---|
| 58 | </p><div class="example"><a id="id3000072"/><p class="title"><b>Example 4.2. Initialize <tt class="varname">newSolution</tt></b></p><pre class="programlisting"> |
|---|
| 59 | |
|---|
| 60 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
|---|
| 61 | CoinBigIndex j; |
|---|
| 62 | double value = solution[iColumn]; |
|---|
| 63 | // Round down integer |
|---|
| 64 | if (fabs(floor(value+0.5)-value)<integerTolerance) |
|---|
| 65 | value=floor(CoinMax(value+1.0e-3,columnLower[iColumn])); |
|---|
| 66 | // make sure clean |
|---|
| 67 | value = CoinMin(value,columnUpper[iColumn]); |
|---|
| 68 | value = CoinMax(value,columnLower[iColumn]); |
|---|
| 69 | newSolution[iColumn]=value; |
|---|
| 70 | if (value) { |
|---|
| 71 | double cost = objective[iColumn]; |
|---|
| 72 | newSolutionValue += value*cost; |
|---|
| 73 | for (j=columnStart[iColumn]; |
|---|
| 74 | j<columnStart[iColumn]+columnLength[iColumn];j++) { |
|---|
| 75 | int iRow=row[j]; |
|---|
| 76 | rowActivity[iRow] += value*element[j]; |
|---|
| 77 | } |
|---|
| 78 | } |
|---|
| 79 | } |
|---|
| 80 | |
|---|
| 81 | </pre></div><p> |
|---|
| 82 | |
|---|
| 83 | |
|---|
| 84 | At this point some row activities are below their lower bound. To correct the infeasibility, the variable which is cheapest in reducing the sum of infeasibilities is found and updated, and the process repeats. This is a finite process. (The implementation could be faster, but is kept simple for illustrative purposes.) |
|---|
| 85 | </p><div class="example"><a id="id3000153"/><p class="title"><b>Example 4.3. Create Feasible <tt class="varname">newSolution</tt> from Initial <tt class="varname">newSolution</tt></b></p><pre class="programlisting"> |
|---|
| 86 | |
|---|
| 87 | while (true) { |
|---|
| 88 | // Get column with best ratio |
|---|
| 89 | int bestColumn=-1; |
|---|
| 90 | double bestRatio=COIN_DBL_MAX; |
|---|
| 91 | for (int iColumn=0;iColumn<numberColumns;iColumn++) { |
|---|
| 92 | CoinBigIndex j; |
|---|
| 93 | double value = newSolution[iColumn]; |
|---|
| 94 | double cost = direction * objective[iColumn]; |
|---|
| 95 | // we could use original upper rather than current |
|---|
| 96 | if (value+0.99<columnUpper[iColumn]) { |
|---|
| 97 | double sum=0.0; // Compute how much we will reduce infeasibility by |
|---|
| 98 | for (j=columnStart[iColumn]; |
|---|
| 99 | j<columnStart[iColumn]+columnLength[iColumn];j++) { |
|---|
| 100 | int iRow=row[j]; |
|---|
| 101 | double gap = rowLower[iRow]-rowActivity[iRow]; |
|---|
| 102 | if (gap>1.0e-7) { |
|---|
| 103 | sum += CoinMin(element[j],gap); |
|---|
| 104 | if (element[j]+rowActivity[iRow]<rowLower[iRow]+1.0e-7) { |
|---|
| 105 | sum += element[j]; |
|---|
| 106 | } |
|---|
| 107 | } |
|---|
| 108 | if (sum>0.0) { |
|---|
| 109 | double ratio = (cost/sum)*(1.0+0.1*CoinDrand48()); |
|---|
| 110 | if (ratio<bestRatio) { |
|---|
| 111 | bestRatio=ratio; |
|---|
| 112 | bestColumn=iColumn; |
|---|
| 113 | } |
|---|
| 114 | } |
|---|
| 115 | } |
|---|
| 116 | } |
|---|
| 117 | if (bestColumn<0) |
|---|
| 118 | break; // we have finished |
|---|
| 119 | // Increase chosen column |
|---|
| 120 | newSolution[bestColumn] += 1.0; |
|---|
| 121 | double cost = direction * objective[bestColumn]; |
|---|
| 122 | newSolutionValue += cost; |
|---|
| 123 | for (CoinBigIndex j=columnStart[bestColumn]; |
|---|
| 124 | j<columnStart[bestColumn]+columnLength[bestColumn];j++) { |
|---|
| 125 | int iRow = row[j]; |
|---|
| 126 | rowActivity[iRow] += element[j]; |
|---|
| 127 | } |
|---|
| 128 | } |
|---|
| 129 | |
|---|
| 130 | </pre></div><p> |
|---|
| 131 | A solution value of <tt class="varname">newSolution</tt> is compared to the best solution value. If <tt class="varname">newSolution</tt> is an improvement, its feasibility is validated. We expect <tt class="varname">newSolution</tt> to be feasible, and are trapping for unexpected numerical errors. |
|---|
| 132 | </p><div class="example"><a id="id3000197"/><p class="title"><b>Example 4.4. Check Solution Quality of <tt class="varname">newSolution</tt></b></p><pre class="programlisting"> |
|---|
| 133 | |
|---|
| 134 | returnCode=0; // 0 means no good solution |
|---|
| 135 | if (newSolutionValue<solutionValue) { // minimization |
|---|
| 136 | // check feasible |
|---|
| 137 | memset(rowActivity,0,numberRows*sizeof(double)); |
|---|
| 138 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
|---|
| 139 | CoinBigIndex j; |
|---|
| 140 | double value = newSolution[iColumn]; |
|---|
| 141 | if (value) { |
|---|
| 142 | for (j=columnStart[iColumn]; |
|---|
| 143 | j<columnStart[iColumn]+columnLength[iColumn];j++) { |
|---|
| 144 | int iRow=row[j]; |
|---|
| 145 | rowActivity[iRow] += value*element[j]; |
|---|
| 146 | } |
|---|
| 147 | } |
|---|
| 148 | } |
|---|
| 149 | // check was approximately feasible |
|---|
| 150 | bool feasible=true; |
|---|
| 151 | for (iRow=0;iRow<numberRows;iRow++) { |
|---|
| 152 | if(rowActivity[iRow]<rowLower[iRow]) { |
|---|
| 153 | if (rowActivity[iRow]<rowLower[iRow]-10.0*primalTolerance) |
|---|
| 154 | feasible = false; |
|---|
| 155 | } |
|---|
| 156 | } |
|---|
| 157 | if (feasible) { |
|---|
| 158 | // new solution |
|---|
| 159 | memcpy(betterSolution,newSolution,numberColumns*sizeof(double)); |
|---|
| 160 | solutionValue = newSolutionValue; |
|---|
| 161 | // We have good solution |
|---|
| 162 | returnCode=1; |
|---|
| 163 | } |
|---|
| 164 | } |
|---|
| 165 | |
|---|
| 166 | </pre></div></div></div><div class="navfooter"><hr/><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="ch03.html">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="index.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="ch05.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">Chapter 3. |
|---|
| 167 | Selecting the Next Node in the Search Tree |
|---|
| 168 | </td><td width="20%" align="center"><a accesskey="h" href="index.html">Home</a></td><td width="40%" align="right" valign="top"> Chapter 5. |
|---|
| 169 | Branching |
|---|
| 170 | </td></tr></table></div></body></html> |
|---|
