1 | // $Id: qmip2.cpp 1902 2013-04-10 16:58:16Z stefan $ |
---|
2 | // Copyright (C) 2006, International Business Machines |
---|
3 | // Corporation and others. All Rights Reserved. |
---|
4 | // This code is licensed under the terms of the Eclipse Public License (EPL). |
---|
5 | |
---|
6 | #include <cassert> |
---|
7 | |
---|
8 | #include "CoinPragma.hpp" |
---|
9 | |
---|
10 | // For Branch and bound |
---|
11 | #include "OsiClpSolverInterface.hpp" |
---|
12 | #include "CbcModel.hpp" |
---|
13 | #include "CbcCutGenerator.hpp" |
---|
14 | #include "CoinHelperFunctions.hpp" |
---|
15 | #include "CbcStrategy.hpp" |
---|
16 | |
---|
17 | // Need stored cuts |
---|
18 | |
---|
19 | #include "CglStored.hpp" |
---|
20 | |
---|
21 | // For saying about solution validity |
---|
22 | #include "OsiAuxInfo.hpp" |
---|
23 | |
---|
24 | |
---|
25 | // Time |
---|
26 | #include "CoinTime.hpp" |
---|
27 | // Class to disallow strong branching solutions |
---|
28 | #include "CbcFeasibilityBase.hpp" |
---|
29 | class CbcFeasibilityNoStrong : public CbcFeasibilityBase{ |
---|
30 | public: |
---|
31 | // Default Constructor |
---|
32 | CbcFeasibilityNoStrong () {} |
---|
33 | |
---|
34 | virtual ~CbcFeasibilityNoStrong() {} |
---|
35 | // Copy constructor |
---|
36 | CbcFeasibilityNoStrong ( const CbcFeasibilityNoStrong &rhs) {} |
---|
37 | |
---|
38 | // Assignment operator |
---|
39 | CbcFeasibilityNoStrong & operator=( const CbcFeasibilityNoStrong& rhs) |
---|
40 | { return * this;} |
---|
41 | |
---|
42 | /// Clone |
---|
43 | virtual CbcFeasibilityBase * clone() const |
---|
44 | { return new CbcFeasibilityNoStrong();} |
---|
45 | |
---|
46 | /** |
---|
47 | On input mode: |
---|
48 | 0 - called after a solve but before any cuts |
---|
49 | -1 - called after strong branching |
---|
50 | Returns : |
---|
51 | 0 - no opinion |
---|
52 | -1 pretend infeasible |
---|
53 | 1 pretend integer solution |
---|
54 | */ |
---|
55 | virtual int feasible(CbcModel * model, int mode) |
---|
56 | {return mode;} |
---|
57 | }; |
---|
58 | |
---|
59 | |
---|
60 | /************************************************************************ |
---|
61 | |
---|
62 | This main program solves the following 0-1 problem: |
---|
63 | |
---|
64 | min -x0 - 2x1 - 3x2 - 4x3 |
---|
65 | |
---|
66 | subject to |
---|
67 | |
---|
68 | x0 + x1 + x2 + x3 <= 2 |
---|
69 | |
---|
70 | and quadratic constraints with positive random numbers |
---|
71 | |
---|
72 | It does it creating extra yij variables and constraints xi + xj -1 <= yij |
---|
73 | and putting quadratic elements on y |
---|
74 | |
---|
75 | The extra constraints are treated as stored cuts. |
---|
76 | |
---|
77 | This is to show how to keep branching even if we have a solution |
---|
78 | |
---|
79 | ************************************************************************/ |
---|
80 | |
---|
81 | int main (int argc, const char *argv[]) |
---|
82 | { |
---|
83 | |
---|
84 | // Define a Solver which inherits from OsiClpsolverInterface -> OsiSolverInterface |
---|
85 | |
---|
86 | OsiClpSolverInterface solver1; |
---|
87 | |
---|
88 | int nX=4; |
---|
89 | |
---|
90 | int nY = (nX * (nX-1)/2); |
---|
91 | // All columns |
---|
92 | double * obj = new double [nX+nY]; |
---|
93 | double * clo = new double[nX+nY]; |
---|
94 | double * cup = new double[nX+nY]; |
---|
95 | int i; |
---|
96 | for (i=0;i<nX;i++) { |
---|
97 | obj[i] = -(i+1); |
---|
98 | clo[i]=0.0; |
---|
99 | cup[i]=1.0; |
---|
100 | } |
---|
101 | for (i=nX;i<nX+nY;i++) { |
---|
102 | obj[i] = 0.0; |
---|
103 | clo[i]=0.0; |
---|
104 | cup[i]=1.0; |
---|
105 | } |
---|
106 | // Just ordinary rows |
---|
107 | int nRow = 1+nX; |
---|
108 | double * rlo = new double[nRow]; |
---|
109 | double * rup = new double[nRow]; |
---|
110 | for (i=0;i<nRow;i++) { |
---|
111 | rlo[i]=-COIN_DBL_MAX; |
---|
112 | rup[i]=1.0; |
---|
113 | } |
---|
114 | // and first row |
---|
115 | rup[0]=nX/2.0; |
---|
116 | // Matrix |
---|
117 | int nEl = nX+nX*nX; |
---|
118 | int * row = new int[nEl]; |
---|
119 | int * col = new int[nEl]; |
---|
120 | double * el = new double[nEl]; |
---|
121 | // X |
---|
122 | nEl=0; |
---|
123 | // May need scale to make plausible |
---|
124 | double scaleFactor = 1.0; |
---|
125 | for (i=0;i<nX;i++) { |
---|
126 | row[nEl]=0; |
---|
127 | col[nEl]=i; |
---|
128 | el[nEl++]=1.0; |
---|
129 | // and diagonal |
---|
130 | row[nEl]=i+1; |
---|
131 | col[nEl]=i; |
---|
132 | double value = CoinDrand48()*scaleFactor; |
---|
133 | // make reasonable (so multiples of 0.000001) |
---|
134 | value *= 1.0e6; |
---|
135 | int iValue = (int) (value+1.0); |
---|
136 | value = iValue; |
---|
137 | value *= 1.0e-6; |
---|
138 | el[nEl++]=value; |
---|
139 | } |
---|
140 | // Y |
---|
141 | nY = nX; |
---|
142 | // And stored cuts |
---|
143 | CglStored stored; |
---|
144 | double cutEls[3]={1.0,1.0,-1.0}; |
---|
145 | int cutIndices[3]; |
---|
146 | for (i=0;i<nX;i++) { |
---|
147 | cutIndices[0]=i; |
---|
148 | for (int j=i+1;j<nX;j++) { |
---|
149 | cutIndices[1]=j; |
---|
150 | cutIndices[2]=nY; |
---|
151 | // add cut |
---|
152 | stored.addCut(-COIN_DBL_MAX,1.0,3,cutIndices,cutEls); |
---|
153 | row[nEl]=i+1; |
---|
154 | col[nEl]=nY; |
---|
155 | double value = CoinDrand48()*scaleFactor; |
---|
156 | // multiply to make ones with most negative objective violated |
---|
157 | // make reasonable (so multiples of 0.000001) |
---|
158 | value *= 1.0e6+1.0e6*j; |
---|
159 | int iValue = (int) (value+1.0); |
---|
160 | value = iValue; |
---|
161 | value *= 1.0e-6; |
---|
162 | el[nEl++]=value; |
---|
163 | // and other |
---|
164 | if (i!=j) { |
---|
165 | row[nEl]=j+1; |
---|
166 | col[nEl]=nY; |
---|
167 | el[nEl++]=value; |
---|
168 | } |
---|
169 | nY++; |
---|
170 | } |
---|
171 | } |
---|
172 | // Create model |
---|
173 | CoinPackedMatrix matrix(true, row, col, el, nEl); |
---|
174 | solver1.loadProblem(matrix, clo, cup, obj, rlo, rup); |
---|
175 | delete [] obj; |
---|
176 | delete [] clo; |
---|
177 | delete [] cup; |
---|
178 | delete [] rlo; |
---|
179 | delete [] rup; |
---|
180 | delete [] row; |
---|
181 | delete [] col; |
---|
182 | delete [] el; |
---|
183 | // Integers |
---|
184 | for (i=0;i<nX;i++) |
---|
185 | solver1.setInteger(i); |
---|
186 | // Reduce printout |
---|
187 | solver1.setHintParam(OsiDoReducePrint,true,OsiHintTry); |
---|
188 | // This clones solver |
---|
189 | CbcModel model(solver1); |
---|
190 | // Add stored cuts (making sure at all depths) |
---|
191 | model.addCutGenerator(&stored,1,"Stored",true,false,false,-100,1,-1); |
---|
192 | /* You need the next few lines - |
---|
193 | a) so that cut generator will always be called again if it generated cuts |
---|
194 | b) it is known that matrix is not enough to define problem so do cuts even |
---|
195 | if it looks integer feasible at continuous optimum. |
---|
196 | c) a solution found by strong branching will be ignored. |
---|
197 | d) don't recompute a solution once found |
---|
198 | */ |
---|
199 | // Make sure cut generator called correctly (a) |
---|
200 | model.cutGenerator(0)->setMustCallAgain(true); |
---|
201 | // Say cuts needed at continuous (b) |
---|
202 | OsiBabSolver oddCuts; |
---|
203 | oddCuts.setSolverType(4); |
---|
204 | model.passInSolverCharacteristics(&oddCuts); |
---|
205 | // Say no to all solutions by strong branching (c) |
---|
206 | CbcFeasibilityNoStrong noStrong; |
---|
207 | model.setProblemFeasibility(noStrong); |
---|
208 | // Say don't recompute solution d) |
---|
209 | model.setSpecialOptions(4); |
---|
210 | |
---|
211 | double time1 = CoinCpuTime(); |
---|
212 | |
---|
213 | // Do complete search |
---|
214 | |
---|
215 | model.branchAndBound(); |
---|
216 | |
---|
217 | std::cout<<argv[1]<<" took "<<CoinCpuTime()-time1<<" seconds, " |
---|
218 | <<model.getNodeCount()<<" nodes with objective " |
---|
219 | <<model.getObjValue() |
---|
220 | <<(!model.status() ? " Finished" : " Not finished") |
---|
221 | <<std::endl; |
---|
222 | |
---|
223 | // Print solution if finished - we can't get names from Osi! |
---|
224 | |
---|
225 | if (!model.status()&&model.getMinimizationObjValue()<1.0e50) { |
---|
226 | int numberColumns = model.solver()->getNumCols(); |
---|
227 | |
---|
228 | //const double * solution = model.bestSolution(); |
---|
229 | const double * solution = model.solver()->getColSolution(); |
---|
230 | |
---|
231 | int iColumn; |
---|
232 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
233 | double value=solution[iColumn]; |
---|
234 | if (fabs(value)>1.0e-7&&model.solver()->isInteger(iColumn)) |
---|
235 | printf("Column %d has value %g\n",iColumn,value); |
---|
236 | } |
---|
237 | } |
---|
238 | return 0; |
---|
239 | } |
---|