1 | // Copyright (C) 2002, International Business Machines |
---|
2 | // Corporation and others. All Rights Reserved. |
---|
3 | #if defined(_MSC_VER) |
---|
4 | // Turn off compiler warning about long names |
---|
5 | # pragma warning(disable:4786) |
---|
6 | #endif |
---|
7 | #include <string> |
---|
8 | //#define CBC_DEBUG 1 |
---|
9 | //#define CHECK_CUT_COUNTS |
---|
10 | //#define CHECK_NODE_FULL |
---|
11 | //#define NODE_LOG |
---|
12 | #include <cassert> |
---|
13 | #include <cmath> |
---|
14 | #include <cfloat> |
---|
15 | |
---|
16 | #ifdef CBC_USE_CLP |
---|
17 | // include Presolve from Clp |
---|
18 | #include "ClpPresolve.hpp" |
---|
19 | #include "OsiClpSolverInterface.hpp" |
---|
20 | #endif |
---|
21 | |
---|
22 | #include "CbcEventHandler.hpp" |
---|
23 | |
---|
24 | #include "OsiSolverInterface.hpp" |
---|
25 | #include "OsiAuxInfo.hpp" |
---|
26 | #include "OsiSolverBranch.hpp" |
---|
27 | #include "CoinWarmStartBasis.hpp" |
---|
28 | #include "CoinPackedMatrix.hpp" |
---|
29 | #include "CoinHelperFunctions.hpp" |
---|
30 | #include "CbcBranchActual.hpp" |
---|
31 | #include "CbcBranchDynamic.hpp" |
---|
32 | #include "CbcHeuristic.hpp" |
---|
33 | #include "CbcModel.hpp" |
---|
34 | #include "CbcTreeLocal.hpp" |
---|
35 | #include "CbcStatistics.hpp" |
---|
36 | #include "CbcStrategy.hpp" |
---|
37 | #include "CbcMessage.hpp" |
---|
38 | #include "OsiRowCut.hpp" |
---|
39 | #include "OsiColCut.hpp" |
---|
40 | #include "OsiRowCutDebugger.hpp" |
---|
41 | #include "OsiCuts.hpp" |
---|
42 | #include "CbcCountRowCut.hpp" |
---|
43 | #include "CbcCutGenerator.hpp" |
---|
44 | #include "CbcFeasibilityBase.hpp" |
---|
45 | // include Probing |
---|
46 | #include "CglProbing.hpp" |
---|
47 | // include preprocessing |
---|
48 | #include "CglPreProcess.hpp" |
---|
49 | |
---|
50 | #include "CoinTime.hpp" |
---|
51 | |
---|
52 | #include "CbcCompareActual.hpp" |
---|
53 | #include "CbcTree.hpp" |
---|
54 | /* Various functions local to CbcModel.cpp */ |
---|
55 | |
---|
56 | namespace { |
---|
57 | |
---|
58 | //------------------------------------------------------------------- |
---|
59 | // Returns the greatest common denominator of two |
---|
60 | // positive integers, a and b, found using Euclid's algorithm |
---|
61 | //------------------------------------------------------------------- |
---|
62 | static int gcd(int a, int b) |
---|
63 | { |
---|
64 | int remainder = -1; |
---|
65 | // make sure a<=b (will always remain so) |
---|
66 | if(a > b) { |
---|
67 | // Swap a and b |
---|
68 | int temp = a; |
---|
69 | a = b; |
---|
70 | b = temp; |
---|
71 | } |
---|
72 | // if zero then gcd is nonzero (zero may occur in rhs of packed) |
---|
73 | if (!a) { |
---|
74 | if (b) { |
---|
75 | return b; |
---|
76 | } else { |
---|
77 | printf("**** gcd given two zeros!!\n"); |
---|
78 | abort(); |
---|
79 | } |
---|
80 | } |
---|
81 | while (remainder) { |
---|
82 | remainder = b % a; |
---|
83 | b = a; |
---|
84 | a = remainder; |
---|
85 | } |
---|
86 | return b; |
---|
87 | } |
---|
88 | |
---|
89 | |
---|
90 | |
---|
91 | #ifdef CHECK_NODE_FULL |
---|
92 | |
---|
93 | /* |
---|
94 | Routine to verify that tree linkage is correct. The invariant that is tested |
---|
95 | is |
---|
96 | |
---|
97 | reference count = (number of actual references) + (number of branches left) |
---|
98 | |
---|
99 | The routine builds a set of paired arrays, info and count, by traversing the |
---|
100 | tree. Each CbcNodeInfo is recorded in info, and the number of times it is |
---|
101 | referenced (via the parent field) is recorded in count. Then a final check is |
---|
102 | made to see if the numberPointingToThis_ field agrees. |
---|
103 | */ |
---|
104 | |
---|
105 | void verifyTreeNodes (const CbcTree * branchingTree, const CbcModel &model) |
---|
106 | |
---|
107 | { printf("*** CHECKING tree after %d nodes\n",model.getNodeCount()) ; |
---|
108 | |
---|
109 | int j ; |
---|
110 | int nNodes = branchingTree->size() ; |
---|
111 | # define MAXINFO 1000 |
---|
112 | int *count = new int [MAXINFO] ; |
---|
113 | CbcNodeInfo **info = new CbcNodeInfo*[MAXINFO] ; |
---|
114 | int nInfo = 0 ; |
---|
115 | /* |
---|
116 | Collect all CbcNodeInfo objects in info, by starting from each live node and |
---|
117 | traversing back to the root. Nodes in the live set should have unexplored |
---|
118 | branches remaining. |
---|
119 | |
---|
120 | TODO: The `while (nodeInfo)' loop could be made to break on reaching a |
---|
121 | common ancester (nodeInfo is found in info[k]). Alternatively, the |
---|
122 | check could change to signal an error if nodeInfo is not found above a |
---|
123 | common ancestor. |
---|
124 | */ |
---|
125 | for (j = 0 ; j < nNodes ; j++) |
---|
126 | { CbcNode *node = branchingTree->nodePointer(j) ; |
---|
127 | if (!node) |
---|
128 | continue; |
---|
129 | CbcNodeInfo *nodeInfo = node->nodeInfo() ; |
---|
130 | int change = node->nodeInfo()->numberBranchesLeft() ; |
---|
131 | assert(change) ; |
---|
132 | while (nodeInfo) |
---|
133 | { int k ; |
---|
134 | for (k = 0 ; k < nInfo ; k++) |
---|
135 | { if (nodeInfo == info[k]) break ; } |
---|
136 | if (k == nInfo) |
---|
137 | { assert(nInfo < MAXINFO) ; |
---|
138 | nInfo++ ; |
---|
139 | info[k] = nodeInfo ; |
---|
140 | count[k] = 0 ; } |
---|
141 | nodeInfo = nodeInfo->parent() ; } } |
---|
142 | /* |
---|
143 | Walk the info array. For each nodeInfo, look up its parent in info and |
---|
144 | increment the corresponding count. |
---|
145 | */ |
---|
146 | for (j = 0 ; j < nInfo ; j++) |
---|
147 | { CbcNodeInfo *nodeInfo = info[j] ; |
---|
148 | nodeInfo = nodeInfo->parent() ; |
---|
149 | if (nodeInfo) |
---|
150 | { int k ; |
---|
151 | for (k = 0 ; k < nInfo ; k++) |
---|
152 | { if (nodeInfo == info[k]) break ; } |
---|
153 | assert (k < nInfo) ; |
---|
154 | count[k]++ ; } } |
---|
155 | /* |
---|
156 | Walk the info array one more time and check that the invariant holds. The |
---|
157 | number of references (numberPointingToThis()) should equal the sum of the |
---|
158 | number of actual references (held in count[]) plus the number of potential |
---|
159 | references (unexplored branches, numberBranchesLeft()). |
---|
160 | */ |
---|
161 | for (j = 0;j < nInfo;j++) { |
---|
162 | CbcNodeInfo * nodeInfo = info[j] ; |
---|
163 | if (nodeInfo) { |
---|
164 | int k ; |
---|
165 | for (k = 0;k < nInfo;k++) |
---|
166 | if (nodeInfo == info[k]) |
---|
167 | break ; |
---|
168 | printf("Nodeinfo %x - %d left, %d count\n", |
---|
169 | nodeInfo, |
---|
170 | nodeInfo->numberBranchesLeft(), |
---|
171 | nodeInfo->numberPointingToThis()) ; |
---|
172 | assert(nodeInfo->numberPointingToThis() == |
---|
173 | count[k]+nodeInfo->numberBranchesLeft()) ; } } |
---|
174 | |
---|
175 | delete [] count ; |
---|
176 | delete [] info ; |
---|
177 | |
---|
178 | return ; } |
---|
179 | |
---|
180 | #endif /* CHECK_NODE_FULL */ |
---|
181 | |
---|
182 | |
---|
183 | |
---|
184 | #ifdef CHECK_CUT_COUNTS |
---|
185 | |
---|
186 | /* |
---|
187 | Routine to verify that cut reference counts are correct. |
---|
188 | */ |
---|
189 | void verifyCutCounts (const CbcTree * branchingTree, CbcModel &model) |
---|
190 | |
---|
191 | { printf("*** CHECKING cuts after %d nodes\n",model.getNodeCount()) ; |
---|
192 | |
---|
193 | int j ; |
---|
194 | int nNodes = branchingTree->size() ; |
---|
195 | |
---|
196 | /* |
---|
197 | cut.tempNumber_ exists for the purpose of doing this verification. Clear it |
---|
198 | in all cuts. We traverse the tree by starting from each live node and working |
---|
199 | back to the root. At each CbcNodeInfo, check for cuts. |
---|
200 | */ |
---|
201 | for (j = 0 ; j < nNodes ; j++) |
---|
202 | { CbcNode *node = branchingTree->nodePointer(j) ; |
---|
203 | CbcNodeInfo * nodeInfo = node->nodeInfo() ; |
---|
204 | assert (node->nodeInfo()->numberBranchesLeft()) ; |
---|
205 | while (nodeInfo) |
---|
206 | { int k ; |
---|
207 | for (k = 0 ; k < nodeInfo->numberCuts() ; k++) |
---|
208 | { CbcCountRowCut *cut = nodeInfo->cuts()[k] ; |
---|
209 | if (cut) cut->tempNumber_ = 0; } |
---|
210 | nodeInfo = nodeInfo->parent() ; } } |
---|
211 | /* |
---|
212 | Walk the live set again, this time collecting the list of cuts in use at each |
---|
213 | node. addCuts1 will collect the cuts in model.addedCuts_. Take into account |
---|
214 | that when we recreate the basis for a node, we compress out the slack cuts. |
---|
215 | */ |
---|
216 | for (j = 0 ; j < nNodes ; j++) |
---|
217 | { CoinWarmStartBasis *debugws = model.getEmptyBasis() ; |
---|
218 | CbcNode *node = branchingTree->nodePointer(j) ; |
---|
219 | CbcNodeInfo *nodeInfo = node->nodeInfo(); |
---|
220 | int change = node->nodeInfo()->numberBranchesLeft() ; |
---|
221 | printf("Node %d %x (info %x) var %d way %d obj %g",j,node, |
---|
222 | node->nodeInfo(),node->variable(),node->way(), |
---|
223 | node->objectiveValue()) ; |
---|
224 | |
---|
225 | model.addCuts1(node,debugws) ; |
---|
226 | |
---|
227 | int i ; |
---|
228 | int numberRowsAtContinuous = model.numberRowsAtContinuous() ; |
---|
229 | CbcCountRowCut **addedCuts = model.addedCuts() ; |
---|
230 | for (i = 0 ; i < model.currentNumberCuts() ; i++) |
---|
231 | { CoinWarmStartBasis::Status status = |
---|
232 | debugws->getArtifStatus(i+numberRowsAtContinuous) ; |
---|
233 | if (status != CoinWarmStartBasis::basic && addedCuts[i]) |
---|
234 | { addedCuts[i]->tempNumber_ += change ; } } |
---|
235 | |
---|
236 | while (nodeInfo) |
---|
237 | { nodeInfo = nodeInfo->parent() ; |
---|
238 | if (nodeInfo) printf(" -> %x",nodeInfo); } |
---|
239 | printf("\n") ; |
---|
240 | delete debugws ; } |
---|
241 | /* |
---|
242 | The moment of truth: We've tallied up the references by direct scan of the |
---|
243 | search tree. Check for agreement with the count in the cut. |
---|
244 | |
---|
245 | TODO: Rewrite to check and print mismatch only when tempNumber_ == 0? |
---|
246 | */ |
---|
247 | for (j = 0 ; j < nNodes ; j++) |
---|
248 | { CbcNode *node = branchingTree->nodePointer(j) ; |
---|
249 | CbcNodeInfo *nodeInfo = node->nodeInfo(); |
---|
250 | while (nodeInfo) |
---|
251 | { int k ; |
---|
252 | for (k = 0 ; k < nodeInfo->numberCuts() ; k++) |
---|
253 | { CbcCountRowCut *cut = nodeInfo->cuts()[k] ; |
---|
254 | if (cut && cut->tempNumber_ >= 0) |
---|
255 | { if (cut->tempNumber_ != cut->numberPointingToThis()) |
---|
256 | printf("mismatch %x %d %x %d %d\n",nodeInfo,k, |
---|
257 | cut,cut->tempNumber_,cut->numberPointingToThis()) ; |
---|
258 | else |
---|
259 | printf(" match %x %d %x %d %d\n", nodeInfo,k, |
---|
260 | cut,cut->tempNumber_,cut->numberPointingToThis()) ; |
---|
261 | cut->tempNumber_ = -1 ; } } |
---|
262 | nodeInfo = nodeInfo->parent() ; } } |
---|
263 | |
---|
264 | return ; } |
---|
265 | |
---|
266 | #endif /* CHECK_CUT_COUNTS */ |
---|
267 | |
---|
268 | } |
---|
269 | |
---|
270 | /* End unnamed namespace for CbcModel.cpp */ |
---|
271 | |
---|
272 | |
---|
273 | |
---|
274 | void |
---|
275 | CbcModel::analyzeObjective () |
---|
276 | /* |
---|
277 | Try to find a minimum change in the objective function. The first scan |
---|
278 | checks that there are no continuous variables with non-zero coefficients, |
---|
279 | and grabs the largest objective coefficient associated with an unfixed |
---|
280 | integer variable. The second scan attempts to scale up the objective |
---|
281 | coefficients to a point where they are sufficiently close to integer that |
---|
282 | we can pretend they are integer, and calculate a gcd over the coefficients |
---|
283 | of interest. This will be the minimum increment for the scaled coefficients. |
---|
284 | The final action is to scale the increment back for the original coefficients |
---|
285 | and install it, if it's better than the existing value. |
---|
286 | |
---|
287 | John's note: We could do better than this. |
---|
288 | |
---|
289 | John's second note - apologies for changing s to z |
---|
290 | */ |
---|
291 | { const double *objective = getObjCoefficients() ; |
---|
292 | const double *lower = getColLower() ; |
---|
293 | const double *upper = getColUpper() ; |
---|
294 | /* |
---|
295 | Take a first scan to see if there are unfixed continuous variables in the |
---|
296 | objective. If so, the minimum objective change could be arbitrarily small. |
---|
297 | Also pick off the maximum coefficient of an unfixed integer variable. |
---|
298 | |
---|
299 | If the objective is found to contain only integer variables, set the |
---|
300 | fathoming discipline to strict. |
---|
301 | */ |
---|
302 | double maximumCost = 0.0 ; |
---|
303 | bool possibleMultiple = true ; |
---|
304 | int iColumn ; |
---|
305 | int numberColumns = getNumCols() ; |
---|
306 | for (iColumn = 0 ; iColumn < numberColumns ; iColumn++) |
---|
307 | { if (upper[iColumn] > lower[iColumn]+1.0e-8) |
---|
308 | { if (isInteger(iColumn)) |
---|
309 | maximumCost = CoinMax(maximumCost,fabs(objective[iColumn])) ; |
---|
310 | else if (objective[iColumn]) |
---|
311 | possibleMultiple = false ; } } |
---|
312 | setIntParam(CbcModel::CbcFathomDiscipline,possibleMultiple) ; |
---|
313 | /* |
---|
314 | If a nontrivial increment is possible, try and figure it out. We're looking |
---|
315 | for gcd(c<j>) for all c<j> that are coefficients of unfixed integer |
---|
316 | variables. Since the c<j> might not be integers, try and inflate them |
---|
317 | sufficiently that they look like integers (and we'll deflate the gcd |
---|
318 | later). |
---|
319 | |
---|
320 | 2520.0 is used as it is a nice multiple of 2,3,5,7 |
---|
321 | */ |
---|
322 | if (possibleMultiple&&maximumCost) |
---|
323 | { int increment = 0 ; |
---|
324 | double multiplier = 2520.0 ; |
---|
325 | while (10.0*multiplier*maximumCost < 1.0e8) |
---|
326 | multiplier *= 10.0 ; |
---|
327 | |
---|
328 | for (iColumn = 0 ; iColumn < numberColumns ; iColumn++) |
---|
329 | { if (upper[iColumn] > lower[iColumn]+1.0e-8) |
---|
330 | { if (isInteger(iColumn)&&objective[iColumn]) |
---|
331 | { double value = fabs(objective[iColumn])*multiplier ; |
---|
332 | int nearest = (int) floor(value+0.5) ; |
---|
333 | if (fabs(value-floor(value+0.5)) > 1.0e-8) |
---|
334 | { increment = 0 ; |
---|
335 | break ; } |
---|
336 | else if (!increment) |
---|
337 | { increment = nearest ; } |
---|
338 | else |
---|
339 | { increment = gcd(increment,nearest) ; } } } } |
---|
340 | /* |
---|
341 | If the increment beats the current value for objective change, install it. |
---|
342 | */ |
---|
343 | if (increment) |
---|
344 | { double value = increment ; |
---|
345 | double cutoff = getDblParam(CbcModel::CbcCutoffIncrement) ; |
---|
346 | value /= multiplier ; |
---|
347 | if (value*0.999 > cutoff) |
---|
348 | { messageHandler()->message(CBC_INTEGERINCREMENT, |
---|
349 | messages()) |
---|
350 | << value << CoinMessageEol ; |
---|
351 | setDblParam(CbcModel::CbcCutoffIncrement,value*0.999) ; } } } |
---|
352 | |
---|
353 | return ; |
---|
354 | } |
---|
355 | |
---|
356 | |
---|
357 | /** |
---|
358 | \todo |
---|
359 | Normally, it looks like we enter here from command dispatch in the main |
---|
360 | routine, after calling the solver for an initial solution |
---|
361 | (CbcModel::initialSolve, which simply calls the solver's initialSolve |
---|
362 | routine.) The first thing we do is call resolve. Presumably there are |
---|
363 | circumstances where this is nontrivial? There's also a call from |
---|
364 | CbcModel::originalModel (tied up with integer presolve), which should be |
---|
365 | checked. |
---|
366 | |
---|
367 | */ |
---|
368 | |
---|
369 | /* |
---|
370 | The overall flow can be divided into three stages: |
---|
371 | * Prep: Check that the lp relaxation remains feasible at the root. If so, |
---|
372 | do all the setup for B&C. |
---|
373 | * Process the root node: Generate cuts, apply heuristics, and in general do |
---|
374 | the best we can to resolve the problem without B&C. |
---|
375 | * Do B&C search until we hit a limit or exhaust the search tree. |
---|
376 | |
---|
377 | Keep in mind that in general there is no node in the search tree that |
---|
378 | corresponds to the active subproblem. The active subproblem is represented |
---|
379 | by the current state of the model, of the solver, and of the constraint |
---|
380 | system held by the solver. |
---|
381 | */ |
---|
382 | |
---|
383 | void CbcModel::branchAndBound(int doStatistics) |
---|
384 | |
---|
385 | { |
---|
386 | /* |
---|
387 | Capture a time stamp before we start. |
---|
388 | */ |
---|
389 | dblParam_[CbcStartSeconds] = CoinCpuTime(); |
---|
390 | strongInfo_[0]=0; |
---|
391 | strongInfo_[1]=0; |
---|
392 | strongInfo_[2]=0; |
---|
393 | numberStrongIterations_ = 0; |
---|
394 | // original solver (only set if pre-processing) |
---|
395 | OsiSolverInterface * originalSolver=NULL; |
---|
396 | int numberOriginalObjects=numberObjects_; |
---|
397 | CbcObject ** originalObject = NULL; |
---|
398 | // Set up strategies |
---|
399 | if (strategy_) { |
---|
400 | // May do preprocessing |
---|
401 | originalSolver = solver_; |
---|
402 | strategy_->setupOther(*this); |
---|
403 | if (strategy_->preProcessState()) { |
---|
404 | // pre-processing done |
---|
405 | if (strategy_->preProcessState()<0) { |
---|
406 | // infeasible |
---|
407 | handler_->message(CBC_INFEAS,messages_)<< CoinMessageEol ; |
---|
408 | status_ = 0 ; |
---|
409 | secondaryStatus_ = 1; |
---|
410 | originalContinuousObjective_ = COIN_DBL_MAX; |
---|
411 | return ; |
---|
412 | } else if (numberObjects_&&object_) { |
---|
413 | numberOriginalObjects=numberObjects_; |
---|
414 | // redo sequence |
---|
415 | numberIntegers_=0; |
---|
416 | int numberColumns = getNumCols(); |
---|
417 | int nOrig = originalSolver->getNumCols(); |
---|
418 | CglPreProcess * process = strategy_->process(); |
---|
419 | assert (process); |
---|
420 | const int * originalColumns = process->originalColumns(); |
---|
421 | // allow for cliques etc |
---|
422 | nOrig = CoinMax(nOrig,originalColumns[numberColumns-1]+1); |
---|
423 | originalObject = object_; |
---|
424 | // object number or -1 |
---|
425 | int * temp = new int[nOrig]; |
---|
426 | int iColumn; |
---|
427 | for (iColumn=0;iColumn<nOrig;iColumn++) |
---|
428 | temp[iColumn]=-1; |
---|
429 | int iObject; |
---|
430 | int nNonInt=0; |
---|
431 | for (iObject=0;iObject<numberOriginalObjects;iObject++) { |
---|
432 | iColumn = originalObject[iObject]->columnNumber(); |
---|
433 | if (iColumn<0) { |
---|
434 | nNonInt++; |
---|
435 | } else { |
---|
436 | temp[iColumn]=iObject; |
---|
437 | } |
---|
438 | } |
---|
439 | int numberNewIntegers=0; |
---|
440 | int numberOldIntegers=0; |
---|
441 | int numberOldOther=0; |
---|
442 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
443 | int jColumn = originalColumns[iColumn]; |
---|
444 | if (temp[jColumn]>=0) { |
---|
445 | int iObject= temp[jColumn]; |
---|
446 | CbcSimpleInteger * obj = |
---|
447 | dynamic_cast <CbcSimpleInteger *>(originalObject[iObject]) ; |
---|
448 | if (obj) |
---|
449 | numberOldIntegers++; |
---|
450 | else |
---|
451 | numberOldOther++; |
---|
452 | } else if (isInteger(iColumn)) { |
---|
453 | numberNewIntegers++; |
---|
454 | } |
---|
455 | } |
---|
456 | /* |
---|
457 | Allocate an array to hold the indices of the integer variables. |
---|
458 | Make a large enough array for all objects |
---|
459 | */ |
---|
460 | numberObjects_= numberNewIntegers+numberOldIntegers+numberOldOther+nNonInt; |
---|
461 | object_ = new CbcObject * [numberObjects_]; |
---|
462 | integerVariable_ = new int [numberNewIntegers+numberOldIntegers]; |
---|
463 | /* |
---|
464 | Walk the variables again, filling in the indices and creating objects for |
---|
465 | the integer variables. Initially, the objects hold the index and upper & |
---|
466 | lower bounds. |
---|
467 | */ |
---|
468 | numberIntegers_=0; |
---|
469 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
470 | int jColumn = originalColumns[iColumn]; |
---|
471 | if (temp[jColumn]>=0) { |
---|
472 | int iObject= temp[jColumn]; |
---|
473 | CbcSimpleInteger * obj = |
---|
474 | dynamic_cast <CbcSimpleInteger *>(originalObject[iObject]) ; |
---|
475 | if (obj) { |
---|
476 | object_[numberIntegers_] = originalObject[iObject]->clone(); |
---|
477 | // redo ids etc |
---|
478 | object_[numberIntegers_]->redoSequenceEtc(this,numberColumns,originalColumns); |
---|
479 | integerVariable_[numberIntegers_++]=iColumn; |
---|
480 | } |
---|
481 | } else if (isInteger(iColumn)) { |
---|
482 | object_[numberIntegers_] = |
---|
483 | new CbcSimpleInteger(this,numberIntegers_,iColumn); |
---|
484 | integerVariable_[numberIntegers_++]=iColumn; |
---|
485 | } |
---|
486 | } |
---|
487 | numberObjects_=numberIntegers_; |
---|
488 | // Now append other column stuff |
---|
489 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
490 | int jColumn = originalColumns[iColumn]; |
---|
491 | if (temp[jColumn]>=0) { |
---|
492 | int iObject= temp[jColumn]; |
---|
493 | CbcSimpleInteger * obj = |
---|
494 | dynamic_cast <CbcSimpleInteger *>(originalObject[iObject]) ; |
---|
495 | if (!obj) { |
---|
496 | object_[numberObjects_] = originalObject[iObject]->clone(); |
---|
497 | // redo ids etc |
---|
498 | object_[numberObjects_]->redoSequenceEtc(this,numberColumns,originalColumns); |
---|
499 | numberObjects_++; |
---|
500 | } |
---|
501 | } |
---|
502 | } |
---|
503 | // now append non column stuff |
---|
504 | for (iObject=0;iObject<numberOriginalObjects;iObject++) { |
---|
505 | iColumn = originalObject[iObject]->columnNumber(); |
---|
506 | if (iColumn<0) { |
---|
507 | object_[numberObjects_] = originalObject[iObject]->clone(); |
---|
508 | // redo ids etc |
---|
509 | object_[numberObjects_]->redoSequenceEtc(this,numberColumns,originalColumns); |
---|
510 | numberObjects_++; |
---|
511 | } |
---|
512 | } |
---|
513 | delete [] temp; |
---|
514 | if (!numberObjects_) |
---|
515 | handler_->message(CBC_NOINT,messages_) << CoinMessageEol ; |
---|
516 | } |
---|
517 | } else { |
---|
518 | //no preprocessing |
---|
519 | originalSolver=NULL; |
---|
520 | } |
---|
521 | strategy_->setupCutGenerators(*this); |
---|
522 | strategy_->setupHeuristics(*this); |
---|
523 | // Set strategy print level to models |
---|
524 | strategy_->setupPrinting(*this,handler_->logLevel()); |
---|
525 | } |
---|
526 | eventHappened_=false; |
---|
527 | CbcEventHandler *eventHandler = getEventHandler() ; |
---|
528 | |
---|
529 | if (!nodeCompare_) |
---|
530 | nodeCompare_=new CbcCompareDefault();; |
---|
531 | // See if hot start wanted |
---|
532 | CbcCompareBase * saveCompare = NULL; |
---|
533 | if (hotstartSolution_) { |
---|
534 | if (strategy_&&strategy_->preProcessState()>0) { |
---|
535 | CglPreProcess * process = strategy_->process(); |
---|
536 | assert (process); |
---|
537 | int n = solver_->getNumCols(); |
---|
538 | const int * originalColumns = process->originalColumns(); |
---|
539 | // columns should be in order ... but |
---|
540 | double * tempS = new double[n]; |
---|
541 | for (int i=0;i<n;i++) { |
---|
542 | int iColumn = originalColumns[i]; |
---|
543 | tempS[i]=hotstartSolution_[iColumn]; |
---|
544 | } |
---|
545 | delete [] hotstartSolution_; |
---|
546 | hotstartSolution_=tempS; |
---|
547 | if (hotstartPriorities_) { |
---|
548 | int * tempP = new int [n]; |
---|
549 | for (int i=0;i<n;i++) { |
---|
550 | int iColumn = originalColumns[i]; |
---|
551 | tempP[i]=hotstartPriorities_[iColumn]; |
---|
552 | } |
---|
553 | delete [] hotstartPriorities_; |
---|
554 | hotstartPriorities_=tempP; |
---|
555 | } |
---|
556 | } |
---|
557 | saveCompare = nodeCompare_; |
---|
558 | // depth first |
---|
559 | nodeCompare_ = new CbcCompareDepth(); |
---|
560 | } |
---|
561 | if (!problemFeasibility_) |
---|
562 | problemFeasibility_=new CbcFeasibilityBase(); |
---|
563 | # ifdef CBC_DEBUG |
---|
564 | std::string problemName ; |
---|
565 | solver_->getStrParam(OsiProbName,problemName) ; |
---|
566 | printf("Problem name - %s\n",problemName.c_str()) ; |
---|
567 | solver_->setHintParam(OsiDoReducePrint,false,OsiHintDo,0) ; |
---|
568 | # endif |
---|
569 | /* |
---|
570 | Assume we're done, and see if we're proven wrong. |
---|
571 | */ |
---|
572 | status_ = 0 ; |
---|
573 | secondaryStatus_ = 0; |
---|
574 | phase_=0; |
---|
575 | /* |
---|
576 | Scan the variables, noting the integer variables. Create an |
---|
577 | CbcSimpleInteger object for each integer variable. |
---|
578 | */ |
---|
579 | findIntegers(false) ; |
---|
580 | // If dynamic pseudo costs then do |
---|
581 | if (numberBeforeTrust_) |
---|
582 | convertToDynamic(); |
---|
583 | // Set up char array to say if integer |
---|
584 | delete [] integerInfo_; |
---|
585 | { |
---|
586 | int n = solver_->getNumCols(); |
---|
587 | integerInfo_ = new char [n]; |
---|
588 | for (int i=0;i<n;i++) { |
---|
589 | if (solver_->isInteger(i)) |
---|
590 | integerInfo_[i]=1; |
---|
591 | else |
---|
592 | integerInfo_[i]=0; |
---|
593 | } |
---|
594 | } |
---|
595 | |
---|
596 | /* |
---|
597 | Ensure that objects on the lists of CbcObjects, heuristics, and cut |
---|
598 | generators attached to this model all refer to this model. |
---|
599 | */ |
---|
600 | synchronizeModel() ; |
---|
601 | assert (!solverCharacteristics_); |
---|
602 | OsiBabSolver * solverCharacteristics = dynamic_cast<OsiBabSolver *> (solver_->getAuxiliaryInfo()); |
---|
603 | if (solverCharacteristics) { |
---|
604 | solverCharacteristics_ = solverCharacteristics; |
---|
605 | } else { |
---|
606 | // replace in solver |
---|
607 | OsiBabSolver defaultC; |
---|
608 | solver_->setAuxiliaryInfo(&defaultC); |
---|
609 | solverCharacteristics_ = dynamic_cast<OsiBabSolver *> (solver_->getAuxiliaryInfo()); |
---|
610 | } |
---|
611 | solverCharacteristics_->setSolver(solver_); |
---|
612 | // Set so we can tell we are in initial phase in resolve |
---|
613 | continuousObjective_ = -COIN_DBL_MAX ; |
---|
614 | /* |
---|
615 | Solve the relaxation. |
---|
616 | |
---|
617 | Apparently there are circumstances where this will be non-trivial --- i.e., |
---|
618 | we've done something since initialSolve that's trashed the solution to the |
---|
619 | continuous relaxation. |
---|
620 | */ |
---|
621 | bool feasible; |
---|
622 | // If NLP then we assume already solved outside branchAndbound |
---|
623 | if (!solverCharacteristics_->solverType()) { |
---|
624 | feasible=resolve(NULL,0) != 0 ; |
---|
625 | } else { |
---|
626 | // pick up given status |
---|
627 | feasible = (solver_->isProvenOptimal() && |
---|
628 | !solver_->isDualObjectiveLimitReached()) ; |
---|
629 | } |
---|
630 | if (problemFeasibility_->feasible(this,0)<0) { |
---|
631 | feasible=false; // pretend infeasible |
---|
632 | } |
---|
633 | /* |
---|
634 | If the linear relaxation of the root is infeasible, bail out now. Otherwise, |
---|
635 | continue with processing the root node. |
---|
636 | */ |
---|
637 | if (!feasible) |
---|
638 | { handler_->message(CBC_INFEAS,messages_)<< CoinMessageEol ; |
---|
639 | status_ = 0 ; |
---|
640 | secondaryStatus_ = 1; |
---|
641 | originalContinuousObjective_ = COIN_DBL_MAX; |
---|
642 | solverCharacteristics_ = NULL; |
---|
643 | return ; } |
---|
644 | // Save objective (just so user can access it) |
---|
645 | originalContinuousObjective_ = solver_->getObjValue(); |
---|
646 | bestPossibleObjective_=originalContinuousObjective_; |
---|
647 | sumChangeObjective1_=0.0; |
---|
648 | sumChangeObjective2_=0.0; |
---|
649 | /* |
---|
650 | OsiRowCutDebugger knows an optimal answer for a subset of MIP problems. |
---|
651 | Assuming it recognises the problem, when called upon it will check a cut to |
---|
652 | see if it cuts off the optimal answer. |
---|
653 | */ |
---|
654 | // If debugger exists set specialOptions_ bit |
---|
655 | if (solver_->getRowCutDebuggerAlways()) |
---|
656 | specialOptions_ |= 1; |
---|
657 | |
---|
658 | # ifdef CBC_DEBUG |
---|
659 | if ((specialOptions_&1)==0) |
---|
660 | solver_->activateRowCutDebugger(problemName.c_str()) ; |
---|
661 | if (solver_->getRowCutDebuggerAlways()) |
---|
662 | specialOptions_ |= 1; |
---|
663 | # endif |
---|
664 | |
---|
665 | /* |
---|
666 | Begin setup to process a feasible root node. |
---|
667 | */ |
---|
668 | bestObjective_ = CoinMin(bestObjective_,1.0e50) ; |
---|
669 | numberSolutions_ = 0 ; |
---|
670 | stateOfSearch_=0; |
---|
671 | numberHeuristicSolutions_ = 0 ; |
---|
672 | // Everything is minimization |
---|
673 | { |
---|
674 | // needed to sync cutoffs |
---|
675 | double value ; |
---|
676 | solver_->getDblParam(OsiDualObjectiveLimit,value) ; |
---|
677 | dblParam_[CbcCurrentCutoff]= value * solver_->getObjSense(); |
---|
678 | } |
---|
679 | double cutoff=getCutoff() ; |
---|
680 | double direction = solver_->getObjSense() ; |
---|
681 | dblParam_[CbcOptimizationDirection]=direction; |
---|
682 | if (cutoff < 1.0e20&&direction<0.0) |
---|
683 | messageHandler()->message(CBC_CUTOFF_WARNING1, |
---|
684 | messages()) |
---|
685 | << cutoff << -cutoff << CoinMessageEol ; |
---|
686 | if (cutoff > bestObjective_) |
---|
687 | cutoff = bestObjective_ ; |
---|
688 | setCutoff(cutoff) ; |
---|
689 | /* |
---|
690 | We probably already have a current solution, but just in case ... |
---|
691 | */ |
---|
692 | int numberColumns = getNumCols() ; |
---|
693 | if (!currentSolution_) |
---|
694 | currentSolution_ = new double[numberColumns] ; |
---|
695 | testSolution_ = currentSolution_; |
---|
696 | /* Tell solver we are in Branch and Cut |
---|
697 | Could use last parameter for subtle differences */ |
---|
698 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,NULL) ; |
---|
699 | /* |
---|
700 | Create a copy of the solver, thus capturing the original (root node) |
---|
701 | constraint system (aka the continuous system). |
---|
702 | */ |
---|
703 | continuousSolver_ = solver_->clone() ; |
---|
704 | #ifdef CBC_USE_CLP |
---|
705 | OsiClpSolverInterface * clpSolver |
---|
706 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
707 | if (clpSolver) { |
---|
708 | ClpSimplex * clpSimplex = clpSolver->getModelPtr(); |
---|
709 | // take off names |
---|
710 | clpSimplex->dropNames(); |
---|
711 | } |
---|
712 | #endif |
---|
713 | |
---|
714 | numberRowsAtContinuous_ = getNumRows() ; |
---|
715 | /* |
---|
716 | Check the objective to see if we can deduce a nontrivial increment. If |
---|
717 | it's better than the current value for CbcCutoffIncrement, it'll be |
---|
718 | installed. |
---|
719 | */ |
---|
720 | if(solverCharacteristics_->reducedCostsAccurate()) |
---|
721 | analyzeObjective() ; |
---|
722 | /* |
---|
723 | Set up for cut generation. addedCuts_ holds the cuts which are relevant for |
---|
724 | the active subproblem. whichGenerator will be used to record the generator |
---|
725 | that produced a given cut. |
---|
726 | */ |
---|
727 | maximumWhich_ = 1000 ; |
---|
728 | delete [] whichGenerator_; |
---|
729 | whichGenerator_ = new int[maximumWhich_] ; |
---|
730 | memset(whichGenerator_,0,maximumWhich_*sizeof(int)); |
---|
731 | int currentNumberCuts = 0 ; |
---|
732 | maximumNumberCuts_ = 0 ; |
---|
733 | currentNumberCuts_ = 0 ; |
---|
734 | delete [] addedCuts_ ; |
---|
735 | addedCuts_ = NULL ; |
---|
736 | /* |
---|
737 | Set up an empty heap and associated data structures to hold the live set |
---|
738 | (problems which require further exploration). |
---|
739 | */ |
---|
740 | tree_->setComparison(*nodeCompare_) ; |
---|
741 | /* |
---|
742 | Used to record the path from a node to the root of the search tree, so that |
---|
743 | we can then traverse from the root to the node when restoring a subproblem. |
---|
744 | */ |
---|
745 | maximumDepth_ = 10 ; |
---|
746 | delete [] walkback_ ; |
---|
747 | walkback_ = new CbcNodeInfo * [maximumDepth_] ; |
---|
748 | /* |
---|
749 | Used to generate bound edits for CbcPartialNodeInfo. |
---|
750 | */ |
---|
751 | double * lowerBefore = new double [numberColumns] ; |
---|
752 | double * upperBefore = new double [numberColumns] ; |
---|
753 | /* |
---|
754 | |
---|
755 | Generate cuts at the root node and reoptimise. solveWithCuts does the heavy |
---|
756 | lifting. It will iterate a generate/reoptimise loop (including reduced cost |
---|
757 | fixing) until no cuts are generated, the change in objective falls off, or |
---|
758 | the limit on the number of rounds of cut generation is exceeded. |
---|
759 | |
---|
760 | At the end of all this, any cuts will be recorded in cuts and also |
---|
761 | installed in the solver's constraint system. We'll have reoptimised, and |
---|
762 | removed any slack cuts (numberOldActiveCuts_ and numberNewCuts_ have been |
---|
763 | adjusted accordingly). |
---|
764 | |
---|
765 | Tell cut generators they can be a bit more aggressive at root node |
---|
766 | |
---|
767 | TODO: Why don't we make a copy of the solution after solveWithCuts? |
---|
768 | TODO: If numberUnsatisfied == 0, don't we have a solution? |
---|
769 | */ |
---|
770 | phase_=1; |
---|
771 | int iCutGenerator; |
---|
772 | for (iCutGenerator = 0;iCutGenerator<numberCutGenerators_;iCutGenerator++) { |
---|
773 | CglCutGenerator * generator = generator_[iCutGenerator]->generator(); |
---|
774 | generator->setAggressiveness(generator->getAggressiveness()+100); |
---|
775 | } |
---|
776 | OsiCuts cuts ; |
---|
777 | int anyAction = -1 ; |
---|
778 | numberOldActiveCuts_ = 0 ; |
---|
779 | numberNewCuts_ = 0 ; |
---|
780 | // Array to mark solution |
---|
781 | delete [] usedInSolution_; |
---|
782 | usedInSolution_ = new int[numberColumns]; |
---|
783 | CoinZeroN(usedInSolution_,numberColumns); |
---|
784 | /* |
---|
785 | For printing totals and for CbcNode (numberNodes_) |
---|
786 | */ |
---|
787 | numberIterations_ = 0 ; |
---|
788 | numberNodes_ = 0 ; |
---|
789 | numberNodes2_ = 0 ; |
---|
790 | maximumStatistics_=0; |
---|
791 | statistics_ = NULL; |
---|
792 | // Do on switch |
---|
793 | if (doStatistics) { |
---|
794 | maximumStatistics_=10000; |
---|
795 | statistics_ = new CbcStatistics * [maximumStatistics_]; |
---|
796 | memset(statistics_,0,maximumStatistics_*sizeof(CbcStatistics *)); |
---|
797 | } |
---|
798 | |
---|
799 | { int iObject ; |
---|
800 | int preferredWay ; |
---|
801 | int numberUnsatisfied = 0 ; |
---|
802 | memcpy(currentSolution_,solver_->getColSolution(), |
---|
803 | numberColumns*sizeof(double)) ; |
---|
804 | |
---|
805 | for (iObject = 0 ; iObject < numberObjects_ ; iObject++) |
---|
806 | { double infeasibility = |
---|
807 | object_[iObject]->infeasibility(preferredWay) ; |
---|
808 | if (infeasibility ) numberUnsatisfied++ ; } |
---|
809 | // replace solverType |
---|
810 | if(solverCharacteristics_->tryCuts()) { |
---|
811 | if (numberUnsatisfied) { |
---|
812 | feasible = solveWithCuts(cuts,maximumCutPassesAtRoot_, |
---|
813 | NULL); |
---|
814 | } else if (solverCharacteristics_->solutionAddsCuts()) { |
---|
815 | // may generate cuts and turn the solution |
---|
816 | //to an infeasible one |
---|
817 | feasible = solveWithCuts(cuts, 1, |
---|
818 | NULL); |
---|
819 | #if 0 |
---|
820 | currentNumberCuts_ = cuts.sizeRowCuts(); |
---|
821 | if (currentNumberCuts_ >= maximumNumberCuts_) { |
---|
822 | maximumNumberCuts_ = currentNumberCuts; |
---|
823 | delete [] addedCuts_; |
---|
824 | addedCuts_ = new CbcCountRowCut * [maximumNumberCuts_]; |
---|
825 | } |
---|
826 | #endif |
---|
827 | } |
---|
828 | } |
---|
829 | // check extra info on feasibility |
---|
830 | if (!solverCharacteristics_->mipFeasible()) |
---|
831 | feasible = false; |
---|
832 | } |
---|
833 | // make cut generators less aggressive |
---|
834 | for (iCutGenerator = 0;iCutGenerator<numberCutGenerators_;iCutGenerator++) { |
---|
835 | CglCutGenerator * generator = generator_[iCutGenerator]->generator(); |
---|
836 | generator->setAggressiveness(generator->getAggressiveness()-100); |
---|
837 | } |
---|
838 | currentNumberCuts_ = numberNewCuts_ ; |
---|
839 | /* |
---|
840 | We've taken the continuous relaxation as far as we can. Time to branch. |
---|
841 | The first order of business is to actually create a node. chooseBranch |
---|
842 | currently uses strong branching to evaluate branch object candidates, |
---|
843 | unless forced back to simple branching. If chooseBranch concludes that a |
---|
844 | branching candidate is monotone (anyAction == -1) or infeasible (anyAction |
---|
845 | == -2) when forced to integer values, it returns here immediately. |
---|
846 | |
---|
847 | Monotone variables trigger a call to resolve(). If the problem remains |
---|
848 | feasible, try again to choose a branching variable. At the end of the loop, |
---|
849 | resolved == true indicates that some variables were fixed. |
---|
850 | |
---|
851 | Loss of feasibility will result in the deletion of newNode. |
---|
852 | */ |
---|
853 | |
---|
854 | bool resolved = false ; |
---|
855 | CbcNode *newNode = NULL ; |
---|
856 | numberFixedAtRoot_=0; |
---|
857 | numberFixedNow_=0; |
---|
858 | int numberIterationsAtContinuous = numberIterations_; |
---|
859 | //solverCharacteristics_->setSolver(solver_); |
---|
860 | if (feasible) { |
---|
861 | newNode = new CbcNode ; |
---|
862 | // Set objective value (not so obvious if NLP etc) |
---|
863 | setObjectiveValue(newNode,NULL); |
---|
864 | anyAction = -1 ; |
---|
865 | // To make depth available we may need a fake node |
---|
866 | CbcNode fakeNode; |
---|
867 | if (!currentNode_) { |
---|
868 | // Not true if sub trees assert (!numberNodes_); |
---|
869 | currentNode_=&fakeNode; |
---|
870 | } |
---|
871 | phase_=3; |
---|
872 | // only allow 1000 passes |
---|
873 | int numberPassesLeft=1000; |
---|
874 | // This is first crude step |
---|
875 | if (numberAnalyzeIterations_) { |
---|
876 | delete [] analyzeResults_; |
---|
877 | analyzeResults_ = new double [4*numberIntegers_]; |
---|
878 | numberFixedAtRoot_=newNode->analyze(this,analyzeResults_); |
---|
879 | if (numberFixedAtRoot_>0) { |
---|
880 | printf("%d fixed by analysis\n",numberFixedAtRoot_); |
---|
881 | setPointers(solver_); |
---|
882 | numberFixedNow_ = numberFixedAtRoot_; |
---|
883 | } else if (numberFixedAtRoot_<0) { |
---|
884 | printf("analysis found to be infeasible\n"); |
---|
885 | anyAction=-2; |
---|
886 | delete newNode ; |
---|
887 | newNode = NULL ; |
---|
888 | feasible = false ; |
---|
889 | } |
---|
890 | } |
---|
891 | while (anyAction == -1) |
---|
892 | { |
---|
893 | // Set objective value (not so obvious if NLP etc) |
---|
894 | setObjectiveValue(newNode,NULL); |
---|
895 | if (numberBeforeTrust_==0 ) { |
---|
896 | anyAction = newNode->chooseBranch(this,NULL,numberPassesLeft) ; |
---|
897 | } else { |
---|
898 | OsiSolverBranch * branches=NULL; |
---|
899 | anyAction = newNode->chooseDynamicBranch(this,NULL,branches,numberPassesLeft) ; |
---|
900 | if (anyAction==-3) |
---|
901 | anyAction = newNode->chooseBranch(this,NULL,numberPassesLeft) ; // dynamic did nothing |
---|
902 | } |
---|
903 | numberPassesLeft--; |
---|
904 | if (anyAction == -1) |
---|
905 | { feasible = resolve(NULL,10) != 0 ; |
---|
906 | if (problemFeasibility_->feasible(this,0)<0) { |
---|
907 | feasible=false; // pretend infeasible |
---|
908 | } |
---|
909 | reducedCostFix() ; |
---|
910 | resolved = true ; |
---|
911 | # ifdef CBC_DEBUG |
---|
912 | printf("Resolve (root) as something fixed, Obj value %g %d rows\n", |
---|
913 | solver_->getObjValue(), |
---|
914 | solver_->getNumRows()) ; |
---|
915 | # endif |
---|
916 | if (!feasible) anyAction = -2 ; } |
---|
917 | if (anyAction == -2||newNode->objectiveValue() >= cutoff) |
---|
918 | { delete newNode ; |
---|
919 | newNode = NULL ; |
---|
920 | feasible = false ; } } } |
---|
921 | /* |
---|
922 | At this point, the root subproblem is infeasible or fathomed by bound |
---|
923 | (newNode == NULL), or we're live with an objective value that satisfies the |
---|
924 | current objective cutoff. |
---|
925 | */ |
---|
926 | assert (!newNode || newNode->objectiveValue() <= cutoff) ; |
---|
927 | // Save address of root node as we don't want to delete it |
---|
928 | CbcNode * rootNode = newNode; |
---|
929 | /* |
---|
930 | The common case is that the lp relaxation is feasible but doesn't satisfy |
---|
931 | integrality (i.e., newNode->variable() >= 0, indicating we've been able to |
---|
932 | select a branching variable). Remove any cuts that have gone slack due to |
---|
933 | forcing monotone variables. Then tack on an CbcFullNodeInfo object and full |
---|
934 | basis (via createInfo()) and stash the new cuts in the nodeInfo (via |
---|
935 | addCuts()). If, by some miracle, we have an integral solution at the root |
---|
936 | (newNode->variable() < 0), takeOffCuts() will ensure that the solver holds |
---|
937 | a valid solution for use by setBestSolution(). |
---|
938 | */ |
---|
939 | CoinWarmStartBasis *lastws = 0 ; |
---|
940 | if (feasible && newNode->variable() >= 0) |
---|
941 | { if (resolved) |
---|
942 | { bool needValidSolution = (newNode->variable() < 0) ; |
---|
943 | takeOffCuts(cuts,needValidSolution,NULL) ; |
---|
944 | # ifdef CHECK_CUT_COUNTS |
---|
945 | { printf("Number of rows after chooseBranch fix (root)" |
---|
946 | "(active only) %d\n", |
---|
947 | numberRowsAtContinuous_+numberNewCuts_+numberOldActiveCuts_) ; |
---|
948 | const CoinWarmStartBasis* debugws = |
---|
949 | dynamic_cast <const CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
950 | debugws->print() ; |
---|
951 | delete debugws ; } |
---|
952 | # endif |
---|
953 | } |
---|
954 | newNode->createInfo(this,NULL,NULL,NULL,NULL,0,0) ; |
---|
955 | newNode->nodeInfo()->addCuts(cuts, |
---|
956 | newNode->numberBranches(),whichGenerator_) ; |
---|
957 | if (lastws) delete lastws ; |
---|
958 | lastws = dynamic_cast<CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
959 | } |
---|
960 | /* |
---|
961 | Continuous data to be used later |
---|
962 | */ |
---|
963 | continuousObjective_ = solver_->getObjValue()*solver_->getObjSense(); |
---|
964 | continuousInfeasibilities_ = 0 ; |
---|
965 | if (newNode) |
---|
966 | { continuousObjective_ = newNode->objectiveValue() ; |
---|
967 | delete [] continuousSolution_; |
---|
968 | continuousSolution_ = CoinCopyOfArray(solver_->getColSolution(), |
---|
969 | numberColumns); |
---|
970 | continuousInfeasibilities_ = newNode->numberUnsatisfied() ; } |
---|
971 | /* |
---|
972 | Bound may have changed so reset in objects |
---|
973 | */ |
---|
974 | { int i ; |
---|
975 | for (i = 0;i < numberObjects_;i++) |
---|
976 | object_[i]->resetBounds() ; } |
---|
977 | stoppedOnGap_ = false ; |
---|
978 | /* |
---|
979 | Feasible? Then we should have either a live node prepped for future |
---|
980 | expansion (indicated by variable() >= 0), or (miracle of miracles) an |
---|
981 | integral solution at the root node. |
---|
982 | |
---|
983 | initializeInfo sets the reference counts in the nodeInfo object. Since |
---|
984 | this node is still live, push it onto the heap that holds the live set. |
---|
985 | */ |
---|
986 | double bestValue = 0.0 ; |
---|
987 | if (newNode) { |
---|
988 | bestValue = newNode->objectiveValue(); |
---|
989 | if (newNode->variable() >= 0) { |
---|
990 | newNode->initializeInfo() ; |
---|
991 | tree_->push(newNode) ; |
---|
992 | if (statistics_) { |
---|
993 | if (numberNodes2_==maximumStatistics_) { |
---|
994 | maximumStatistics_ = 2*maximumStatistics_; |
---|
995 | CbcStatistics ** temp = new CbcStatistics * [maximumStatistics_]; |
---|
996 | memset(temp,0,maximumStatistics_*sizeof(CbcStatistics *)); |
---|
997 | memcpy(temp,statistics_,numberNodes2_*sizeof(CbcStatistics *)); |
---|
998 | delete [] statistics_; |
---|
999 | statistics_=temp; |
---|
1000 | } |
---|
1001 | assert (!statistics_[numberNodes2_]); |
---|
1002 | statistics_[numberNodes2_]=new CbcStatistics(newNode); |
---|
1003 | } |
---|
1004 | numberNodes2_++; |
---|
1005 | # ifdef CHECK_NODE |
---|
1006 | printf("Node %x on tree\n",newNode) ; |
---|
1007 | # endif |
---|
1008 | } else { |
---|
1009 | // continuous is integer |
---|
1010 | double objectiveValue = newNode->objectiveValue(); |
---|
1011 | setBestSolution(CBC_SOLUTION,objectiveValue, |
---|
1012 | solver_->getColSolution()) ; |
---|
1013 | delete newNode ; |
---|
1014 | newNode = NULL ; |
---|
1015 | } |
---|
1016 | } |
---|
1017 | |
---|
1018 | if (printFrequency_ <= 0) { |
---|
1019 | printFrequency_ = 1000 ; |
---|
1020 | if (getNumCols() > 2000) |
---|
1021 | printFrequency_ = 100 ; |
---|
1022 | } |
---|
1023 | /* |
---|
1024 | It is possible that strong branching fixes one variable and then the code goes round |
---|
1025 | again and again. This can take too long. So we need to warn user - just once. |
---|
1026 | */ |
---|
1027 | numberLongStrong_=0; |
---|
1028 | /* |
---|
1029 | At last, the actual branch-and-cut search loop, which will iterate until |
---|
1030 | the live set is empty or we hit some limit (integrality gap, time, node |
---|
1031 | count, etc.). The overall flow is to rebuild a subproblem, reoptimise using |
---|
1032 | solveWithCuts(), choose a branching pattern with chooseBranch(), and finally |
---|
1033 | add the node to the live set. |
---|
1034 | |
---|
1035 | The first action is to winnow the live set to remove nodes which are worse |
---|
1036 | than the current objective cutoff. |
---|
1037 | */ |
---|
1038 | while (!tree_->empty()) { |
---|
1039 | if (cutoff > getCutoff()) { |
---|
1040 | double newCutoff = getCutoff(); |
---|
1041 | if (analyzeResults_) { |
---|
1042 | // see if we could fix any (more) |
---|
1043 | int n=0; |
---|
1044 | double * newLower = analyzeResults_; |
---|
1045 | double * objLower = newLower+numberIntegers_; |
---|
1046 | double * newUpper = objLower+numberIntegers_; |
---|
1047 | double * objUpper = newUpper+numberIntegers_; |
---|
1048 | for (int i=0;i<numberIntegers_;i++) { |
---|
1049 | if (objLower[i]>newCutoff) { |
---|
1050 | n++; |
---|
1051 | if (objUpper[i]>newCutoff) { |
---|
1052 | newCutoff = -COIN_DBL_MAX; |
---|
1053 | break; |
---|
1054 | } |
---|
1055 | } else if (objUpper[i]>newCutoff) { |
---|
1056 | n++; |
---|
1057 | } |
---|
1058 | } |
---|
1059 | if (newCutoff==-COIN_DBL_MAX) { |
---|
1060 | printf("Root analysis says finished\n"); |
---|
1061 | } else if (n>numberFixedNow_) { |
---|
1062 | printf("%d more fixed by analysis - now %d\n",n-numberFixedNow_,n); |
---|
1063 | numberFixedNow_=n; |
---|
1064 | } |
---|
1065 | } |
---|
1066 | if (eventHandler) { |
---|
1067 | if (!eventHandler->event(CbcEventHandler::solution)) { |
---|
1068 | eventHappened_=true; // exit |
---|
1069 | } |
---|
1070 | } |
---|
1071 | // Do from deepest |
---|
1072 | tree_->cleanTree(this, newCutoff,bestPossibleObjective_) ; |
---|
1073 | nodeCompare_->newSolution(this) ; |
---|
1074 | nodeCompare_->newSolution(this,continuousObjective_, |
---|
1075 | continuousInfeasibilities_) ; |
---|
1076 | tree_->setComparison(*nodeCompare_) ; |
---|
1077 | if (tree_->empty()) |
---|
1078 | break; // finished |
---|
1079 | } |
---|
1080 | cutoff = getCutoff() ; |
---|
1081 | /* |
---|
1082 | Periodic activities: Opportunities to |
---|
1083 | + tweak the nodeCompare criteria, |
---|
1084 | + check if we've closed the integrality gap enough to quit, |
---|
1085 | + print a summary line to let the user know we're working |
---|
1086 | */ |
---|
1087 | if ((numberNodes_%1000) == 0) { |
---|
1088 | bool redoTree=nodeCompare_->every1000Nodes(this, numberNodes_) ; |
---|
1089 | // redo tree if wanted |
---|
1090 | if (redoTree) |
---|
1091 | tree_->setComparison(*nodeCompare_) ; |
---|
1092 | } |
---|
1093 | if (saveCompare&&!hotstartSolution_) { |
---|
1094 | // hotstart switched off |
---|
1095 | delete nodeCompare_; // off depth first |
---|
1096 | nodeCompare_=saveCompare; |
---|
1097 | saveCompare=NULL; |
---|
1098 | // redo tree |
---|
1099 | tree_->setComparison(*nodeCompare_) ; |
---|
1100 | } |
---|
1101 | if ((numberNodes_%printFrequency_) == 0) { |
---|
1102 | int j ; |
---|
1103 | int nNodes = tree_->size() ; |
---|
1104 | bestPossibleObjective_ = 1.0e100 ; |
---|
1105 | for (j = 0;j < nNodes;j++) { |
---|
1106 | CbcNode * node = tree_->nodePointer(j) ; |
---|
1107 | if (node&&node->objectiveValue() < bestPossibleObjective_) |
---|
1108 | bestPossibleObjective_ = node->objectiveValue() ; |
---|
1109 | } |
---|
1110 | messageHandler()->message(CBC_STATUS,messages()) |
---|
1111 | << numberNodes_<< nNodes<< bestObjective_<< bestPossibleObjective_ |
---|
1112 | <<getCurrentSeconds() |
---|
1113 | << CoinMessageEol ; |
---|
1114 | if (!eventHandler->event(CbcEventHandler::treeStatus)) { |
---|
1115 | eventHappened_=true; // exit |
---|
1116 | } |
---|
1117 | } |
---|
1118 | // If no solution but many nodes - signal change in strategy |
---|
1119 | if (numberNodes_>2*numberObjects_+1000&&stateOfSearch_!=2) |
---|
1120 | stateOfSearch_=3; |
---|
1121 | // See if can stop on gap |
---|
1122 | double testGap = CoinMax(dblParam_[CbcAllowableGap], |
---|
1123 | CoinMax(fabs(bestObjective_),fabs(bestPossibleObjective_)) |
---|
1124 | *dblParam_[CbcAllowableFractionGap]); |
---|
1125 | if (bestObjective_-bestPossibleObjective_ < testGap && getCutoffIncrement()>=0.0) { |
---|
1126 | stoppedOnGap_ = true ; |
---|
1127 | } |
---|
1128 | |
---|
1129 | # ifdef CHECK_NODE_FULL |
---|
1130 | verifyTreeNodes(tree_,*this) ; |
---|
1131 | # endif |
---|
1132 | # ifdef CHECK_CUT_COUNTS |
---|
1133 | verifyCutCounts(tree_,*this) ; |
---|
1134 | # endif |
---|
1135 | |
---|
1136 | /* |
---|
1137 | Now we come to the meat of the loop. To create the active subproblem, we'll |
---|
1138 | pop the most promising node in the live set, rebuild the subproblem it |
---|
1139 | represents, and then execute the current arm of the branch to create the |
---|
1140 | active subproblem. |
---|
1141 | */ |
---|
1142 | CbcNode *node = tree_->bestNode(cutoff) ; |
---|
1143 | // Possible one on tree worse than cutoff |
---|
1144 | if (!node) |
---|
1145 | continue; |
---|
1146 | currentNode_=node; // so can be accessed elsewhere |
---|
1147 | #ifdef CBC_DEBUG |
---|
1148 | printf("%d unsat, way %d, obj %g est %g\n", |
---|
1149 | node->numberUnsatisfied(),node->way(),node->objectiveValue(), |
---|
1150 | node->guessedObjectiveValue()); |
---|
1151 | #endif |
---|
1152 | // Save clone in branching decision |
---|
1153 | if(branchingMethod_) |
---|
1154 | branchingMethod_->saveBranchingObject(node->modifiableBranchingObject()); |
---|
1155 | bool nodeOnTree=false; // Node has been popped |
---|
1156 | // Say not on optimal path |
---|
1157 | bool onOptimalPath=false; |
---|
1158 | # ifdef CHECK_NODE |
---|
1159 | /* |
---|
1160 | WARNING: The use of integerVariable_[*] here will break as soon as the |
---|
1161 | branching object is something other than an integer variable. |
---|
1162 | This needs some thought. |
---|
1163 | */ |
---|
1164 | printf("Node %x popped from tree - %d left, %d count\n",node, |
---|
1165 | node->nodeInfo()->numberBranchesLeft(), |
---|
1166 | node->nodeInfo()->numberPointingToThis()) ; |
---|
1167 | printf("\tdepth = %d, z = %g, unsat = %d, var = %d.\n", |
---|
1168 | node->depth(),node->objectiveValue(), |
---|
1169 | node->numberUnsatisfied(), |
---|
1170 | integerVariable_[node->variable()]) ; |
---|
1171 | # endif |
---|
1172 | |
---|
1173 | /* |
---|
1174 | Rebuild the subproblem for this node: Call addCuts() to adjust the model |
---|
1175 | to recreate the subproblem for this node (set proper variable bounds, add |
---|
1176 | cuts, create a basis). This may result in the problem being fathomed by |
---|
1177 | bound or infeasibility. Returns 1 if node is fathomed. |
---|
1178 | Execute the current arm of the branch: If the problem survives, save the |
---|
1179 | resulting variable bounds and call branch() to modify variable bounds |
---|
1180 | according to the current arm of the branching object. If we're processing |
---|
1181 | the final arm of the branching object, flag the node for removal from the |
---|
1182 | live set. |
---|
1183 | */ |
---|
1184 | CbcNodeInfo * nodeInfo = node->nodeInfo() ; |
---|
1185 | newNode = NULL ; |
---|
1186 | if (!addCuts(node,lastws,numberFixedNow_>numberFixedAtRoot_)) |
---|
1187 | { int i ; |
---|
1188 | const double * lower = getColLower() ; |
---|
1189 | const double * upper = getColUpper() ; |
---|
1190 | for (i = 0 ; i < numberColumns ; i++) |
---|
1191 | { lowerBefore[i]= lower[i] ; |
---|
1192 | upperBefore[i]= upper[i] ; } |
---|
1193 | bool deleteNode ; |
---|
1194 | if (node->branch()) |
---|
1195 | { |
---|
1196 | // set nodenumber correctly |
---|
1197 | node->nodeInfo()->setNodeNumber(numberNodes2_); |
---|
1198 | tree_->push(node) ; |
---|
1199 | if (statistics_) { |
---|
1200 | if (numberNodes2_==maximumStatistics_) { |
---|
1201 | maximumStatistics_ = 2*maximumStatistics_; |
---|
1202 | CbcStatistics ** temp = new CbcStatistics * [maximumStatistics_]; |
---|
1203 | memset(temp,0,maximumStatistics_*sizeof(CbcStatistics *)); |
---|
1204 | memcpy(temp,statistics_,numberNodes2_*sizeof(CbcStatistics *)); |
---|
1205 | delete [] statistics_; |
---|
1206 | statistics_=temp; |
---|
1207 | } |
---|
1208 | assert (!statistics_[numberNodes2_]); |
---|
1209 | statistics_[numberNodes2_]=new CbcStatistics(node); |
---|
1210 | } |
---|
1211 | numberNodes2_++; |
---|
1212 | nodeOnTree=true; // back on tree |
---|
1213 | deleteNode = false ; |
---|
1214 | # ifdef CHECK_NODE |
---|
1215 | printf("Node %x pushed back on tree - %d left, %d count\n",node, |
---|
1216 | nodeInfo->numberBranchesLeft(), |
---|
1217 | nodeInfo->numberPointingToThis()) ; |
---|
1218 | # endif |
---|
1219 | } |
---|
1220 | else |
---|
1221 | { deleteNode = true ; |
---|
1222 | if (!nodeInfo->numberBranchesLeft()) |
---|
1223 | nodeInfo->allBranchesGone(); // can clean up |
---|
1224 | } |
---|
1225 | |
---|
1226 | if ((specialOptions_&1)!=0) { |
---|
1227 | /* |
---|
1228 | This doesn't work as intended --- getRowCutDebugger will return null |
---|
1229 | unless the current feasible solution region includes the optimal solution |
---|
1230 | that RowCutDebugger knows. There's no way to tell inactive from off the |
---|
1231 | optimal path. |
---|
1232 | */ |
---|
1233 | const OsiRowCutDebugger *debugger = solver_->getRowCutDebugger() ; |
---|
1234 | if (debugger) { |
---|
1235 | onOptimalPath=true; |
---|
1236 | printf("On optimal path\n") ; |
---|
1237 | } |
---|
1238 | } |
---|
1239 | /* |
---|
1240 | Reoptimize, possibly generating cuts and/or using heuristics to find |
---|
1241 | solutions. Cut reference counts are unaffected unless we lose feasibility, |
---|
1242 | in which case solveWithCuts() will make the adjustment. |
---|
1243 | */ |
---|
1244 | phase_=2; |
---|
1245 | cuts = OsiCuts() ; |
---|
1246 | currentNumberCuts = solver_->getNumRows()-numberRowsAtContinuous_ ; |
---|
1247 | int saveNumber = numberIterations_; |
---|
1248 | if(solverCharacteristics_->solutionAddsCuts()) { |
---|
1249 | int returnCode=resolve(node ? node->nodeInfo() : NULL,1); |
---|
1250 | feasible = returnCode != 0; |
---|
1251 | if (feasible) { |
---|
1252 | int iObject ; |
---|
1253 | int preferredWay ; |
---|
1254 | int numberUnsatisfied = 0 ; |
---|
1255 | memcpy(currentSolution_,solver_->getColSolution(), |
---|
1256 | numberColumns*sizeof(double)) ; |
---|
1257 | |
---|
1258 | for (iObject = 0 ; iObject < numberObjects_ ; iObject++) { |
---|
1259 | double infeasibility = |
---|
1260 | object_[iObject]->infeasibility(preferredWay) ; |
---|
1261 | if (infeasibility ) numberUnsatisfied++ ; |
---|
1262 | } |
---|
1263 | if (returnCode>0) { |
---|
1264 | if (numberUnsatisfied) { |
---|
1265 | feasible = solveWithCuts(cuts,maximumCutPasses_,node); |
---|
1266 | } else { |
---|
1267 | // may generate cuts and turn the solution |
---|
1268 | //to an infeasible one |
---|
1269 | feasible = solveWithCuts(cuts, 1, |
---|
1270 | node); |
---|
1271 | #if 0 |
---|
1272 | currentNumberCuts_ = cuts.sizeRowCuts(); |
---|
1273 | if (currentNumberCuts_ >= maximumNumberCuts_) { |
---|
1274 | maximumNumberCuts_ = currentNumberCuts; |
---|
1275 | delete [] addedCuts_; |
---|
1276 | addedCuts_ = new CbcCountRowCut * [maximumNumberCuts_]; |
---|
1277 | } |
---|
1278 | #endif |
---|
1279 | } |
---|
1280 | } |
---|
1281 | // check extra info on feasibility |
---|
1282 | if (!solverCharacteristics_->mipFeasible()) |
---|
1283 | feasible = false; |
---|
1284 | } |
---|
1285 | } else { |
---|
1286 | // normal |
---|
1287 | feasible = solveWithCuts(cuts,maximumCutPasses_,node); |
---|
1288 | } |
---|
1289 | if ((specialOptions_&1)!=0&&onOptimalPath) { |
---|
1290 | const OsiRowCutDebugger *debugger = solver_->getRowCutDebugger() ; |
---|
1291 | assert (debugger) ; |
---|
1292 | } |
---|
1293 | if (statistics_) { |
---|
1294 | assert (numberNodes2_); |
---|
1295 | assert (statistics_[numberNodes2_-1]); |
---|
1296 | assert (statistics_[numberNodes2_-1]->node()==numberNodes2_-1); |
---|
1297 | statistics_[numberNodes2_-1]->endOfBranch(numberIterations_-saveNumber, |
---|
1298 | feasible ? solver_->getObjValue() |
---|
1299 | : COIN_DBL_MAX); |
---|
1300 | } |
---|
1301 | /* |
---|
1302 | Check for abort on limits: node count, solution count, time, integrality gap. |
---|
1303 | */ |
---|
1304 | double totalTime = CoinCpuTime()-dblParam_[CbcStartSeconds] ; |
---|
1305 | if (numberNodes_ < intParam_[CbcMaxNumNode] && |
---|
1306 | numberSolutions_ < intParam_[CbcMaxNumSol] && |
---|
1307 | totalTime < dblParam_[CbcMaximumSeconds] && |
---|
1308 | !stoppedOnGap_&&!eventHappened_) |
---|
1309 | { |
---|
1310 | /* |
---|
1311 | Are we still feasible? If so, create a node and do the work to attach a |
---|
1312 | branching object, reoptimising as needed if chooseBranch() identifies |
---|
1313 | monotone objects. |
---|
1314 | |
---|
1315 | Finally, attach a partial nodeInfo object and store away any cuts that we |
---|
1316 | created back in solveWithCuts. addCuts() will initialise the reference |
---|
1317 | counts for these new cuts. |
---|
1318 | |
---|
1319 | TODO: (lh) I'm confused. We create a nodeInfo without checking whether we |
---|
1320 | have a solution or not. Then we use numberUnsatisfied() to decide |
---|
1321 | whether to stash the cuts and bump reference counts. Other places we |
---|
1322 | use variable() (i.e., presence of a branching variable). Equivalent? |
---|
1323 | */ |
---|
1324 | if (onOptimalPath) { |
---|
1325 | if (!feasible) { |
---|
1326 | printf("infeas2\n"); |
---|
1327 | solver_->writeMps("infeas"); |
---|
1328 | CoinWarmStartBasis *slack = |
---|
1329 | dynamic_cast<CoinWarmStartBasis *>(solver_->getEmptyWarmStart()) ; |
---|
1330 | solver_->setWarmStart(slack); |
---|
1331 | delete slack ; |
---|
1332 | solver_->setHintParam(OsiDoReducePrint,false,OsiHintDo,0) ; |
---|
1333 | solver_->initialSolve(); |
---|
1334 | assert (!solver_->isProvenOptimal()); |
---|
1335 | } |
---|
1336 | assert (feasible); |
---|
1337 | } |
---|
1338 | bool checkingNode=false; |
---|
1339 | if (feasible) { |
---|
1340 | newNode = new CbcNode ; |
---|
1341 | // Set objective value (not so obvious if NLP etc) |
---|
1342 | setObjectiveValue(newNode,node); |
---|
1343 | anyAction =-1 ; |
---|
1344 | resolved = false ; |
---|
1345 | if (newNode->objectiveValue() >= getCutoff()) |
---|
1346 | anyAction=-2; |
---|
1347 | // only allow twenty passes |
---|
1348 | int numberPassesLeft=20; |
---|
1349 | checkingNode=true; |
---|
1350 | while (anyAction == -1) |
---|
1351 | { |
---|
1352 | // Set objective value (not so obvious if NLP etc) |
---|
1353 | setObjectiveValue(newNode,node); |
---|
1354 | if (numberBeforeTrust_==0 ) { |
---|
1355 | anyAction = newNode->chooseBranch(this,node,numberPassesLeft) ; |
---|
1356 | } else { |
---|
1357 | OsiSolverBranch * branches=NULL; |
---|
1358 | anyAction = newNode->chooseDynamicBranch(this,node,branches,numberPassesLeft) ; |
---|
1359 | if (anyAction==-3) |
---|
1360 | anyAction = newNode->chooseBranch(this,node,numberPassesLeft) ; // dynamic did nothing |
---|
1361 | } |
---|
1362 | /* |
---|
1363 | Yep, false positives for sure. And no easy way to distinguish honest |
---|
1364 | infeasibility from `found a solution and tightened objective target.' |
---|
1365 | |
---|
1366 | if (onOptimalPath) |
---|
1367 | assert (anyAction!=-2); // can be useful but gives false positives on strong |
---|
1368 | */ |
---|
1369 | numberPassesLeft--; |
---|
1370 | if (numberPassesLeft<=-1) { |
---|
1371 | if (!numberLongStrong_) |
---|
1372 | messageHandler()->message(CBC_WARNING_STRONG, |
---|
1373 | messages()) << CoinMessageEol ; |
---|
1374 | numberLongStrong_++; |
---|
1375 | } |
---|
1376 | if (anyAction == -1) |
---|
1377 | { |
---|
1378 | // can do quick optimality check |
---|
1379 | int easy=2; |
---|
1380 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,&easy) ; |
---|
1381 | feasible = resolve(node ? node->nodeInfo() : NULL,11) != 0 ; |
---|
1382 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,NULL) ; |
---|
1383 | resolved = true ; |
---|
1384 | if (problemFeasibility_->feasible(this,0)<0) { |
---|
1385 | feasible=false; // pretend infeasible |
---|
1386 | } |
---|
1387 | if (feasible) |
---|
1388 | { |
---|
1389 | // Set objective value (not so obvious if NLP etc) |
---|
1390 | setObjectiveValue(newNode,node); |
---|
1391 | reducedCostFix() ; |
---|
1392 | if (newNode->objectiveValue() >= getCutoff()) |
---|
1393 | anyAction=-2; |
---|
1394 | } |
---|
1395 | else |
---|
1396 | { anyAction = -2 ; } } } |
---|
1397 | if (anyAction >= 0) |
---|
1398 | { if (resolved) |
---|
1399 | { bool needValidSolution = (newNode->variable() < 0) ; |
---|
1400 | takeOffCuts(cuts,needValidSolution,NULL) ; |
---|
1401 | # ifdef CHECK_CUT_COUNTS |
---|
1402 | { printf("Number of rows after chooseBranch fix (node)" |
---|
1403 | "(active only) %d\n", |
---|
1404 | numberRowsAtContinuous_+numberNewCuts_+ |
---|
1405 | numberOldActiveCuts_) ; |
---|
1406 | const CoinWarmStartBasis* debugws = |
---|
1407 | dynamic_cast<const CoinWarmStartBasis*> |
---|
1408 | (solver_->getWarmStart()) ; |
---|
1409 | debugws->print() ; |
---|
1410 | delete debugws ; } |
---|
1411 | # endif |
---|
1412 | } |
---|
1413 | newNode->createInfo(this,node,lastws,lowerBefore,upperBefore, |
---|
1414 | numberOldActiveCuts_,numberNewCuts_) ; |
---|
1415 | if (newNode->numberUnsatisfied()) { |
---|
1416 | newNode->initializeInfo() ; |
---|
1417 | newNode->nodeInfo()->addCuts(cuts,newNode->numberBranches(), |
---|
1418 | whichGenerator_) ; } } } |
---|
1419 | else |
---|
1420 | { anyAction = -2 ; } |
---|
1421 | // May have slipped through i.e. anyAction == 0 and objective above cutoff |
---|
1422 | // I think this will screw up cut reference counts if executed. |
---|
1423 | // We executed addCuts just above. (lh) |
---|
1424 | if ( anyAction >=0 ) { |
---|
1425 | assert (newNode); |
---|
1426 | if (newNode->objectiveValue() >= getCutoff()) |
---|
1427 | anyAction = -2; // say bad after all |
---|
1428 | } |
---|
1429 | /* |
---|
1430 | If we end up infeasible, we can delete the new node immediately. Since this |
---|
1431 | node won't be needing the cuts we collected, decrement the reference counts. |
---|
1432 | If we are feasible, then we'll be placing this node into the live set, so |
---|
1433 | increment the reference count in the current (parent) nodeInfo. |
---|
1434 | */ |
---|
1435 | if (anyAction == -2) |
---|
1436 | { delete newNode ; |
---|
1437 | newNode = NULL ; |
---|
1438 | // say strong doing well |
---|
1439 | if (checkingNode) |
---|
1440 | setSpecialOptions(specialOptions_|8); |
---|
1441 | for (i = 0 ; i < currentNumberCuts_ ; i++) |
---|
1442 | { if (addedCuts_[i]) |
---|
1443 | { if (!addedCuts_[i]->decrement(1)) |
---|
1444 | delete addedCuts_[i] ; } } } |
---|
1445 | else |
---|
1446 | { nodeInfo->increment() ; |
---|
1447 | if ((numberNodes_%20)==0) { |
---|
1448 | // say strong not doing as well |
---|
1449 | setSpecialOptions(specialOptions_&~8); |
---|
1450 | } |
---|
1451 | } |
---|
1452 | /* |
---|
1453 | At this point, there are three possibilities: |
---|
1454 | * newNode is live and will require further branching to resolve |
---|
1455 | (variable() >= 0). Increment the cut reference counts by |
---|
1456 | numberBranches() to allow for use by children of this node, and |
---|
1457 | decrement by 1 because we've executed one arm of the branch of our |
---|
1458 | parent (consuming one reference). Before we push newNode onto the |
---|
1459 | search tree, try for a heuristic solution. |
---|
1460 | * We have a solution, in which case newNode is non-null but we have no |
---|
1461 | branching variable. Decrement the cut counts and save the solution. |
---|
1462 | * The node was found to be infeasible, in which case it's already been |
---|
1463 | deleted, and newNode is null. |
---|
1464 | */ |
---|
1465 | if (!eventHandler->event(CbcEventHandler::node)) { |
---|
1466 | eventHappened_=true; // exit |
---|
1467 | } |
---|
1468 | assert (!newNode || newNode->objectiveValue() <= getCutoff()) ; |
---|
1469 | if (statistics_) { |
---|
1470 | assert (numberNodes2_); |
---|
1471 | assert (statistics_[numberNodes2_-1]); |
---|
1472 | assert (statistics_[numberNodes2_-1]->node()==numberNodes2_-1); |
---|
1473 | if (newNode) |
---|
1474 | statistics_[numberNodes2_-1]->updateInfeasibility(newNode->numberUnsatisfied()); |
---|
1475 | else |
---|
1476 | statistics_[numberNodes2_-1]->sayInfeasible(); |
---|
1477 | } |
---|
1478 | if (newNode) |
---|
1479 | { if (newNode->variable() >= 0) |
---|
1480 | { handler_->message(CBC_BRANCH,messages_) |
---|
1481 | << numberNodes_<< newNode->objectiveValue() |
---|
1482 | << newNode->numberUnsatisfied()<< newNode->depth() |
---|
1483 | << CoinMessageEol ; |
---|
1484 | // Increment cut counts (taking off current) |
---|
1485 | int numberLeft = newNode->numberBranches() ; |
---|
1486 | for (i = 0;i < currentNumberCuts_;i++) |
---|
1487 | { if (addedCuts_[i]) |
---|
1488 | { |
---|
1489 | # ifdef CHECK_CUT_COUNTS |
---|
1490 | printf("Count on cut %x increased by %d\n",addedCuts_[i], |
---|
1491 | numberLeft-1) ; |
---|
1492 | # endif |
---|
1493 | addedCuts_[i]->increment(numberLeft-1) ; } } |
---|
1494 | |
---|
1495 | double estValue = newNode->guessedObjectiveValue() ; |
---|
1496 | int found = -1 ; |
---|
1497 | // no - overhead on small problems solver_->resolve() ; // double check current optimal |
---|
1498 | // assert (!solver_->getIterationCount()); |
---|
1499 | double * newSolution = new double [numberColumns] ; |
---|
1500 | double heurValue = getCutoff() ; |
---|
1501 | int iHeur ; |
---|
1502 | for (iHeur = 0 ; iHeur < numberHeuristics_ ; iHeur++) |
---|
1503 | { double saveValue = heurValue ; |
---|
1504 | int ifSol = heuristic_[iHeur]->solution(heurValue,newSolution) ; |
---|
1505 | if (ifSol > 0) { |
---|
1506 | // new solution found |
---|
1507 | found = iHeur ; |
---|
1508 | incrementUsed(newSolution); |
---|
1509 | } |
---|
1510 | else |
---|
1511 | if (ifSol < 0) // just returning an estimate |
---|
1512 | { estValue = CoinMin(heurValue,estValue) ; |
---|
1513 | heurValue = saveValue ; } } |
---|
1514 | if (found >= 0) { |
---|
1515 | setBestSolution(CBC_ROUNDING,heurValue,newSolution) ; |
---|
1516 | lastHeuristic_ = heuristic_[found]; |
---|
1517 | } |
---|
1518 | delete [] newSolution ; |
---|
1519 | newNode->setGuessedObjectiveValue(estValue) ; |
---|
1520 | tree_->push(newNode) ; |
---|
1521 | if (statistics_) { |
---|
1522 | if (numberNodes2_==maximumStatistics_) { |
---|
1523 | maximumStatistics_ = 2*maximumStatistics_; |
---|
1524 | CbcStatistics ** temp = new CbcStatistics * [maximumStatistics_]; |
---|
1525 | memset(temp,0,maximumStatistics_*sizeof(CbcStatistics *)); |
---|
1526 | memcpy(temp,statistics_,numberNodes2_*sizeof(CbcStatistics *)); |
---|
1527 | delete [] statistics_; |
---|
1528 | statistics_=temp; |
---|
1529 | } |
---|
1530 | assert (!statistics_[numberNodes2_]); |
---|
1531 | statistics_[numberNodes2_]=new CbcStatistics(newNode); |
---|
1532 | } |
---|
1533 | numberNodes2_++; |
---|
1534 | # ifdef CHECK_NODE |
---|
1535 | printf("Node %x pushed on tree c\n",newNode) ; |
---|
1536 | # endif |
---|
1537 | } |
---|
1538 | else |
---|
1539 | { for (i = 0 ; i < currentNumberCuts_ ; i++) |
---|
1540 | { if (addedCuts_[i]) |
---|
1541 | { if (!addedCuts_[i]->decrement(1)) |
---|
1542 | delete addedCuts_[i] ; } } |
---|
1543 | double objectiveValue = newNode->objectiveValue(); |
---|
1544 | setBestSolution(CBC_SOLUTION,objectiveValue, |
---|
1545 | solver_->getColSolution()) ; |
---|
1546 | lastHeuristic_ = NULL; |
---|
1547 | incrementUsed(solver_->getColSolution()); |
---|
1548 | //assert(nodeInfo->numberPointingToThis() <= 2) ; |
---|
1549 | // avoid accidental pruning, if newNode was final branch arm |
---|
1550 | nodeInfo->increment(); |
---|
1551 | delete newNode ; |
---|
1552 | nodeInfo->decrement() ; } } |
---|
1553 | /* |
---|
1554 | This node has been completely expanded and can be removed from the live |
---|
1555 | set. |
---|
1556 | */ |
---|
1557 | if (deleteNode) |
---|
1558 | delete node ; } |
---|
1559 | /* |
---|
1560 | End of the non-abort actions. The next block of code is executed if we've |
---|
1561 | aborted because we hit one of the limits. Clean up by deleting the live set |
---|
1562 | and break out of the node processing loop. Note that on an abort, node may |
---|
1563 | have been pushed back onto the tree for further processing, in which case |
---|
1564 | it'll be deleted in cleanTree. We need to check. |
---|
1565 | */ |
---|
1566 | else |
---|
1567 | { |
---|
1568 | tree_->cleanTree(this,-COIN_DBL_MAX,bestPossibleObjective_) ; |
---|
1569 | delete nextRowCut_; |
---|
1570 | // We need to get rid of node if is has already been popped from tree |
---|
1571 | if (!nodeOnTree&&!stoppedOnGap_&&node!=rootNode) |
---|
1572 | delete node; |
---|
1573 | if (stoppedOnGap_) |
---|
1574 | { messageHandler()->message(CBC_GAP,messages()) |
---|
1575 | << bestObjective_-bestPossibleObjective_ |
---|
1576 | << dblParam_[CbcAllowableGap] |
---|
1577 | << dblParam_[CbcAllowableFractionGap]*100.0 |
---|
1578 | << CoinMessageEol ; |
---|
1579 | secondaryStatus_ = 2; |
---|
1580 | status_ = 0 ; } |
---|
1581 | else |
---|
1582 | if (isNodeLimitReached()) |
---|
1583 | { handler_->message(CBC_MAXNODES,messages_) << CoinMessageEol ; |
---|
1584 | secondaryStatus_ = 3; |
---|
1585 | status_ = 1 ; } |
---|
1586 | else |
---|
1587 | if (totalTime >= dblParam_[CbcMaximumSeconds]) |
---|
1588 | { handler_->message(CBC_MAXTIME,messages_) << CoinMessageEol ; |
---|
1589 | secondaryStatus_ = 4; |
---|
1590 | status_ = 1 ; } |
---|
1591 | else |
---|
1592 | if (eventHappened_) |
---|
1593 | { handler_->message(CBC_EVENT,messages_) << CoinMessageEol ; |
---|
1594 | secondaryStatus_ = 5; |
---|
1595 | status_ = 5 ; } |
---|
1596 | else |
---|
1597 | { handler_->message(CBC_MAXSOLS,messages_) << CoinMessageEol ; |
---|
1598 | secondaryStatus_ = 6; |
---|
1599 | status_ = 1 ; } |
---|
1600 | break ; } |
---|
1601 | /* |
---|
1602 | Delete cuts to get back to the original system. |
---|
1603 | |
---|
1604 | I'm thinking this is redundant --- the call to addCuts that conditions entry |
---|
1605 | to this code block also performs this action. |
---|
1606 | */ |
---|
1607 | int numberToDelete = getNumRows()-numberRowsAtContinuous_ ; |
---|
1608 | if (numberToDelete) |
---|
1609 | { int * delRows = new int[numberToDelete] ; |
---|
1610 | int i ; |
---|
1611 | for (i = 0 ; i < numberToDelete ; i++) |
---|
1612 | { delRows[i] = i+numberRowsAtContinuous_ ; } |
---|
1613 | solver_->deleteRows(numberToDelete,delRows) ; |
---|
1614 | delete [] delRows ; } } |
---|
1615 | /* |
---|
1616 | This node fathomed when addCuts atttempted to revive it. Toss it. |
---|
1617 | */ |
---|
1618 | else |
---|
1619 | { delete node ; } } |
---|
1620 | /* |
---|
1621 | That's it, we've exhausted the search tree, or broken out of the loop because |
---|
1622 | we hit some limit on evaluation. |
---|
1623 | |
---|
1624 | We may have got an intelligent tree so give it one more chance |
---|
1625 | */ |
---|
1626 | // Tell solver we are not in Branch and Cut |
---|
1627 | solver_->setHintParam(OsiDoInBranchAndCut,false,OsiHintDo,NULL) ; |
---|
1628 | tree_->endSearch(); |
---|
1629 | // If we did any sub trees - did we give up on any? |
---|
1630 | if ( numberStoppedSubTrees_) |
---|
1631 | status_=1; |
---|
1632 | if (!status_) { |
---|
1633 | bestPossibleObjective_=bestObjective_; |
---|
1634 | handler_->message(CBC_END_GOOD,messages_) |
---|
1635 | << bestObjective_ << numberIterations_ << numberNodes_<<getCurrentSeconds() |
---|
1636 | << CoinMessageEol ; |
---|
1637 | } else { |
---|
1638 | handler_->message(CBC_END,messages_) |
---|
1639 | << bestObjective_ <<bestPossibleObjective_ |
---|
1640 | << numberIterations_ << numberNodes_<<getCurrentSeconds() |
---|
1641 | << CoinMessageEol ; |
---|
1642 | } |
---|
1643 | if (numberStrongIterations_) |
---|
1644 | handler_->message(CBC_STRONG_STATS,messages_) |
---|
1645 | << strongInfo_[0] << numberStrongIterations_ << strongInfo_[2] |
---|
1646 | << strongInfo_[1] << CoinMessageEol ; |
---|
1647 | if (statistics_) { |
---|
1648 | // report in some way |
---|
1649 | int * lookup = new int[numberObjects_]; |
---|
1650 | int i; |
---|
1651 | for (i=0;i<numberObjects_;i++) |
---|
1652 | lookup[i]=-1; |
---|
1653 | bool goodIds=true; |
---|
1654 | for (i=0;i<numberObjects_;i++) { |
---|
1655 | int id = object_[i]->id(); |
---|
1656 | int iColumn = object_[i]->columnNumber(); |
---|
1657 | if (iColumn<0) |
---|
1658 | iColumn = id+numberColumns; |
---|
1659 | if(id>=0&&id<numberObjects_) { |
---|
1660 | if (lookup[id]==-1) { |
---|
1661 | lookup[id]=iColumn; |
---|
1662 | } else { |
---|
1663 | goodIds=false; |
---|
1664 | break; |
---|
1665 | } |
---|
1666 | } else { |
---|
1667 | goodIds=false; |
---|
1668 | break; |
---|
1669 | } |
---|
1670 | } |
---|
1671 | if (!goodIds) { |
---|
1672 | delete [] lookup; |
---|
1673 | lookup=NULL; |
---|
1674 | } |
---|
1675 | if (doStatistics==3) { |
---|
1676 | printf(" node parent depth column value obj inf\n"); |
---|
1677 | for ( i=0;i<numberNodes2_;i++) { |
---|
1678 | statistics_[i]->print(lookup); |
---|
1679 | } |
---|
1680 | } |
---|
1681 | if (doStatistics>1) { |
---|
1682 | // Find last solution |
---|
1683 | int k; |
---|
1684 | for (k=numberNodes2_-1;k>=0;k--) { |
---|
1685 | if (statistics_[k]->endingObjective()!=COIN_DBL_MAX&& |
---|
1686 | !statistics_[k]->endingInfeasibility()) |
---|
1687 | break; |
---|
1688 | } |
---|
1689 | if (k>=0) { |
---|
1690 | int depth=statistics_[k]->depth(); |
---|
1691 | int * which = new int[depth+1]; |
---|
1692 | for (i=depth;i>=0;i--) { |
---|
1693 | which[i]=k; |
---|
1694 | k=statistics_[k]->parentNode(); |
---|
1695 | } |
---|
1696 | printf(" node parent depth column value obj inf\n"); |
---|
1697 | for (i=0;i<=depth;i++) { |
---|
1698 | statistics_[which[i]]->print(lookup); |
---|
1699 | } |
---|
1700 | delete [] which; |
---|
1701 | } |
---|
1702 | } |
---|
1703 | // now summary |
---|
1704 | int maxDepth=0; |
---|
1705 | double averageSolutionDepth=0.0; |
---|
1706 | int numberSolutions=0; |
---|
1707 | double averageCutoffDepth=0.0; |
---|
1708 | double averageSolvedDepth=0.0; |
---|
1709 | int numberCutoff=0; |
---|
1710 | int numberDown=0; |
---|
1711 | int numberFirstDown=0; |
---|
1712 | double averageInfDown=0.0; |
---|
1713 | double averageObjDown=0.0; |
---|
1714 | int numberCutoffDown=0; |
---|
1715 | int numberUp=0; |
---|
1716 | int numberFirstUp=0; |
---|
1717 | double averageInfUp=0.0; |
---|
1718 | double averageObjUp=0.0; |
---|
1719 | int numberCutoffUp=0; |
---|
1720 | double averageNumberIterations1=0.0; |
---|
1721 | double averageValue=0.0; |
---|
1722 | for ( i=0;i<numberNodes2_;i++) { |
---|
1723 | int depth = statistics_[i]->depth(); |
---|
1724 | int way = statistics_[i]->way(); |
---|
1725 | double value = statistics_[i]->value(); |
---|
1726 | double startingObjective = statistics_[i]->startingObjective(); |
---|
1727 | int startingInfeasibility = statistics_[i]->startingInfeasibility(); |
---|
1728 | double endingObjective = statistics_[i]->endingObjective(); |
---|
1729 | int endingInfeasibility = statistics_[i]->endingInfeasibility(); |
---|
1730 | maxDepth = CoinMax(depth,maxDepth); |
---|
1731 | // Only for completed |
---|
1732 | averageNumberIterations1 += statistics_[i]->numberIterations(); |
---|
1733 | averageValue += value; |
---|
1734 | if (endingObjective!=COIN_DBL_MAX&&!endingInfeasibility) { |
---|
1735 | numberSolutions++; |
---|
1736 | averageSolutionDepth += depth; |
---|
1737 | } |
---|
1738 | if (endingObjective==COIN_DBL_MAX) { |
---|
1739 | numberCutoff++; |
---|
1740 | averageCutoffDepth += depth; |
---|
1741 | if (way<0) { |
---|
1742 | numberDown++; |
---|
1743 | numberCutoffDown++; |
---|
1744 | if (way==-1) |
---|
1745 | numberFirstDown++; |
---|
1746 | } else { |
---|
1747 | numberUp++; |
---|
1748 | numberCutoffUp++; |
---|
1749 | if (way==1) |
---|
1750 | numberFirstUp++; |
---|
1751 | } |
---|
1752 | } else { |
---|
1753 | averageSolvedDepth += depth; |
---|
1754 | if (way<0) { |
---|
1755 | numberDown++; |
---|
1756 | averageInfDown += startingInfeasibility-endingInfeasibility; |
---|
1757 | averageObjDown += endingObjective-startingObjective; |
---|
1758 | if (way==-1) |
---|
1759 | numberFirstDown++; |
---|
1760 | } else { |
---|
1761 | numberUp++; |
---|
1762 | averageInfUp += startingInfeasibility-endingInfeasibility; |
---|
1763 | averageObjUp += endingObjective-startingObjective; |
---|
1764 | if (way==1) |
---|
1765 | numberFirstUp++; |
---|
1766 | } |
---|
1767 | } |
---|
1768 | } |
---|
1769 | // Now print |
---|
1770 | if (numberSolutions) |
---|
1771 | averageSolutionDepth /= (double) numberSolutions; |
---|
1772 | int numberSolved = numberNodes2_-numberCutoff; |
---|
1773 | double averageNumberIterations2=numberIterations_-averageNumberIterations1 |
---|
1774 | -numberIterationsAtContinuous; |
---|
1775 | if(numberCutoff) { |
---|
1776 | averageCutoffDepth /= (double) numberCutoff; |
---|
1777 | averageNumberIterations2 /= (double) numberCutoff; |
---|
1778 | } |
---|
1779 | if (numberNodes2_) |
---|
1780 | averageValue /= (double) numberNodes2_; |
---|
1781 | if (numberSolved) { |
---|
1782 | averageNumberIterations1 /= (double) numberSolved; |
---|
1783 | averageSolvedDepth /= (double) numberSolved; |
---|
1784 | } |
---|
1785 | printf("%d solution(s) were found (by branching) at an average depth of %g\n", |
---|
1786 | numberSolutions,averageSolutionDepth); |
---|
1787 | printf("average value of variable being branched on was %g\n", |
---|
1788 | averageValue); |
---|
1789 | printf("%d nodes were cutoff at an average depth of %g with iteration count of %g\n", |
---|
1790 | numberCutoff,averageCutoffDepth,averageNumberIterations2); |
---|
1791 | printf("%d nodes were solved at an average depth of %g with iteration count of %g\n", |
---|
1792 | numberSolved,averageSolvedDepth,averageNumberIterations1); |
---|
1793 | if (numberDown) { |
---|
1794 | averageInfDown /= (double) numberDown; |
---|
1795 | averageObjDown /= (double) numberDown; |
---|
1796 | } |
---|
1797 | printf("Down %d nodes (%d first, %d second) - %d cutoff, rest decrease numinf %g increase obj %g\n", |
---|
1798 | numberDown,numberFirstDown,numberDown-numberFirstDown,numberCutoffDown, |
---|
1799 | averageInfDown,averageObjDown); |
---|
1800 | if (numberUp) { |
---|
1801 | averageInfUp /= (double) numberUp; |
---|
1802 | averageObjUp /= (double) numberUp; |
---|
1803 | } |
---|
1804 | printf("Up %d nodes (%d first, %d second) - %d cutoff, rest decrease numinf %g increase obj %g\n", |
---|
1805 | numberUp,numberFirstUp,numberUp-numberFirstUp,numberCutoffUp, |
---|
1806 | averageInfUp,averageObjUp); |
---|
1807 | for ( i=0;i<numberNodes2_;i++) |
---|
1808 | delete statistics_[i]; |
---|
1809 | delete [] statistics_; |
---|
1810 | statistics_=NULL; |
---|
1811 | maximumStatistics_=0; |
---|
1812 | delete [] lookup; |
---|
1813 | } |
---|
1814 | /* |
---|
1815 | If we think we have a solution, restore and confirm it with a call to |
---|
1816 | setBestSolution(). We need to reset the cutoff value so as not to fathom |
---|
1817 | the solution on bounds. Note that calling setBestSolution( ..., true) |
---|
1818 | leaves the continuousSolver_ bounds vectors fixed at the solution value. |
---|
1819 | |
---|
1820 | Running resolve() here is a failsafe --- setBestSolution has already |
---|
1821 | reoptimised using the continuousSolver_. If for some reason we fail to |
---|
1822 | prove optimality, run the problem again after instructing the solver to |
---|
1823 | tell us more. |
---|
1824 | |
---|
1825 | If all looks good, replace solver_ with continuousSolver_, so that the |
---|
1826 | outside world will be able to obtain information about the solution using |
---|
1827 | public methods. |
---|
1828 | */ |
---|
1829 | if (bestSolution_&&solverCharacteristics_->solverType()<2) |
---|
1830 | { setCutoff(1.0e50) ; // As best solution should be worse than cutoff |
---|
1831 | phase_=5; |
---|
1832 | double increment = getDblParam(CbcModel::CbcCutoffIncrement) ; |
---|
1833 | bestObjective_ += 100.0*increment+1.0e-3; |
---|
1834 | setBestSolution(CBC_END_SOLUTION,bestObjective_,bestSolution_,true) ; |
---|
1835 | continuousSolver_->resolve() ; |
---|
1836 | if (!continuousSolver_->isProvenOptimal()) |
---|
1837 | { continuousSolver_->messageHandler()->setLogLevel(2) ; |
---|
1838 | continuousSolver_->initialSolve() ; } |
---|
1839 | delete solver_ ; |
---|
1840 | // above deletes solverCharacteristics_ |
---|
1841 | solverCharacteristics_ = NULL; |
---|
1842 | solver_ = continuousSolver_ ; |
---|
1843 | setPointers(solver_); |
---|
1844 | continuousSolver_ = NULL ; } |
---|
1845 | /* |
---|
1846 | Clean up dangling objects. continuousSolver_ may already be toast. |
---|
1847 | */ |
---|
1848 | delete lastws ; |
---|
1849 | delete [] whichGenerator_ ; |
---|
1850 | whichGenerator_=NULL; |
---|
1851 | delete [] lowerBefore ; |
---|
1852 | delete [] upperBefore ; |
---|
1853 | delete [] walkback_ ; |
---|
1854 | walkback_ = NULL ; |
---|
1855 | delete [] addedCuts_ ; |
---|
1856 | addedCuts_ = NULL ; |
---|
1857 | // Get rid of characteristics |
---|
1858 | solverCharacteristics_=NULL; |
---|
1859 | if (continuousSolver_) |
---|
1860 | { delete continuousSolver_ ; |
---|
1861 | continuousSolver_ = NULL ; } |
---|
1862 | /* |
---|
1863 | Destroy global cuts by replacing with an empty OsiCuts object. |
---|
1864 | */ |
---|
1865 | globalCuts_= OsiCuts() ; |
---|
1866 | if (strategy_&&strategy_->preProcessState()>0) { |
---|
1867 | // undo preprocessing |
---|
1868 | CglPreProcess * process = strategy_->process(); |
---|
1869 | assert (process); |
---|
1870 | int n = originalSolver->getNumCols(); |
---|
1871 | if (bestSolution_) { |
---|
1872 | delete [] bestSolution_; |
---|
1873 | bestSolution_ = new double [n]; |
---|
1874 | process->postProcess(*solver_); |
---|
1875 | } |
---|
1876 | strategy_->deletePreProcess(); |
---|
1877 | // Solution now back in originalSolver |
---|
1878 | delete solver_; |
---|
1879 | solver_=originalSolver; |
---|
1880 | if (bestSolution_) |
---|
1881 | memcpy(bestSolution_,solver_->getColSolution(),n*sizeof(double)); |
---|
1882 | // put back original objects if there were any |
---|
1883 | if (originalObject) { |
---|
1884 | int iColumn; |
---|
1885 | for (iColumn=0;iColumn<numberObjects_;iColumn++) |
---|
1886 | delete object_[iColumn]; |
---|
1887 | delete [] object_; |
---|
1888 | numberObjects_ = numberOriginalObjects; |
---|
1889 | object_=originalObject; |
---|
1890 | delete [] integerVariable_; |
---|
1891 | numberIntegers_=0; |
---|
1892 | for (iColumn=0;iColumn<n;iColumn++) { |
---|
1893 | if (solver_->isInteger(iColumn)) |
---|
1894 | numberIntegers_++; |
---|
1895 | } |
---|
1896 | integerVariable_ = new int[numberIntegers_]; |
---|
1897 | numberIntegers_=0; |
---|
1898 | for (iColumn=0;iColumn<n;iColumn++) { |
---|
1899 | if (solver_->isInteger(iColumn)) |
---|
1900 | integerVariable_[numberIntegers_++]=iColumn; |
---|
1901 | } |
---|
1902 | } |
---|
1903 | } |
---|
1904 | return ; } |
---|
1905 | |
---|
1906 | |
---|
1907 | // Solve the initial LP relaxation |
---|
1908 | void |
---|
1909 | CbcModel::initialSolve() |
---|
1910 | { |
---|
1911 | assert (solver_); |
---|
1912 | assert (!solverCharacteristics_); |
---|
1913 | OsiBabSolver * solverCharacteristics = dynamic_cast<OsiBabSolver *> (solver_->getAuxiliaryInfo()); |
---|
1914 | if (solverCharacteristics) { |
---|
1915 | solverCharacteristics_ = solverCharacteristics; |
---|
1916 | } else { |
---|
1917 | // replace in solver |
---|
1918 | OsiBabSolver defaultC; |
---|
1919 | solver_->setAuxiliaryInfo(&defaultC); |
---|
1920 | solverCharacteristics_ = dynamic_cast<OsiBabSolver *> (solver_->getAuxiliaryInfo()); |
---|
1921 | } |
---|
1922 | solverCharacteristics_->setSolver(solver_); |
---|
1923 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,NULL) ; |
---|
1924 | solver_->initialSolve(); |
---|
1925 | solver_->setHintParam(OsiDoInBranchAndCut,false,OsiHintDo,NULL) ; |
---|
1926 | // But set up so Jon Lee will be happy |
---|
1927 | status_=-1; |
---|
1928 | secondaryStatus_ = -1; |
---|
1929 | originalContinuousObjective_ = solver_->getObjValue()*solver_->getObjSense(); |
---|
1930 | delete [] continuousSolution_; |
---|
1931 | continuousSolution_ = CoinCopyOfArray(solver_->getColSolution(), |
---|
1932 | solver_->getNumCols()); |
---|
1933 | setPointers(solver_); |
---|
1934 | solverCharacteristics_ = NULL; |
---|
1935 | } |
---|
1936 | |
---|
1937 | /*! \brief Get an empty basis object |
---|
1938 | |
---|
1939 | Return an empty CoinWarmStartBasis object with the requested capacity, |
---|
1940 | appropriate for the current solver. The object is cloned from the object |
---|
1941 | cached as emptyWarmStart_. If there is no cached object, the routine |
---|
1942 | queries the solver for a warm start object, empties it, and caches the |
---|
1943 | result. |
---|
1944 | */ |
---|
1945 | |
---|
1946 | CoinWarmStartBasis *CbcModel::getEmptyBasis (int ns, int na) const |
---|
1947 | |
---|
1948 | { CoinWarmStartBasis *emptyBasis ; |
---|
1949 | /* |
---|
1950 | Acquire an empty basis object, if we don't yet have one. |
---|
1951 | */ |
---|
1952 | if (emptyWarmStart_ == 0) |
---|
1953 | { if (solver_ == 0) |
---|
1954 | { throw CoinError("Cannot construct basis without solver!", |
---|
1955 | "getEmptyBasis","CbcModel") ; } |
---|
1956 | emptyBasis = |
---|
1957 | dynamic_cast<CoinWarmStartBasis *>(solver_->getEmptyWarmStart()) ; |
---|
1958 | if (emptyBasis == 0) |
---|
1959 | { throw CoinError( |
---|
1960 | "Solver does not appear to use a basis-oriented warm start.", |
---|
1961 | "getEmptyBasis","CbcModel") ; } |
---|
1962 | emptyBasis->setSize(0,0) ; |
---|
1963 | emptyWarmStart_ = dynamic_cast<CoinWarmStart *>(emptyBasis) ; } |
---|
1964 | /* |
---|
1965 | Clone the empty basis object, resize it as requested, and return. |
---|
1966 | */ |
---|
1967 | emptyBasis = dynamic_cast<CoinWarmStartBasis *>(emptyWarmStart_->clone()) ; |
---|
1968 | assert(emptyBasis) ; |
---|
1969 | if (ns != 0 || na != 0) emptyBasis->setSize(ns,na) ; |
---|
1970 | |
---|
1971 | return (emptyBasis) ; } |
---|
1972 | |
---|
1973 | |
---|
1974 | /** Default Constructor |
---|
1975 | |
---|
1976 | Creates an empty model without an associated solver. |
---|
1977 | */ |
---|
1978 | CbcModel::CbcModel() |
---|
1979 | |
---|
1980 | : |
---|
1981 | solver_(NULL), |
---|
1982 | ourSolver_(true), |
---|
1983 | continuousSolver_(NULL), |
---|
1984 | referenceSolver_(NULL), |
---|
1985 | defaultHandler_(true), |
---|
1986 | emptyWarmStart_(NULL), |
---|
1987 | bestObjective_(COIN_DBL_MAX), |
---|
1988 | bestPossibleObjective_(COIN_DBL_MAX), |
---|
1989 | sumChangeObjective1_(0.0), |
---|
1990 | sumChangeObjective2_(0.0), |
---|
1991 | bestSolution_(NULL), |
---|
1992 | currentSolution_(NULL), |
---|
1993 | testSolution_(NULL), |
---|
1994 | minimumDrop_(1.0e-4), |
---|
1995 | numberSolutions_(0), |
---|
1996 | stateOfSearch_(0), |
---|
1997 | hotstartSolution_(NULL), |
---|
1998 | hotstartPriorities_(NULL), |
---|
1999 | numberHeuristicSolutions_(0), |
---|
2000 | numberNodes_(0), |
---|
2001 | numberNodes2_(0), |
---|
2002 | numberIterations_(0), |
---|
2003 | status_(-1), |
---|
2004 | secondaryStatus_(-1), |
---|
2005 | numberIntegers_(0), |
---|
2006 | numberRowsAtContinuous_(0), |
---|
2007 | maximumNumberCuts_(0), |
---|
2008 | phase_(0), |
---|
2009 | currentNumberCuts_(0), |
---|
2010 | maximumDepth_(0), |
---|
2011 | walkback_(NULL), |
---|
2012 | addedCuts_(NULL), |
---|
2013 | nextRowCut_(NULL), |
---|
2014 | currentNode_(NULL), |
---|
2015 | integerVariable_(NULL), |
---|
2016 | integerInfo_(NULL), |
---|
2017 | continuousSolution_(NULL), |
---|
2018 | usedInSolution_(NULL), |
---|
2019 | specialOptions_(0), |
---|
2020 | subTreeModel_(NULL), |
---|
2021 | numberStoppedSubTrees_(0), |
---|
2022 | presolve_(0), |
---|
2023 | numberStrong_(5), |
---|
2024 | numberBeforeTrust_(0), |
---|
2025 | numberPenalties_(20), |
---|
2026 | penaltyScaleFactor_(3.0), |
---|
2027 | numberAnalyzeIterations_(0), |
---|
2028 | analyzeResults_(NULL), |
---|
2029 | numberInfeasibleNodes_(0), |
---|
2030 | problemType_(0), |
---|
2031 | printFrequency_(0), |
---|
2032 | numberCutGenerators_(0), |
---|
2033 | generator_(NULL), |
---|
2034 | virginGenerator_(NULL), |
---|
2035 | numberHeuristics_(0), |
---|
2036 | heuristic_(NULL), |
---|
2037 | lastHeuristic_(NULL), |
---|
2038 | eventHandler_(0), |
---|
2039 | numberObjects_(0), |
---|
2040 | object_(NULL), |
---|
2041 | originalColumns_(NULL), |
---|
2042 | howOftenGlobalScan_(1), |
---|
2043 | numberGlobalViolations_(0), |
---|
2044 | continuousObjective_(COIN_DBL_MAX), |
---|
2045 | originalContinuousObjective_(COIN_DBL_MAX), |
---|
2046 | continuousInfeasibilities_(INT_MAX), |
---|
2047 | maximumCutPassesAtRoot_(20), |
---|
2048 | maximumCutPasses_(10), |
---|
2049 | currentPassNumber_(0), |
---|
2050 | maximumWhich_(1000), |
---|
2051 | whichGenerator_(NULL), |
---|
2052 | maximumStatistics_(0), |
---|
2053 | statistics_(NULL), |
---|
2054 | numberFixedAtRoot_(0), |
---|
2055 | numberFixedNow_(0), |
---|
2056 | stoppedOnGap_(false), |
---|
2057 | eventHappened_(false), |
---|
2058 | numberLongStrong_(0), |
---|
2059 | numberOldActiveCuts_(0), |
---|
2060 | numberNewCuts_(0), |
---|
2061 | sizeMiniTree_(0), |
---|
2062 | searchStrategy_(-1), |
---|
2063 | numberStrongIterations_(0), |
---|
2064 | resolveAfterTakeOffCuts_(true) |
---|
2065 | { |
---|
2066 | intParam_[CbcMaxNumNode] = 2147483647; |
---|
2067 | intParam_[CbcMaxNumSol] = 9999999; |
---|
2068 | intParam_[CbcFathomDiscipline] = 0; |
---|
2069 | |
---|
2070 | dblParam_[CbcIntegerTolerance] = 1e-6; |
---|
2071 | dblParam_[CbcInfeasibilityWeight] = 0.0; |
---|
2072 | dblParam_[CbcCutoffIncrement] = 1e-5; |
---|
2073 | dblParam_[CbcAllowableGap] = 1.0e-10; |
---|
2074 | dblParam_[CbcAllowableFractionGap] = 0.0; |
---|
2075 | dblParam_[CbcMaximumSeconds] = 1.0e100; |
---|
2076 | dblParam_[CbcCurrentCutoff] = 1.0e100; |
---|
2077 | dblParam_[CbcOptimizationDirection] = 1.0; |
---|
2078 | dblParam_[CbcCurrentObjectiveValue] = 1.0e100; |
---|
2079 | dblParam_[CbcCurrentMinimizationObjectiveValue] = 1.0e100; |
---|
2080 | dblParam_[CbcStartSeconds] = 0.0; |
---|
2081 | strongInfo_[0]=0; |
---|
2082 | strongInfo_[1]=0; |
---|
2083 | strongInfo_[2]=0; |
---|
2084 | solverCharacteristics_ = NULL; |
---|
2085 | nodeCompare_=new CbcCompareDefault();; |
---|
2086 | problemFeasibility_=new CbcFeasibilityBase(); |
---|
2087 | tree_= new CbcTree(); |
---|
2088 | branchingMethod_=NULL; |
---|
2089 | strategy_=NULL; |
---|
2090 | parentModel_=NULL; |
---|
2091 | cbcColLower_ = NULL; |
---|
2092 | cbcColUpper_ = NULL; |
---|
2093 | cbcRowLower_ = NULL; |
---|
2094 | cbcRowUpper_ = NULL; |
---|
2095 | cbcColSolution_ = NULL; |
---|
2096 | cbcRowPrice_ = NULL; |
---|
2097 | cbcReducedCost_ = NULL; |
---|
2098 | cbcRowActivity_ = NULL; |
---|
2099 | appData_=NULL; |
---|
2100 | handler_ = new CoinMessageHandler(); |
---|
2101 | handler_->setLogLevel(2); |
---|
2102 | messages_ = CbcMessage(); |
---|
2103 | eventHandler_ = new CbcEventHandler() ; |
---|
2104 | } |
---|
2105 | |
---|
2106 | /** Constructor from solver. |
---|
2107 | |
---|
2108 | Creates a model complete with a clone of the solver passed as a parameter. |
---|
2109 | */ |
---|
2110 | |
---|
2111 | CbcModel::CbcModel(const OsiSolverInterface &rhs) |
---|
2112 | : |
---|
2113 | continuousSolver_(NULL), |
---|
2114 | referenceSolver_(NULL), |
---|
2115 | defaultHandler_(true), |
---|
2116 | emptyWarmStart_(NULL), |
---|
2117 | bestObjective_(COIN_DBL_MAX), |
---|
2118 | bestPossibleObjective_(COIN_DBL_MAX), |
---|
2119 | sumChangeObjective1_(0.0), |
---|
2120 | sumChangeObjective2_(0.0), |
---|
2121 | minimumDrop_(1.0e-4), |
---|
2122 | numberSolutions_(0), |
---|
2123 | stateOfSearch_(0), |
---|
2124 | hotstartSolution_(NULL), |
---|
2125 | hotstartPriorities_(NULL), |
---|
2126 | numberHeuristicSolutions_(0), |
---|
2127 | numberNodes_(0), |
---|
2128 | numberNodes2_(0), |
---|
2129 | numberIterations_(0), |
---|
2130 | status_(-1), |
---|
2131 | secondaryStatus_(-1), |
---|
2132 | numberRowsAtContinuous_(0), |
---|
2133 | maximumNumberCuts_(0), |
---|
2134 | phase_(0), |
---|
2135 | currentNumberCuts_(0), |
---|
2136 | maximumDepth_(0), |
---|
2137 | walkback_(NULL), |
---|
2138 | addedCuts_(NULL), |
---|
2139 | nextRowCut_(NULL), |
---|
2140 | currentNode_(NULL), |
---|
2141 | integerInfo_(NULL), |
---|
2142 | specialOptions_(0), |
---|
2143 | subTreeModel_(NULL), |
---|
2144 | numberStoppedSubTrees_(0), |
---|
2145 | presolve_(0), |
---|
2146 | numberStrong_(5), |
---|
2147 | numberBeforeTrust_(0), |
---|
2148 | numberPenalties_(20), |
---|
2149 | penaltyScaleFactor_(3.0), |
---|
2150 | numberAnalyzeIterations_(0), |
---|
2151 | analyzeResults_(NULL), |
---|
2152 | numberInfeasibleNodes_(0), |
---|
2153 | problemType_(0), |
---|
2154 | printFrequency_(0), |
---|
2155 | numberCutGenerators_(0), |
---|
2156 | generator_(NULL), |
---|
2157 | virginGenerator_(NULL), |
---|
2158 | numberHeuristics_(0), |
---|
2159 | heuristic_(NULL), |
---|
2160 | lastHeuristic_(NULL), |
---|
2161 | eventHandler_(0), |
---|
2162 | numberObjects_(0), |
---|
2163 | object_(NULL), |
---|
2164 | originalColumns_(NULL), |
---|
2165 | howOftenGlobalScan_(1), |
---|
2166 | numberGlobalViolations_(0), |
---|
2167 | continuousObjective_(COIN_DBL_MAX), |
---|
2168 | originalContinuousObjective_(COIN_DBL_MAX), |
---|
2169 | continuousInfeasibilities_(INT_MAX), |
---|
2170 | maximumCutPassesAtRoot_(20), |
---|
2171 | maximumCutPasses_(10), |
---|
2172 | currentPassNumber_(0), |
---|
2173 | maximumWhich_(1000), |
---|
2174 | whichGenerator_(NULL), |
---|
2175 | maximumStatistics_(0), |
---|
2176 | statistics_(NULL), |
---|
2177 | numberFixedAtRoot_(0), |
---|
2178 | numberFixedNow_(0), |
---|
2179 | stoppedOnGap_(false), |
---|
2180 | eventHappened_(false), |
---|
2181 | numberLongStrong_(0), |
---|
2182 | numberOldActiveCuts_(0), |
---|
2183 | numberNewCuts_(0), |
---|
2184 | sizeMiniTree_(0), |
---|
2185 | searchStrategy_(-1), |
---|
2186 | numberStrongIterations_(0), |
---|
2187 | resolveAfterTakeOffCuts_(true) |
---|
2188 | { |
---|
2189 | intParam_[CbcMaxNumNode] = 2147483647; |
---|
2190 | intParam_[CbcMaxNumSol] = 9999999; |
---|
2191 | intParam_[CbcFathomDiscipline] = 0; |
---|
2192 | |
---|
2193 | dblParam_[CbcIntegerTolerance] = 1e-6; |
---|
2194 | dblParam_[CbcInfeasibilityWeight] = 0.0; |
---|
2195 | dblParam_[CbcCutoffIncrement] = 1e-5; |
---|
2196 | dblParam_[CbcAllowableGap] = 1.0e-10; |
---|
2197 | dblParam_[CbcAllowableFractionGap] = 0.0; |
---|
2198 | dblParam_[CbcMaximumSeconds] = 1.0e100; |
---|
2199 | dblParam_[CbcCurrentCutoff] = 1.0e100; |
---|
2200 | dblParam_[CbcOptimizationDirection] = 1.0; |
---|
2201 | dblParam_[CbcCurrentObjectiveValue] = 1.0e100; |
---|
2202 | dblParam_[CbcCurrentMinimizationObjectiveValue] = 1.0e100; |
---|
2203 | dblParam_[CbcStartSeconds] = 0.0; |
---|
2204 | strongInfo_[0]=0; |
---|
2205 | strongInfo_[1]=0; |
---|
2206 | strongInfo_[2]=0; |
---|
2207 | solverCharacteristics_ = NULL; |
---|
2208 | |
---|
2209 | nodeCompare_=new CbcCompareDefault();; |
---|
2210 | problemFeasibility_=new CbcFeasibilityBase(); |
---|
2211 | tree_= new CbcTree(); |
---|
2212 | branchingMethod_=NULL; |
---|
2213 | strategy_=NULL; |
---|
2214 | parentModel_=NULL; |
---|
2215 | appData_=NULL; |
---|
2216 | handler_ = new CoinMessageHandler(); |
---|
2217 | handler_->setLogLevel(2); |
---|
2218 | messages_ = CbcMessage(); |
---|
2219 | eventHandler_ = new CbcEventHandler() ; |
---|
2220 | solver_ = rhs.clone(); |
---|
2221 | referenceSolver_ = solver_->clone(); |
---|
2222 | ourSolver_ = true ; |
---|
2223 | cbcColLower_ = NULL; |
---|
2224 | cbcColUpper_ = NULL; |
---|
2225 | cbcRowLower_ = NULL; |
---|
2226 | cbcRowUpper_ = NULL; |
---|
2227 | cbcColSolution_ = NULL; |
---|
2228 | cbcRowPrice_ = NULL; |
---|
2229 | cbcReducedCost_ = NULL; |
---|
2230 | cbcRowActivity_ = NULL; |
---|
2231 | |
---|
2232 | // Initialize solution and integer variable vectors |
---|
2233 | bestSolution_ = NULL; // to say no solution found |
---|
2234 | numberIntegers_=0; |
---|
2235 | int numberColumns = solver_->getNumCols(); |
---|
2236 | int iColumn; |
---|
2237 | if (numberColumns) { |
---|
2238 | // Space for current solution |
---|
2239 | currentSolution_ = new double[numberColumns]; |
---|
2240 | continuousSolution_ = new double[numberColumns]; |
---|
2241 | usedInSolution_ = new int[numberColumns]; |
---|
2242 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
2243 | if( solver_->isInteger(iColumn)) |
---|
2244 | numberIntegers_++; |
---|
2245 | } |
---|
2246 | } else { |
---|
2247 | // empty model |
---|
2248 | currentSolution_=NULL; |
---|
2249 | continuousSolution_=NULL; |
---|
2250 | usedInSolution_=NULL; |
---|
2251 | } |
---|
2252 | testSolution_=currentSolution_; |
---|
2253 | if (numberIntegers_) { |
---|
2254 | integerVariable_ = new int [numberIntegers_]; |
---|
2255 | numberIntegers_=0; |
---|
2256 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
2257 | if( solver_->isInteger(iColumn)) |
---|
2258 | integerVariable_[numberIntegers_++]=iColumn; |
---|
2259 | } |
---|
2260 | } else { |
---|
2261 | integerVariable_ = NULL; |
---|
2262 | } |
---|
2263 | } |
---|
2264 | |
---|
2265 | /* |
---|
2266 | Assign a solver to the model (model assumes ownership) |
---|
2267 | |
---|
2268 | The integer variable vector is initialized if it's not already present. |
---|
2269 | If deleteSolver then current solver deleted (if model owned) |
---|
2270 | |
---|
2271 | Assuming ownership matches usage in OsiSolverInterface |
---|
2272 | (cf. assignProblem, loadProblem). |
---|
2273 | |
---|
2274 | TODO: What to do about solver parameters? A simple copy likely won't do it, |
---|
2275 | because the SI must push the settings into the underlying solver. In |
---|
2276 | the context of switching solvers in cbc, this means that command line |
---|
2277 | settings will get lost. Stash the command line somewhere and reread it |
---|
2278 | here, maybe? |
---|
2279 | |
---|
2280 | TODO: More generally, how much state should be transferred from the old |
---|
2281 | solver to the new solver? Best perhaps to see how usage develops. |
---|
2282 | What's done here mimics the CbcModel(OsiSolverInterface) constructor. |
---|
2283 | */ |
---|
2284 | void |
---|
2285 | CbcModel::assignSolver(OsiSolverInterface *&solver, bool deleteSolver) |
---|
2286 | |
---|
2287 | { |
---|
2288 | // Keep the current message level for solver (if solver exists) |
---|
2289 | if (solver_) |
---|
2290 | solver->messageHandler()->setLogLevel(solver_->messageHandler()->logLevel()) ; |
---|
2291 | |
---|
2292 | if (ourSolver_&&deleteSolver) delete solver_ ; |
---|
2293 | solver_ = solver; |
---|
2294 | solver = NULL ; |
---|
2295 | ourSolver_ = true ; |
---|
2296 | /* |
---|
2297 | Basis information is solver-specific. |
---|
2298 | */ |
---|
2299 | if (emptyWarmStart_) |
---|
2300 | { delete emptyWarmStart_ ; |
---|
2301 | emptyWarmStart_ = 0 ; } |
---|
2302 | /* |
---|
2303 | Initialize integer variable vector. |
---|
2304 | */ |
---|
2305 | numberIntegers_=0; |
---|
2306 | int numberColumns = solver_->getNumCols(); |
---|
2307 | int iColumn; |
---|
2308 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
2309 | if( solver_->isInteger(iColumn)) |
---|
2310 | numberIntegers_++; |
---|
2311 | } |
---|
2312 | if (numberIntegers_) { |
---|
2313 | delete [] integerVariable_; |
---|
2314 | integerVariable_ = new int [numberIntegers_]; |
---|
2315 | numberIntegers_=0; |
---|
2316 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
2317 | if( solver_->isInteger(iColumn)) |
---|
2318 | integerVariable_[numberIntegers_++]=iColumn; |
---|
2319 | } |
---|
2320 | } else { |
---|
2321 | integerVariable_ = NULL; |
---|
2322 | } |
---|
2323 | |
---|
2324 | return ; |
---|
2325 | } |
---|
2326 | |
---|
2327 | // Copy constructor. |
---|
2328 | |
---|
2329 | CbcModel::CbcModel(const CbcModel & rhs, bool noTree) |
---|
2330 | : |
---|
2331 | continuousSolver_(NULL), |
---|
2332 | referenceSolver_(NULL), |
---|
2333 | defaultHandler_(rhs.defaultHandler_), |
---|
2334 | emptyWarmStart_(NULL), |
---|
2335 | bestObjective_(rhs.bestObjective_), |
---|
2336 | bestPossibleObjective_(rhs.bestPossibleObjective_), |
---|
2337 | sumChangeObjective1_(rhs.sumChangeObjective1_), |
---|
2338 | sumChangeObjective2_(rhs.sumChangeObjective2_), |
---|
2339 | minimumDrop_(rhs.minimumDrop_), |
---|
2340 | numberSolutions_(rhs.numberSolutions_), |
---|
2341 | stateOfSearch_(rhs.stateOfSearch_), |
---|
2342 | numberHeuristicSolutions_(rhs.numberHeuristicSolutions_), |
---|
2343 | numberNodes_(rhs.numberNodes_), |
---|
2344 | numberNodes2_(rhs.numberNodes2_), |
---|
2345 | numberIterations_(rhs.numberIterations_), |
---|
2346 | status_(rhs.status_), |
---|
2347 | secondaryStatus_(rhs.secondaryStatus_), |
---|
2348 | specialOptions_(rhs.specialOptions_), |
---|
2349 | subTreeModel_(rhs.subTreeModel_), |
---|
2350 | numberStoppedSubTrees_(rhs.numberStoppedSubTrees_), |
---|
2351 | presolve_(rhs.presolve_), |
---|
2352 | numberStrong_(rhs.numberStrong_), |
---|
2353 | numberBeforeTrust_(rhs.numberBeforeTrust_), |
---|
2354 | numberPenalties_(rhs.numberPenalties_), |
---|
2355 | penaltyScaleFactor_(rhs.penaltyScaleFactor_), |
---|
2356 | numberAnalyzeIterations_(rhs.numberAnalyzeIterations_), |
---|
2357 | analyzeResults_(NULL), |
---|
2358 | numberInfeasibleNodes_(rhs.numberInfeasibleNodes_), |
---|
2359 | problemType_(rhs.problemType_), |
---|
2360 | printFrequency_(rhs.printFrequency_), |
---|
2361 | howOftenGlobalScan_(rhs.howOftenGlobalScan_), |
---|
2362 | numberGlobalViolations_(rhs.numberGlobalViolations_), |
---|
2363 | continuousObjective_(rhs.continuousObjective_), |
---|
2364 | originalContinuousObjective_(rhs.originalContinuousObjective_), |
---|
2365 | continuousInfeasibilities_(rhs.continuousInfeasibilities_), |
---|
2366 | maximumCutPassesAtRoot_(rhs.maximumCutPassesAtRoot_), |
---|
2367 | maximumCutPasses_( rhs.maximumCutPasses_), |
---|
2368 | currentPassNumber_(rhs.currentPassNumber_), |
---|
2369 | maximumWhich_(rhs.maximumWhich_), |
---|
2370 | whichGenerator_(NULL), |
---|
2371 | maximumStatistics_(0), |
---|
2372 | statistics_(NULL), |
---|
2373 | numberFixedAtRoot_(rhs.numberFixedAtRoot_), |
---|
2374 | numberFixedNow_(rhs.numberFixedNow_), |
---|
2375 | stoppedOnGap_(rhs.stoppedOnGap_), |
---|
2376 | eventHappened_(rhs.eventHappened_), |
---|
2377 | numberLongStrong_(rhs.numberLongStrong_), |
---|
2378 | numberOldActiveCuts_(rhs.numberOldActiveCuts_), |
---|
2379 | numberNewCuts_(rhs.numberNewCuts_), |
---|
2380 | sizeMiniTree_(rhs.sizeMiniTree_), |
---|
2381 | searchStrategy_(rhs.searchStrategy_), |
---|
2382 | numberStrongIterations_(rhs.numberStrongIterations_), |
---|
2383 | resolveAfterTakeOffCuts_(rhs.resolveAfterTakeOffCuts_) |
---|
2384 | { |
---|
2385 | intParam_[CbcMaxNumNode] = rhs.intParam_[CbcMaxNumNode]; |
---|
2386 | intParam_[CbcMaxNumSol] = rhs.intParam_[CbcMaxNumSol]; |
---|
2387 | intParam_[CbcFathomDiscipline] = rhs.intParam_[CbcFathomDiscipline]; |
---|
2388 | dblParam_[CbcIntegerTolerance] = rhs.dblParam_[CbcIntegerTolerance]; |
---|
2389 | dblParam_[CbcInfeasibilityWeight] = rhs.dblParam_[CbcInfeasibilityWeight]; |
---|
2390 | dblParam_[CbcCutoffIncrement] = rhs.dblParam_[CbcCutoffIncrement]; |
---|
2391 | dblParam_[CbcAllowableGap] = rhs.dblParam_[CbcAllowableGap]; |
---|
2392 | dblParam_[CbcAllowableFractionGap] = rhs.dblParam_[CbcAllowableFractionGap]; |
---|
2393 | dblParam_[CbcMaximumSeconds] = rhs.dblParam_[CbcMaximumSeconds]; |
---|
2394 | dblParam_[CbcCurrentCutoff] = rhs.dblParam_[CbcCurrentCutoff]; |
---|
2395 | dblParam_[CbcOptimizationDirection] = rhs.dblParam_[CbcOptimizationDirection]; |
---|
2396 | dblParam_[CbcCurrentObjectiveValue] = rhs.dblParam_[CbcCurrentObjectiveValue]; |
---|
2397 | dblParam_[CbcCurrentMinimizationObjectiveValue] = rhs.dblParam_[CbcCurrentMinimizationObjectiveValue]; |
---|
2398 | dblParam_[CbcStartSeconds] = dblParam_[CbcStartSeconds]; // will be overwritten hopefully |
---|
2399 | strongInfo_[0]=rhs.strongInfo_[0]; |
---|
2400 | strongInfo_[1]=rhs.strongInfo_[1]; |
---|
2401 | strongInfo_[2]=rhs.strongInfo_[2]; |
---|
2402 | solverCharacteristics_ = NULL; |
---|
2403 | if (rhs.emptyWarmStart_) emptyWarmStart_ = rhs.emptyWarmStart_->clone() ; |
---|
2404 | if (defaultHandler_) { |
---|
2405 | handler_ = new CoinMessageHandler(); |
---|
2406 | handler_->setLogLevel(2); |
---|
2407 | } else { |
---|
2408 | handler_ = rhs.handler_; |
---|
2409 | } |
---|
2410 | messageHandler()->setLogLevel(rhs.messageHandler()->logLevel()); |
---|
2411 | numberCutGenerators_ = rhs.numberCutGenerators_; |
---|
2412 | if (numberCutGenerators_) { |
---|
2413 | generator_ = new CbcCutGenerator * [numberCutGenerators_]; |
---|
2414 | virginGenerator_ = new CbcCutGenerator * [numberCutGenerators_]; |
---|
2415 | int i; |
---|
2416 | for (i=0;i<numberCutGenerators_;i++) { |
---|
2417 | generator_[i]=new CbcCutGenerator(*rhs.generator_[i]); |
---|
2418 | virginGenerator_[i]=new CbcCutGenerator(*rhs.virginGenerator_[i]); |
---|
2419 | } |
---|
2420 | } else { |
---|
2421 | generator_=NULL; |
---|
2422 | virginGenerator_=NULL; |
---|
2423 | } |
---|
2424 | if (!noTree) |
---|
2425 | globalCuts_ = rhs.globalCuts_; |
---|
2426 | numberHeuristics_ = rhs.numberHeuristics_; |
---|
2427 | if (numberHeuristics_) { |
---|
2428 | heuristic_ = new CbcHeuristic * [numberHeuristics_]; |
---|
2429 | int i; |
---|
2430 | for (i=0;i<numberHeuristics_;i++) { |
---|
2431 | heuristic_[i]=rhs.heuristic_[i]->clone(); |
---|
2432 | } |
---|
2433 | } else { |
---|
2434 | heuristic_=NULL; |
---|
2435 | } |
---|
2436 | lastHeuristic_ = NULL; |
---|
2437 | if (rhs.eventHandler_) |
---|
2438 | { eventHandler_ = new CbcEventHandler(*rhs.eventHandler_) ; } |
---|
2439 | else |
---|
2440 | { eventHandler_ = NULL ; } |
---|
2441 | numberObjects_=rhs.numberObjects_; |
---|
2442 | if (numberObjects_) { |
---|
2443 | object_ = new CbcObject * [numberObjects_]; |
---|
2444 | int i; |
---|
2445 | for (i=0;i<numberObjects_;i++) |
---|
2446 | object_[i]=(rhs.object_[i])->clone(); |
---|
2447 | } else { |
---|
2448 | object_=NULL; |
---|
2449 | } |
---|
2450 | if (rhs.referenceSolver_) |
---|
2451 | referenceSolver_ = rhs.referenceSolver_->clone(); |
---|
2452 | else |
---|
2453 | referenceSolver_=NULL; |
---|
2454 | if (!noTree||!rhs.continuousSolver_) |
---|
2455 | solver_ = rhs.solver_->clone(); |
---|
2456 | else |
---|
2457 | solver_ = rhs.continuousSolver_->clone(); |
---|
2458 | if (rhs.originalColumns_) { |
---|
2459 | int numberColumns = solver_->getNumCols(); |
---|
2460 | originalColumns_= new int [numberColumns]; |
---|
2461 | memcpy(originalColumns_,rhs.originalColumns_,numberColumns*sizeof(int)); |
---|
2462 | } else { |
---|
2463 | originalColumns_=NULL; |
---|
2464 | } |
---|
2465 | if (maximumWhich_&&rhs.whichGenerator_) |
---|
2466 | whichGenerator_ = CoinCopyOfArray(rhs.whichGenerator_,maximumWhich_); |
---|
2467 | nodeCompare_=rhs.nodeCompare_->clone(); |
---|
2468 | problemFeasibility_=rhs.problemFeasibility_->clone(); |
---|
2469 | tree_= rhs.tree_->clone(); |
---|
2470 | branchingMethod_=rhs.branchingMethod_; |
---|
2471 | cbcColLower_ = NULL; |
---|
2472 | cbcColUpper_ = NULL; |
---|
2473 | cbcRowLower_ = NULL; |
---|
2474 | cbcRowUpper_ = NULL; |
---|
2475 | cbcColSolution_ = NULL; |
---|
2476 | cbcRowPrice_ = NULL; |
---|
2477 | cbcReducedCost_ = NULL; |
---|
2478 | cbcRowActivity_ = NULL; |
---|
2479 | if (rhs.strategy_) |
---|
2480 | strategy_=rhs.strategy_->clone(); |
---|
2481 | else |
---|
2482 | strategy_=NULL; |
---|
2483 | parentModel_=rhs.parentModel_; |
---|
2484 | appData_=rhs.appData_; |
---|
2485 | messages_ = rhs.messages_; |
---|
2486 | ourSolver_ = true ; |
---|
2487 | messageHandler()->setLogLevel(rhs.messageHandler()->logLevel()); |
---|
2488 | numberIntegers_=rhs.numberIntegers_; |
---|
2489 | if (numberIntegers_) { |
---|
2490 | integerVariable_ = new int [numberIntegers_]; |
---|
2491 | memcpy(integerVariable_,rhs.integerVariable_,numberIntegers_*sizeof(int)); |
---|
2492 | integerInfo_ = CoinCopyOfArray(rhs.integerInfo_,solver_->getNumCols()); |
---|
2493 | } else { |
---|
2494 | integerVariable_ = NULL; |
---|
2495 | integerInfo_=NULL; |
---|
2496 | } |
---|
2497 | if (rhs.hotstartSolution_) { |
---|
2498 | int numberColumns = solver_->getNumCols(); |
---|
2499 | hotstartSolution_ = CoinCopyOfArray(rhs.hotstartSolution_,numberColumns); |
---|
2500 | hotstartPriorities_ = CoinCopyOfArray(rhs.hotstartPriorities_,numberColumns); |
---|
2501 | } else { |
---|
2502 | hotstartSolution_ = NULL; |
---|
2503 | hotstartPriorities_ =NULL; |
---|
2504 | } |
---|
2505 | if (rhs.bestSolution_&&!noTree) { |
---|
2506 | int numberColumns = solver_->getNumCols(); |
---|
2507 | bestSolution_ = new double[numberColumns]; |
---|
2508 | memcpy(bestSolution_,rhs.bestSolution_,numberColumns*sizeof(double)); |
---|
2509 | } else { |
---|
2510 | bestSolution_=NULL; |
---|
2511 | } |
---|
2512 | if (!noTree) { |
---|
2513 | int numberColumns = solver_->getNumCols(); |
---|
2514 | currentSolution_ = CoinCopyOfArray(rhs.currentSolution_,numberColumns); |
---|
2515 | continuousSolution_ = CoinCopyOfArray(rhs.continuousSolution_,numberColumns); |
---|
2516 | usedInSolution_ = CoinCopyOfArray(rhs.usedInSolution_,numberColumns); |
---|
2517 | } else { |
---|
2518 | currentSolution_=NULL; |
---|
2519 | continuousSolution_=NULL; |
---|
2520 | usedInSolution_=NULL; |
---|
2521 | } |
---|
2522 | testSolution_=currentSolution_; |
---|
2523 | numberRowsAtContinuous_ = rhs.numberRowsAtContinuous_; |
---|
2524 | maximumNumberCuts_=rhs.maximumNumberCuts_; |
---|
2525 | phase_ = rhs.phase_; |
---|
2526 | currentNumberCuts_=rhs.currentNumberCuts_; |
---|
2527 | maximumDepth_= rhs.maximumDepth_; |
---|
2528 | if (noTree) { |
---|
2529 | bestObjective_ = COIN_DBL_MAX; |
---|
2530 | numberSolutions_ =0; |
---|
2531 | stateOfSearch_= 0; |
---|
2532 | numberHeuristicSolutions_=0; |
---|
2533 | numberNodes_=0; |
---|
2534 | numberNodes2_=0; |
---|
2535 | numberIterations_=0; |
---|
2536 | status_=0; |
---|
2537 | subTreeModel_=NULL; |
---|
2538 | numberStoppedSubTrees_=0; |
---|
2539 | continuousObjective_=COIN_DBL_MAX; |
---|
2540 | originalContinuousObjective_=COIN_DBL_MAX; |
---|
2541 | continuousInfeasibilities_=INT_MAX; |
---|
2542 | maximumNumberCuts_=0; |
---|
2543 | tree_->cleanTree(this,-COIN_DBL_MAX,bestPossibleObjective_); |
---|
2544 | bestPossibleObjective_ = COIN_DBL_MAX; |
---|
2545 | } |
---|
2546 | // These are only used as temporary arrays so need not be filled |
---|
2547 | if (maximumNumberCuts_) { |
---|
2548 | addedCuts_ = new CbcCountRowCut * [maximumNumberCuts_]; |
---|
2549 | } else { |
---|
2550 | addedCuts_ = NULL; |
---|
2551 | } |
---|
2552 | nextRowCut_ = NULL; |
---|
2553 | currentNode_ = NULL; |
---|
2554 | if (maximumDepth_) |
---|
2555 | walkback_ = new CbcNodeInfo * [maximumDepth_]; |
---|
2556 | else |
---|
2557 | walkback_ = NULL; |
---|
2558 | synchronizeModel(); |
---|
2559 | } |
---|
2560 | |
---|
2561 | // Assignment operator |
---|
2562 | CbcModel & |
---|
2563 | CbcModel::operator=(const CbcModel& rhs) |
---|
2564 | { |
---|
2565 | if (this!=&rhs) { |
---|
2566 | gutsOfDestructor(); |
---|
2567 | if (defaultHandler_) |
---|
2568 | { delete handler_; |
---|
2569 | handler_ = NULL; } |
---|
2570 | defaultHandler_ = rhs.defaultHandler_; |
---|
2571 | if (defaultHandler_) |
---|
2572 | { handler_ = new CoinMessageHandler(); |
---|
2573 | handler_->setLogLevel(2); } |
---|
2574 | else |
---|
2575 | { handler_ = rhs.handler_; } |
---|
2576 | messages_ = rhs.messages_; |
---|
2577 | messageHandler()->setLogLevel(rhs.messageHandler()->logLevel()); |
---|
2578 | |
---|
2579 | delete solver_; |
---|
2580 | if (rhs.solver_) |
---|
2581 | { solver_ = rhs.solver_->clone() ; } |
---|
2582 | else |
---|
2583 | { solver_ = 0 ; } |
---|
2584 | ourSolver_ = true ; |
---|
2585 | delete continuousSolver_ ; |
---|
2586 | if (rhs.continuousSolver_) |
---|
2587 | { solver_ = rhs.continuousSolver_->clone() ; } |
---|
2588 | else |
---|
2589 | { continuousSolver_ = 0 ; } |
---|
2590 | |
---|
2591 | if (rhs.referenceSolver_) |
---|
2592 | { solver_ = rhs.referenceSolver_->clone() ; } |
---|
2593 | else |
---|
2594 | { referenceSolver_ = NULL ; } |
---|
2595 | |
---|
2596 | delete emptyWarmStart_ ; |
---|
2597 | if (rhs.emptyWarmStart_) |
---|
2598 | { emptyWarmStart_ = rhs.emptyWarmStart_->clone() ; } |
---|
2599 | else |
---|
2600 | { emptyWarmStart_ = 0 ; } |
---|
2601 | |
---|
2602 | bestObjective_ = rhs.bestObjective_; |
---|
2603 | bestPossibleObjective_=rhs.bestPossibleObjective_; |
---|
2604 | sumChangeObjective1_=rhs.sumChangeObjective1_; |
---|
2605 | sumChangeObjective2_=rhs.sumChangeObjective2_; |
---|
2606 | delete [] bestSolution_; |
---|
2607 | if (rhs.bestSolution_) { |
---|
2608 | int numberColumns = rhs.getNumCols(); |
---|
2609 | bestSolution_ = new double[numberColumns]; |
---|
2610 | memcpy(bestSolution_,rhs.bestSolution_,numberColumns*sizeof(double)); |
---|
2611 | } else { |
---|
2612 | bestSolution_=NULL; |
---|
2613 | } |
---|
2614 | int numberColumns = solver_->getNumCols(); |
---|
2615 | currentSolution_ = CoinCopyOfArray(rhs.currentSolution_,numberColumns); |
---|
2616 | continuousSolution_ = CoinCopyOfArray(rhs.continuousSolution_,numberColumns); |
---|
2617 | usedInSolution_ = CoinCopyOfArray(rhs.usedInSolution_,numberColumns); |
---|
2618 | testSolution_=currentSolution_; |
---|
2619 | minimumDrop_ = rhs.minimumDrop_; |
---|
2620 | numberSolutions_=rhs.numberSolutions_; |
---|
2621 | stateOfSearch_= rhs.stateOfSearch_; |
---|
2622 | numberHeuristicSolutions_=rhs.numberHeuristicSolutions_; |
---|
2623 | numberNodes_ = rhs.numberNodes_; |
---|
2624 | numberNodes2_ = rhs.numberNodes2_; |
---|
2625 | numberIterations_ = rhs.numberIterations_; |
---|
2626 | status_ = rhs.status_; |
---|
2627 | secondaryStatus_ = rhs.secondaryStatus_; |
---|
2628 | specialOptions_ = rhs.specialOptions_; |
---|
2629 | subTreeModel_ = rhs.subTreeModel_; |
---|
2630 | numberStoppedSubTrees_ = rhs.numberStoppedSubTrees_; |
---|
2631 | presolve_ = rhs.presolve_; |
---|
2632 | numberStrong_ = rhs.numberStrong_; |
---|
2633 | numberBeforeTrust_ = rhs.numberBeforeTrust_; |
---|
2634 | numberPenalties_ = rhs.numberPenalties_; |
---|
2635 | penaltyScaleFactor_ = rhs.penaltyScaleFactor_; |
---|
2636 | numberAnalyzeIterations_ = rhs.numberAnalyzeIterations_; |
---|
2637 | delete [] analyzeResults_; |
---|
2638 | analyzeResults_ = NULL; |
---|
2639 | numberInfeasibleNodes_ = rhs.numberInfeasibleNodes_; |
---|
2640 | problemType_ = rhs.problemType_; |
---|
2641 | printFrequency_ = rhs.printFrequency_; |
---|
2642 | howOftenGlobalScan_=rhs.howOftenGlobalScan_; |
---|
2643 | numberGlobalViolations_=rhs.numberGlobalViolations_; |
---|
2644 | continuousObjective_=rhs.continuousObjective_; |
---|
2645 | originalContinuousObjective_ = rhs.originalContinuousObjective_; |
---|
2646 | continuousInfeasibilities_ = rhs.continuousInfeasibilities_; |
---|
2647 | maximumCutPassesAtRoot_ = rhs.maximumCutPassesAtRoot_; |
---|
2648 | maximumCutPasses_ = rhs.maximumCutPasses_; |
---|
2649 | currentPassNumber_ = rhs.currentPassNumber_; |
---|
2650 | intParam_[CbcMaxNumNode] = rhs.intParam_[CbcMaxNumNode]; |
---|
2651 | intParam_[CbcMaxNumSol] = rhs.intParam_[CbcMaxNumSol]; |
---|
2652 | intParam_[CbcFathomDiscipline] = rhs.intParam_[CbcFathomDiscipline]; |
---|
2653 | dblParam_[CbcIntegerTolerance] = rhs.dblParam_[CbcIntegerTolerance]; |
---|
2654 | dblParam_[CbcInfeasibilityWeight] = rhs.dblParam_[CbcInfeasibilityWeight]; |
---|
2655 | dblParam_[CbcCutoffIncrement] = rhs.dblParam_[CbcCutoffIncrement]; |
---|
2656 | dblParam_[CbcAllowableGap] = rhs.dblParam_[CbcAllowableGap]; |
---|
2657 | dblParam_[CbcAllowableFractionGap] = rhs.dblParam_[CbcAllowableFractionGap]; |
---|
2658 | dblParam_[CbcMaximumSeconds] = rhs.dblParam_[CbcMaximumSeconds]; |
---|
2659 | dblParam_[CbcCurrentCutoff] = rhs.dblParam_[CbcCurrentCutoff]; |
---|
2660 | dblParam_[CbcOptimizationDirection] = rhs.dblParam_[CbcOptimizationDirection]; |
---|
2661 | dblParam_[CbcCurrentObjectiveValue] = rhs.dblParam_[CbcCurrentObjectiveValue]; |
---|
2662 | dblParam_[CbcCurrentMinimizationObjectiveValue] = rhs.dblParam_[CbcCurrentMinimizationObjectiveValue]; |
---|
2663 | dblParam_[CbcStartSeconds] = dblParam_[CbcStartSeconds]; // will be overwritten hopefully |
---|
2664 | globalCuts_ = rhs.globalCuts_; |
---|
2665 | int i; |
---|
2666 | for (i=0;i<numberCutGenerators_;i++) { |
---|
2667 | delete generator_[i]; |
---|
2668 | delete virginGenerator_[i]; |
---|
2669 | } |
---|
2670 | delete [] generator_; |
---|
2671 | delete [] virginGenerator_; |
---|
2672 | delete [] heuristic_; |
---|
2673 | maximumWhich_ = rhs.maximumWhich_; |
---|
2674 | delete [] whichGenerator_; |
---|
2675 | whichGenerator_ = NULL; |
---|
2676 | if (maximumWhich_&&rhs.whichGenerator_) |
---|
2677 | whichGenerator_ = CoinCopyOfArray(rhs.whichGenerator_,maximumWhich_); |
---|
2678 | for (i=0;i<maximumStatistics_;i++) |
---|
2679 | delete statistics_[i]; |
---|
2680 | delete [] statistics_; |
---|
2681 | maximumStatistics_ = 0; |
---|
2682 | statistics_ = NULL; |
---|
2683 | numberFixedAtRoot_ = rhs.numberFixedAtRoot_; |
---|
2684 | numberFixedNow_ = rhs.numberFixedNow_; |
---|
2685 | stoppedOnGap_ = rhs.stoppedOnGap_; |
---|
2686 | eventHappened_ = rhs.eventHappened_; |
---|
2687 | numberLongStrong_ = rhs.numberLongStrong_; |
---|
2688 | numberOldActiveCuts_ = rhs.numberOldActiveCuts_; |
---|
2689 | numberNewCuts_ = rhs.numberNewCuts_; |
---|
2690 | resolveAfterTakeOffCuts_=rhs.resolveAfterTakeOffCuts_; |
---|
2691 | sizeMiniTree_ = rhs.sizeMiniTree_; |
---|
2692 | searchStrategy_ = rhs.searchStrategy_; |
---|
2693 | numberStrongIterations_ = rhs.numberStrongIterations_; |
---|
2694 | strongInfo_[0]=rhs.strongInfo_[0]; |
---|
2695 | strongInfo_[1]=rhs.strongInfo_[1]; |
---|
2696 | strongInfo_[2]=rhs.strongInfo_[2]; |
---|
2697 | solverCharacteristics_ = NULL; |
---|
2698 | lastHeuristic_ = NULL; |
---|
2699 | numberCutGenerators_ = rhs.numberCutGenerators_; |
---|
2700 | if (numberCutGenerators_) { |
---|
2701 | generator_ = new CbcCutGenerator * [numberCutGenerators_]; |
---|
2702 | virginGenerator_ = new CbcCutGenerator * [numberCutGenerators_]; |
---|
2703 | int i; |
---|
2704 | for (i=0;i<numberCutGenerators_;i++) { |
---|
2705 | generator_[i]=new CbcCutGenerator(*rhs.generator_[i]); |
---|
2706 | virginGenerator_[i]=new CbcCutGenerator(*rhs.virginGenerator_[i]); |
---|
2707 | } |
---|
2708 | } else { |
---|
2709 | generator_=NULL; |
---|
2710 | virginGenerator_=NULL; |
---|
2711 | } |
---|
2712 | numberHeuristics_ = rhs.numberHeuristics_; |
---|
2713 | if (numberHeuristics_) { |
---|
2714 | heuristic_ = new CbcHeuristic * [numberHeuristics_]; |
---|
2715 | memcpy(heuristic_,rhs.heuristic_, |
---|
2716 | numberHeuristics_*sizeof(CbcHeuristic *)); |
---|
2717 | } else { |
---|
2718 | heuristic_=NULL; |
---|
2719 | } |
---|
2720 | lastHeuristic_ = NULL; |
---|
2721 | if (eventHandler_) |
---|
2722 | delete eventHandler_ ; |
---|
2723 | if (rhs.eventHandler_) |
---|
2724 | { eventHandler_ = new CbcEventHandler(*rhs.eventHandler_) ; } |
---|
2725 | else |
---|
2726 | { eventHandler_ = NULL ; } |
---|
2727 | for (i=0;i<numberObjects_;i++) |
---|
2728 | delete object_[i]; |
---|
2729 | delete [] object_; |
---|
2730 | numberObjects_=rhs.numberObjects_; |
---|
2731 | if (numberObjects_) { |
---|
2732 | object_ = new CbcObject * [numberObjects_]; |
---|
2733 | int i; |
---|
2734 | for (i=0;i<numberObjects_;i++) |
---|
2735 | object_[i]=(rhs.object_[i])->clone(); |
---|
2736 | } else { |
---|
2737 | object_=NULL; |
---|
2738 | } |
---|
2739 | delete [] originalColumns_; |
---|
2740 | if (rhs.originalColumns_) { |
---|
2741 | int numberColumns = rhs.getNumCols(); |
---|
2742 | originalColumns_= new int [numberColumns]; |
---|
2743 | memcpy(originalColumns_,rhs.originalColumns_,numberColumns*sizeof(int)); |
---|
2744 | } else { |
---|
2745 | originalColumns_=NULL; |
---|
2746 | } |
---|
2747 | nodeCompare_=rhs.nodeCompare_->clone(); |
---|
2748 | problemFeasibility_=rhs.problemFeasibility_->clone(); |
---|
2749 | delete tree_; |
---|
2750 | tree_= rhs.tree_->clone(); |
---|
2751 | branchingMethod_=rhs.branchingMethod_; |
---|
2752 | delete strategy_; |
---|
2753 | if (rhs.strategy_) |
---|
2754 | strategy_=rhs.strategy_->clone(); |
---|
2755 | else |
---|
2756 | strategy_=NULL; |
---|
2757 | parentModel_=rhs.parentModel_; |
---|
2758 | appData_=rhs.appData_; |
---|
2759 | |
---|
2760 | delete [] integerVariable_; |
---|
2761 | numberIntegers_=rhs.numberIntegers_; |
---|
2762 | if (numberIntegers_) { |
---|
2763 | integerVariable_ = new int [numberIntegers_]; |
---|
2764 | memcpy(integerVariable_,rhs.integerVariable_, |
---|
2765 | numberIntegers_*sizeof(int)); |
---|
2766 | integerInfo_ = CoinCopyOfArray(rhs.integerInfo_,solver_->getNumCols()); |
---|
2767 | } else { |
---|
2768 | integerVariable_ = NULL; |
---|
2769 | integerInfo_=NULL; |
---|
2770 | } |
---|
2771 | if (rhs.hotstartSolution_) { |
---|
2772 | int numberColumns = solver_->getNumCols(); |
---|
2773 | hotstartSolution_ = CoinCopyOfArray(rhs.hotstartSolution_,numberColumns); |
---|
2774 | hotstartPriorities_ = CoinCopyOfArray(rhs.hotstartPriorities_,numberColumns); |
---|
2775 | } else { |
---|
2776 | hotstartSolution_ = NULL; |
---|
2777 | hotstartPriorities_ =NULL; |
---|
2778 | } |
---|
2779 | numberRowsAtContinuous_ = rhs.numberRowsAtContinuous_; |
---|
2780 | maximumNumberCuts_=rhs.maximumNumberCuts_; |
---|
2781 | phase_ = rhs.phase_; |
---|
2782 | currentNumberCuts_=rhs.currentNumberCuts_; |
---|
2783 | maximumDepth_= rhs.maximumDepth_; |
---|
2784 | delete [] addedCuts_; |
---|
2785 | delete [] walkback_; |
---|
2786 | // These are only used as temporary arrays so need not be filled |
---|
2787 | if (maximumNumberCuts_) { |
---|
2788 | addedCuts_ = new CbcCountRowCut * [maximumNumberCuts_]; |
---|
2789 | } else { |
---|
2790 | addedCuts_ = NULL; |
---|
2791 | } |
---|
2792 | nextRowCut_ = NULL; |
---|
2793 | currentNode_ = NULL; |
---|
2794 | if (maximumDepth_) |
---|
2795 | walkback_ = new CbcNodeInfo * [maximumDepth_]; |
---|
2796 | else |
---|
2797 | walkback_ = NULL; |
---|
2798 | synchronizeModel(); |
---|
2799 | cbcColLower_ = NULL; |
---|
2800 | cbcColUpper_ = NULL; |
---|
2801 | cbcRowLower_ = NULL; |
---|
2802 | cbcRowUpper_ = NULL; |
---|
2803 | cbcColSolution_ = NULL; |
---|
2804 | cbcRowPrice_ = NULL; |
---|
2805 | cbcReducedCost_ = NULL; |
---|
2806 | cbcRowActivity_ = NULL; |
---|
2807 | } |
---|
2808 | return *this; |
---|
2809 | } |
---|
2810 | |
---|
2811 | // Destructor |
---|
2812 | CbcModel::~CbcModel () |
---|
2813 | { |
---|
2814 | if (defaultHandler_) { |
---|
2815 | delete handler_; |
---|
2816 | handler_ = NULL; |
---|
2817 | } |
---|
2818 | delete tree_; |
---|
2819 | tree_=NULL; |
---|
2820 | if (ourSolver_) delete solver_; |
---|
2821 | gutsOfDestructor(); |
---|
2822 | delete eventHandler_ ; |
---|
2823 | eventHandler_ = NULL ; |
---|
2824 | } |
---|
2825 | // Clears out as much as possible (except solver) |
---|
2826 | void |
---|
2827 | CbcModel::gutsOfDestructor() |
---|
2828 | { |
---|
2829 | delete referenceSolver_; |
---|
2830 | referenceSolver_=NULL; |
---|
2831 | int i; |
---|
2832 | for (i=0;i<numberCutGenerators_;i++) { |
---|
2833 | delete generator_[i]; |
---|
2834 | delete virginGenerator_[i]; |
---|
2835 | } |
---|
2836 | delete [] generator_; |
---|
2837 | delete [] virginGenerator_; |
---|
2838 | generator_=NULL; |
---|
2839 | virginGenerator_=NULL; |
---|
2840 | for (i=0;i<numberHeuristics_;i++) |
---|
2841 | delete heuristic_[i]; |
---|
2842 | delete [] heuristic_; |
---|
2843 | heuristic_=NULL; |
---|
2844 | delete nodeCompare_; |
---|
2845 | nodeCompare_=NULL; |
---|
2846 | delete problemFeasibility_; |
---|
2847 | problemFeasibility_=NULL; |
---|
2848 | delete [] originalColumns_; |
---|
2849 | originalColumns_=NULL; |
---|
2850 | delete strategy_; |
---|
2851 | gutsOfDestructor2(); |
---|
2852 | } |
---|
2853 | // Clears out enough to reset CbcModel |
---|
2854 | void |
---|
2855 | CbcModel::gutsOfDestructor2() |
---|
2856 | { |
---|
2857 | delete [] integerInfo_; |
---|
2858 | integerInfo_=NULL; |
---|
2859 | delete [] integerVariable_; |
---|
2860 | integerVariable_=NULL; |
---|
2861 | int i; |
---|
2862 | for (i=0;i<numberObjects_;i++) |
---|
2863 | delete object_[i]; |
---|
2864 | delete [] object_; |
---|
2865 | object_=NULL; |
---|
2866 | numberIntegers_=0; |
---|
2867 | numberObjects_=0; |
---|
2868 | delete emptyWarmStart_ ; |
---|
2869 | emptyWarmStart_ =NULL; |
---|
2870 | delete continuousSolver_; |
---|
2871 | continuousSolver_=NULL; |
---|
2872 | delete [] bestSolution_; |
---|
2873 | bestSolution_=NULL; |
---|
2874 | delete [] currentSolution_; |
---|
2875 | currentSolution_=NULL; |
---|
2876 | delete [] continuousSolution_; |
---|
2877 | continuousSolution_=NULL; |
---|
2878 | solverCharacteristics_=NULL; |
---|
2879 | delete [] usedInSolution_; |
---|
2880 | usedInSolution_ = NULL; |
---|
2881 | testSolution_=NULL; |
---|
2882 | lastHeuristic_ = NULL; |
---|
2883 | delete [] addedCuts_; |
---|
2884 | addedCuts_=NULL; |
---|
2885 | nextRowCut_ = NULL; |
---|
2886 | currentNode_ = NULL; |
---|
2887 | delete [] walkback_; |
---|
2888 | walkback_=NULL; |
---|
2889 | delete [] whichGenerator_; |
---|
2890 | whichGenerator_ = NULL; |
---|
2891 | for (i=0;i<maximumStatistics_;i++) |
---|
2892 | delete statistics_[i]; |
---|
2893 | delete [] statistics_; |
---|
2894 | statistics_=NULL; |
---|
2895 | maximumStatistics_=0; |
---|
2896 | delete [] analyzeResults_; |
---|
2897 | analyzeResults_=NULL; |
---|
2898 | // Below here is whatever consensus is |
---|
2899 | ourSolver_=true; |
---|
2900 | bestObjective_=COIN_DBL_MAX; |
---|
2901 | bestPossibleObjective_=COIN_DBL_MAX; |
---|
2902 | sumChangeObjective1_=0.0; |
---|
2903 | sumChangeObjective2_=0.0; |
---|
2904 | numberSolutions_=0; |
---|
2905 | stateOfSearch_=0; |
---|
2906 | delete [] hotstartSolution_; |
---|
2907 | hotstartSolution_=NULL; |
---|
2908 | delete [] hotstartPriorities_; |
---|
2909 | hotstartPriorities_=NULL; |
---|
2910 | numberHeuristicSolutions_=0; |
---|
2911 | numberNodes_=0; |
---|
2912 | numberNodes2_=0; |
---|
2913 | numberIterations_=0; |
---|
2914 | status_=-1; |
---|
2915 | secondaryStatus_=-1; |
---|
2916 | maximumNumberCuts_=0; |
---|
2917 | phase_=0; |
---|
2918 | currentNumberCuts_=0; |
---|
2919 | maximumDepth_=0; |
---|
2920 | nextRowCut_=NULL; |
---|
2921 | currentNode_=NULL; |
---|
2922 | // clear out tree |
---|
2923 | if (tree_&&tree_->size()) |
---|
2924 | tree_->cleanTree(this, -1.0e100,bestPossibleObjective_) ; |
---|
2925 | subTreeModel_=NULL; |
---|
2926 | numberStoppedSubTrees_=0; |
---|
2927 | numberInfeasibleNodes_=0; |
---|
2928 | numberGlobalViolations_=0; |
---|
2929 | continuousObjective_=0.0; |
---|
2930 | originalContinuousObjective_=0.0; |
---|
2931 | continuousInfeasibilities_=0; |
---|
2932 | numberFixedAtRoot_=0; |
---|
2933 | numberFixedNow_=0; |
---|
2934 | stoppedOnGap_=false; |
---|
2935 | eventHappened_=false; |
---|
2936 | numberLongStrong_=0; |
---|
2937 | numberOldActiveCuts_=0; |
---|
2938 | numberNewCuts_=0; |
---|
2939 | searchStrategy_=-1; |
---|
2940 | numberStrongIterations_=0; |
---|
2941 | // Parameters which need to be reset |
---|
2942 | dblParam_[CbcCutoffIncrement] = 1e-5; |
---|
2943 | dblParam_[CbcCurrentCutoff] = 1.0e100; |
---|
2944 | dblParam_[CbcCurrentObjectiveValue] = 1.0e100; |
---|
2945 | dblParam_[CbcCurrentMinimizationObjectiveValue] = 1.0e100; |
---|
2946 | } |
---|
2947 | // Save a copy of the current solver so can be reset to |
---|
2948 | void |
---|
2949 | CbcModel::saveReferenceSolver() |
---|
2950 | { |
---|
2951 | delete referenceSolver_; |
---|
2952 | referenceSolver_= solver_->clone(); |
---|
2953 | } |
---|
2954 | |
---|
2955 | // Uses a copy of reference solver to be current solver |
---|
2956 | void |
---|
2957 | CbcModel::resetToReferenceSolver() |
---|
2958 | { |
---|
2959 | delete solver_; |
---|
2960 | solver_ = referenceSolver_->clone(); |
---|
2961 | // clear many things |
---|
2962 | gutsOfDestructor2(); |
---|
2963 | // Reset cutoff |
---|
2964 | // Solvers know about direction |
---|
2965 | double direction = solver_->getObjSense(); |
---|
2966 | double value; |
---|
2967 | solver_->getDblParam(OsiDualObjectiveLimit,value); |
---|
2968 | setCutoff(value*direction); |
---|
2969 | } |
---|
2970 | |
---|
2971 | // Are there a numerical difficulties? |
---|
2972 | bool |
---|
2973 | CbcModel::isAbandoned() const |
---|
2974 | { |
---|
2975 | return status_ == 2; |
---|
2976 | } |
---|
2977 | // Is optimality proven? |
---|
2978 | bool |
---|
2979 | CbcModel::isProvenOptimal() const |
---|
2980 | { |
---|
2981 | if (!status_ && bestObjective_<1.0e30) |
---|
2982 | return true; |
---|
2983 | else |
---|
2984 | return false; |
---|
2985 | } |
---|
2986 | // Is infeasiblity proven (or none better than cutoff)? |
---|
2987 | bool |
---|
2988 | CbcModel::isProvenInfeasible() const |
---|
2989 | { |
---|
2990 | if (!status_ && bestObjective_>=1.0e30) |
---|
2991 | return true; |
---|
2992 | else |
---|
2993 | return false; |
---|
2994 | } |
---|
2995 | // Node limit reached? |
---|
2996 | bool |
---|
2997 | CbcModel::isNodeLimitReached() const |
---|
2998 | { |
---|
2999 | return numberNodes_ >= intParam_[CbcMaxNumNode]; |
---|
3000 | } |
---|
3001 | // Time limit reached? |
---|
3002 | bool |
---|
3003 | CbcModel::isSecondsLimitReached() const |
---|
3004 | { |
---|
3005 | if (status_==1&&secondaryStatus_==4) |
---|
3006 | return true; |
---|
3007 | else |
---|
3008 | return false; |
---|
3009 | } |
---|
3010 | // Solution limit reached? |
---|
3011 | bool |
---|
3012 | CbcModel::isSolutionLimitReached() const |
---|
3013 | { |
---|
3014 | return numberSolutions_ >= intParam_[CbcMaxNumSol]; |
---|
3015 | } |
---|
3016 | // Set language |
---|
3017 | void |
---|
3018 | CbcModel::newLanguage(CoinMessages::Language language) |
---|
3019 | { |
---|
3020 | messages_ = CbcMessage(language); |
---|
3021 | } |
---|
3022 | void |
---|
3023 | CbcModel::setNumberStrong(int number) |
---|
3024 | { |
---|
3025 | if (number<0) |
---|
3026 | numberStrong_=0; |
---|
3027 | else |
---|
3028 | numberStrong_=number; |
---|
3029 | } |
---|
3030 | void |
---|
3031 | CbcModel::setNumberBeforeTrust(int number) |
---|
3032 | { |
---|
3033 | if (number<-1) { |
---|
3034 | numberBeforeTrust_=0; |
---|
3035 | } else { |
---|
3036 | numberBeforeTrust_=number; |
---|
3037 | //numberStrong_ = CoinMax(numberStrong_,1); |
---|
3038 | } |
---|
3039 | } |
---|
3040 | void |
---|
3041 | CbcModel::setNumberPenalties(int number) |
---|
3042 | { |
---|
3043 | if (number<=0) { |
---|
3044 | numberPenalties_=0; |
---|
3045 | } else { |
---|
3046 | numberPenalties_=number; |
---|
3047 | } |
---|
3048 | } |
---|
3049 | void |
---|
3050 | CbcModel::setPenaltyScaleFactor(double value) |
---|
3051 | { |
---|
3052 | if (value<=0) { |
---|
3053 | penaltyScaleFactor_=3.0; |
---|
3054 | } else { |
---|
3055 | penaltyScaleFactor_=value; |
---|
3056 | } |
---|
3057 | } |
---|
3058 | void |
---|
3059 | CbcModel::setHowOftenGlobalScan(int number) |
---|
3060 | { |
---|
3061 | if (number<-1) |
---|
3062 | howOftenGlobalScan_=0; |
---|
3063 | else |
---|
3064 | howOftenGlobalScan_=number; |
---|
3065 | } |
---|
3066 | |
---|
3067 | // Add one generator |
---|
3068 | void |
---|
3069 | CbcModel::addCutGenerator(CglCutGenerator * generator, |
---|
3070 | int howOften, const char * name, |
---|
3071 | bool normal, bool atSolution, |
---|
3072 | bool whenInfeasible,int howOftenInSub, |
---|
3073 | int whatDepth, int whatDepthInSub) |
---|
3074 | { |
---|
3075 | CbcCutGenerator ** temp = generator_; |
---|
3076 | generator_ = new CbcCutGenerator * [numberCutGenerators_+1]; |
---|
3077 | memcpy(generator_,temp,numberCutGenerators_*sizeof(CbcCutGenerator *)); |
---|
3078 | delete[] temp ; |
---|
3079 | generator_[numberCutGenerators_]= |
---|
3080 | new CbcCutGenerator(this,generator, howOften, name, |
---|
3081 | normal,atSolution,whenInfeasible,howOftenInSub, |
---|
3082 | whatDepth, whatDepthInSub); |
---|
3083 | // and before any cahnges |
---|
3084 | temp = virginGenerator_; |
---|
3085 | virginGenerator_ = new CbcCutGenerator * [numberCutGenerators_+1]; |
---|
3086 | memcpy(virginGenerator_,temp,numberCutGenerators_*sizeof(CbcCutGenerator *)); |
---|
3087 | delete[] temp ; |
---|
3088 | virginGenerator_[numberCutGenerators_++]= |
---|
3089 | new CbcCutGenerator(this,generator, howOften, name, |
---|
3090 | normal,atSolution,whenInfeasible,howOftenInSub, |
---|
3091 | whatDepth, whatDepthInSub); |
---|
3092 | |
---|
3093 | } |
---|
3094 | // Add one heuristic |
---|
3095 | void |
---|
3096 | CbcModel::addHeuristic(CbcHeuristic * generator) |
---|
3097 | { |
---|
3098 | CbcHeuristic ** temp = heuristic_; |
---|
3099 | heuristic_ = new CbcHeuristic * [numberHeuristics_+1]; |
---|
3100 | memcpy(heuristic_,temp,numberHeuristics_*sizeof(CbcHeuristic *)); |
---|
3101 | delete [] temp; |
---|
3102 | heuristic_[numberHeuristics_++]=generator->clone(); |
---|
3103 | } |
---|
3104 | |
---|
3105 | /* |
---|
3106 | The last subproblem handled by the solver is not necessarily related to the |
---|
3107 | one being recreated, so the first action is to remove all cuts from the |
---|
3108 | constraint system. Next, traverse the tree from node to the root to |
---|
3109 | determine the basis size required for this subproblem and create an empty |
---|
3110 | basis with the right capacity. Finally, traverse the tree from root to |
---|
3111 | node, adjusting bounds in the constraint system, adjusting the basis, and |
---|
3112 | collecting the cuts that must be added to the constraint system. |
---|
3113 | applyToModel does the heavy lifting. |
---|
3114 | |
---|
3115 | addCuts1 is used in contexts where all that's desired is the list of cuts: |
---|
3116 | the node is already fathomed, and we're collecting cuts so that we can |
---|
3117 | adjust reference counts as we prune nodes. Arguably the two functions |
---|
3118 | should be separated. The culprit is applyToModel, which performs cut |
---|
3119 | collection and model adjustment. |
---|
3120 | |
---|
3121 | Certainly in the contexts where all we need is a list of cuts, there's no |
---|
3122 | point in passing in a valid basis --- an empty basis will do just fine. |
---|
3123 | */ |
---|
3124 | void CbcModel::addCuts1 (CbcNode * node, CoinWarmStartBasis *&lastws) |
---|
3125 | { int i; |
---|
3126 | int nNode=0; |
---|
3127 | int numberColumns = getNumCols(); |
---|
3128 | CbcNodeInfo * nodeInfo = node->nodeInfo(); |
---|
3129 | |
---|
3130 | /* |
---|
3131 | Remove all cuts from the constraint system. |
---|
3132 | (original comment includes ``see note below for later efficiency'', but |
---|
3133 | the reference isn't clear to me). |
---|
3134 | */ |
---|
3135 | int currentNumberCuts = solver_->getNumRows()-numberRowsAtContinuous_; |
---|
3136 | int *which = new int[currentNumberCuts]; |
---|
3137 | for (i = 0 ; i < currentNumberCuts ; i++) |
---|
3138 | which[i] = i+numberRowsAtContinuous_; |
---|
3139 | solver_->deleteRows(currentNumberCuts,which); |
---|
3140 | delete [] which; |
---|
3141 | /* |
---|
3142 | Accumulate the path from node to the root in walkback_, and accumulate a |
---|
3143 | cut count in currentNumberCuts. |
---|
3144 | |
---|
3145 | original comment: when working then just unwind until where new node joins |
---|
3146 | old node (for cuts?) |
---|
3147 | */ |
---|
3148 | currentNumberCuts = 0; |
---|
3149 | while (nodeInfo) { |
---|
3150 | //printf("nNode = %d, nodeInfo = %x\n",nNode,nodeInfo); |
---|
3151 | walkback_[nNode++]=nodeInfo; |
---|
3152 | currentNumberCuts += nodeInfo->numberCuts() ; |
---|
3153 | nodeInfo = nodeInfo->parent() ; |
---|
3154 | if (nNode==maximumDepth_) { |
---|
3155 | maximumDepth_ *= 2; |
---|
3156 | CbcNodeInfo ** temp = new CbcNodeInfo * [maximumDepth_]; |
---|
3157 | for (i=0;i<nNode;i++) |
---|
3158 | temp[i] = walkback_[i]; |
---|
3159 | delete [] walkback_; |
---|
3160 | walkback_ = temp; |
---|
3161 | } |
---|
3162 | } |
---|
3163 | /* |
---|
3164 | Create an empty basis with sufficient capacity for the constraint system |
---|
3165 | we'll construct: original system plus cuts. Make sure we have capacity to |
---|
3166 | record those cuts in addedCuts_. |
---|
3167 | |
---|
3168 | The method of adjusting the basis at a FullNodeInfo object (the root, for |
---|
3169 | example) is to use a copy constructor to duplicate the basis held in the |
---|
3170 | nodeInfo, then resize it and return the new basis object. Guaranteed, |
---|
3171 | lastws will point to a different basis when it returns. We pass in a basis |
---|
3172 | because we need the parameter to return the allocated basis, and it's an |
---|
3173 | easy way to pass in the size. But we take a hit for memory allocation. |
---|
3174 | */ |
---|
3175 | currentNumberCuts_=currentNumberCuts; |
---|
3176 | if (currentNumberCuts >= maximumNumberCuts_) { |
---|
3177 | maximumNumberCuts_ = currentNumberCuts; |
---|
3178 | delete [] addedCuts_; |
---|
3179 | addedCuts_ = new CbcCountRowCut * [maximumNumberCuts_]; |
---|
3180 | } |
---|
3181 | lastws->setSize(numberColumns,numberRowsAtContinuous_+currentNumberCuts); |
---|
3182 | /* |
---|
3183 | This last bit of code traverses the path collected in walkback_ from the |
---|
3184 | root back to node. At the end of the loop, |
---|
3185 | * lastws will be an appropriate basis for node; |
---|
3186 | * variable bounds in the constraint system will be set to be correct for |
---|
3187 | node; and |
---|
3188 | * addedCuts_ will be set to a list of cuts that need to be added to the |
---|
3189 | constraint system at node. |
---|
3190 | applyToModel does all the heavy lifting. |
---|
3191 | */ |
---|
3192 | currentNumberCuts=0; |
---|
3193 | while (nNode) { |
---|
3194 | --nNode; |
---|
3195 | walkback_[nNode]->applyToModel(this,lastws,addedCuts_,currentNumberCuts); |
---|
3196 | } |
---|
3197 | } |
---|
3198 | |
---|
3199 | /* |
---|
3200 | adjustCuts might be a better name: If the node is feasible, we sift through |
---|
3201 | the cuts collected by addCuts1, add the ones that are tight and omit the |
---|
3202 | ones that are loose. If the node is infeasible, we just adjust the |
---|
3203 | reference counts to reflect that we're about to prune this node and its |
---|
3204 | descendants. |
---|
3205 | */ |
---|
3206 | int CbcModel::addCuts (CbcNode *node, CoinWarmStartBasis *&lastws,bool canFix) |
---|
3207 | { |
---|
3208 | /* |
---|
3209 | addCuts1 performs step 1 of restoring the subproblem at this node; see the |
---|
3210 | comments there. |
---|
3211 | */ |
---|
3212 | addCuts1(node,lastws); |
---|
3213 | int i; |
---|
3214 | int numberColumns = getNumCols(); |
---|
3215 | CbcNodeInfo * nodeInfo = node->nodeInfo(); |
---|
3216 | double cutoff = getCutoff() ; |
---|
3217 | int currentNumberCuts=currentNumberCuts_; |
---|
3218 | if (canFix) { |
---|
3219 | bool feasible=true; |
---|
3220 | const double *lower = solver_->getColLower() ; |
---|
3221 | const double *upper = solver_->getColUpper() ; |
---|
3222 | double * newLower = analyzeResults_; |
---|
3223 | double * objLower = newLower+numberIntegers_; |
---|
3224 | double * newUpper = objLower+numberIntegers_; |
---|
3225 | double * objUpper = newUpper+numberIntegers_; |
---|
3226 | int n=0; |
---|
3227 | for (i=0;i<numberIntegers_;i++) { |
---|
3228 | int iColumn = integerVariable_[i]; |
---|
3229 | bool changed=false; |
---|
3230 | double lo = 0.0; |
---|
3231 | double up = 0.0; |
---|
3232 | if (objLower[i]>cutoff) { |
---|
3233 | lo = lower[iColumn]; |
---|
3234 | up = upper[iColumn]; |
---|
3235 | if (lo<newLower[i]) { |
---|
3236 | lo = newLower[i]; |
---|
3237 | solver_->setColLower(iColumn,lo); |
---|
3238 | changed=true; |
---|
3239 | n++; |
---|
3240 | } |
---|
3241 | if (objUpper[i]>cutoff) { |
---|
3242 | if (up>newUpper[i]) { |
---|
3243 | up = newUpper[i]; |
---|
3244 | solver_->setColUpper(iColumn,up); |
---|
3245 | changed=true; |
---|
3246 | n++; |
---|
3247 | } |
---|
3248 | } |
---|
3249 | } else if (objUpper[i]>cutoff) { |
---|
3250 | lo = lower[iColumn]; |
---|
3251 | up = upper[iColumn]; |
---|
3252 | if (up>newUpper[i]) { |
---|
3253 | up = newUpper[i]; |
---|
3254 | solver_->setColUpper(iColumn,up); |
---|
3255 | changed=true; |
---|
3256 | n++; |
---|
3257 | } |
---|
3258 | } |
---|
3259 | if (changed&&lo>up) { |
---|
3260 | feasible=false; |
---|
3261 | break; |
---|
3262 | } |
---|
3263 | } |
---|
3264 | if (!feasible) { |
---|
3265 | printf("analysis says node infeas\n"); |
---|
3266 | cutoff=-COIN_DBL_MAX; |
---|
3267 | } |
---|
3268 | } |
---|
3269 | /* |
---|
3270 | If the node can't be fathomed by bound, reinstall tight cuts in the |
---|
3271 | constraint system. Even if there are no cuts, we'll want to set the |
---|
3272 | reconstructed basis in the solver. |
---|
3273 | */ |
---|
3274 | if (node->objectiveValue() < cutoff) |
---|
3275 | { |
---|
3276 | # ifdef CBC_CHECK_BASIS |
---|
3277 | printf("addCuts: expanded basis; rows %d+%d\n", |
---|
3278 | numberRowsAtContinuous_,currentNumberCuts); |
---|
3279 | lastws->print(); |
---|
3280 | # endif |
---|
3281 | /* |
---|
3282 | Adjust the basis and constraint system so that we retain only active cuts. |
---|
3283 | There are three steps: |
---|
3284 | 1) Scan the basis. Sort the cuts into effective cuts to be kept and |
---|
3285 | loose cuts to be dropped. |
---|
3286 | 2) Drop the loose cuts and resize the basis to fit. |
---|
3287 | 3) Install the tight cuts in the constraint system (applyRowCuts) and |
---|
3288 | and install the basis (setWarmStart). |
---|
3289 | Use of compressRows conveys we're compressing the basis and not just |
---|
3290 | tweaking the artificialStatus_ array. |
---|
3291 | */ |
---|
3292 | if (currentNumberCuts > 0) { |
---|
3293 | int numberToAdd = 0; |
---|
3294 | const OsiRowCut **addCuts; |
---|
3295 | int numberToDrop = 0 ; |
---|
3296 | int *cutsToDrop ; |
---|
3297 | addCuts = new const OsiRowCut* [currentNumberCuts]; |
---|
3298 | cutsToDrop = new int[currentNumberCuts] ; |
---|
3299 | int nxrow = lastws->getNumArtificial(); |
---|
3300 | for (i=0;i<currentNumberCuts;i++) { |
---|
3301 | assert (i+numberRowsAtContinuous_<nxrow); |
---|
3302 | CoinWarmStartBasis::Status status = |
---|
3303 | lastws->getArtifStatus(i+numberRowsAtContinuous_); |
---|
3304 | if (addedCuts_[i] && |
---|
3305 | (status != CoinWarmStartBasis::basic || |
---|
3306 | addedCuts_[i]->effectiveness()==COIN_DBL_MAX)) { |
---|
3307 | # ifdef CHECK_CUT_COUNTS |
---|
3308 | printf("Using cut %d %x as row %d\n",i,addedCuts_[i], |
---|
3309 | numberRowsAtContinuous_+numberToAdd); |
---|
3310 | # endif |
---|
3311 | addCuts[numberToAdd++] = new OsiRowCut(*addedCuts_[i]); |
---|
3312 | } else { |
---|
3313 | # ifdef CHECK_CUT_COUNTS |
---|
3314 | printf("Dropping cut %d %x\n",i,addedCuts_[i]); |
---|
3315 | # endif |
---|
3316 | addedCuts_[i]=NULL; |
---|
3317 | cutsToDrop[numberToDrop++] = numberRowsAtContinuous_+i ; |
---|
3318 | } |
---|
3319 | } |
---|
3320 | int numberRowsNow=numberRowsAtContinuous_+numberToAdd; |
---|
3321 | lastws->compressRows(numberToDrop,cutsToDrop) ; |
---|
3322 | lastws->resize(numberRowsNow,numberColumns); |
---|
3323 | solver_->applyRowCuts(numberToAdd,addCuts); |
---|
3324 | # ifdef CBC_CHECK_BASIS |
---|
3325 | printf("addCuts: stripped basis; rows %d + %d\n", |
---|
3326 | numberRowsAtContinuous_,numberToAdd); |
---|
3327 | lastws->print(); |
---|
3328 | # endif |
---|
3329 | for (i=0;i<numberToAdd;i++) |
---|
3330 | delete addCuts[i]; |
---|
3331 | delete [] addCuts; |
---|
3332 | delete [] cutsToDrop ; |
---|
3333 | } |
---|
3334 | /* |
---|
3335 | Set the basis in the solver. |
---|
3336 | */ |
---|
3337 | solver_->setWarmStart(lastws); |
---|
3338 | #if 0 |
---|
3339 | if ((numberNodes_%printFrequency_)==0) { |
---|
3340 | printf("Objective %g, depth %d, unsatisfied %d\n", |
---|
3341 | node->objectiveValue(), |
---|
3342 | node->depth(),node->numberUnsatisfied()); |
---|
3343 | } |
---|
3344 | #endif |
---|
3345 | /* |
---|
3346 | Clean up and we're out of here. |
---|
3347 | */ |
---|
3348 | numberNodes_++; |
---|
3349 | return 0; |
---|
3350 | } |
---|
3351 | /* |
---|
3352 | This node has been fathomed by bound as we try to revive it out of the live |
---|
3353 | set. Adjust the cut reference counts to reflect that we no longer need to |
---|
3354 | explore the remaining branch arms, hence they will no longer reference any |
---|
3355 | cuts. Cuts whose reference count falls to zero are deleted. |
---|
3356 | */ |
---|
3357 | else |
---|
3358 | { int i; |
---|
3359 | int numberLeft = nodeInfo->numberBranchesLeft(); |
---|
3360 | for (i = 0 ; i < currentNumberCuts ; i++) |
---|
3361 | { if (addedCuts_[i]) |
---|
3362 | { if (!addedCuts_[i]->decrement(numberLeft)) |
---|
3363 | { delete addedCuts_[i]; |
---|
3364 | addedCuts_[i] = NULL; } } } |
---|
3365 | return 1 ; } |
---|
3366 | } |
---|
3367 | |
---|
3368 | |
---|
3369 | /* |
---|
3370 | Perform reduced cost fixing on integer variables. |
---|
3371 | |
---|
3372 | The variables in question are already nonbasic at bound. We're just nailing |
---|
3373 | down the current situation. |
---|
3374 | */ |
---|
3375 | |
---|
3376 | int CbcModel::reducedCostFix () |
---|
3377 | |
---|
3378 | { |
---|
3379 | if(!solverCharacteristics_->reducedCostsAccurate()) |
---|
3380 | return 0; //NLP |
---|
3381 | double cutoff = getCutoff() ; |
---|
3382 | double direction = solver_->getObjSense() ; |
---|
3383 | double gap = cutoff - solver_->getObjValue()*direction ; |
---|
3384 | double tolerance; |
---|
3385 | solver_->getDblParam(OsiDualTolerance,tolerance) ; |
---|
3386 | if (gap<=0.0) |
---|
3387 | return 0; |
---|
3388 | gap += 100.0*tolerance; |
---|
3389 | double integerTolerance = getDblParam(CbcIntegerTolerance) ; |
---|
3390 | |
---|
3391 | const double *lower = solver_->getColLower() ; |
---|
3392 | const double *upper = solver_->getColUpper() ; |
---|
3393 | const double *solution = solver_->getColSolution() ; |
---|
3394 | const double *reducedCost = solver_->getReducedCost() ; |
---|
3395 | |
---|
3396 | int numberFixed = 0 ; |
---|
3397 | |
---|
3398 | # ifdef CBC_USE_CLP |
---|
3399 | OsiClpSolverInterface * clpSolver |
---|
3400 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
3401 | ClpSimplex * clpSimplex=NULL; |
---|
3402 | if (clpSolver) |
---|
3403 | clpSimplex = clpSolver->getModelPtr(); |
---|
3404 | # endif |
---|
3405 | for (int i = 0 ; i < numberIntegers_ ; i++) |
---|
3406 | { int iColumn = integerVariable_[i] ; |
---|
3407 | double djValue = direction*reducedCost[iColumn] ; |
---|
3408 | if (upper[iColumn]-lower[iColumn] > integerTolerance) |
---|
3409 | { if (solution[iColumn] < lower[iColumn]+integerTolerance && djValue > gap) |
---|
3410 | { solver_->setColUpper(iColumn,lower[iColumn]) ; |
---|
3411 | #ifdef CBC_USE_CLP |
---|
3412 | if (clpSimplex) |
---|
3413 | assert (clpSimplex->getColumnStatus(iColumn)==ClpSimplex::atLowerBound); |
---|
3414 | #endif |
---|
3415 | numberFixed++ ; } |
---|
3416 | else |
---|
3417 | if (solution[iColumn] > upper[iColumn]-integerTolerance && -djValue > gap) |
---|
3418 | { solver_->setColLower(iColumn,upper[iColumn]) ; |
---|
3419 | #ifdef CBC_USE_CLP |
---|
3420 | if (clpSimplex) |
---|
3421 | assert (clpSimplex->getColumnStatus(iColumn)==ClpSimplex::atUpperBound); |
---|
3422 | #endif |
---|
3423 | numberFixed++ ; } } } |
---|
3424 | |
---|
3425 | return numberFixed; } |
---|
3426 | |
---|
3427 | // Collect coding to replace whichGenerator |
---|
3428 | void |
---|
3429 | CbcModel::resizeWhichGenerator(int numberNow, int numberAfter) |
---|
3430 | { |
---|
3431 | if (numberAfter > maximumWhich_) { |
---|
3432 | maximumWhich_ = CoinMax(maximumWhich_*2+100,numberAfter) ; |
---|
3433 | int * temp = new int[2*maximumWhich_] ; |
---|
3434 | memcpy(temp,whichGenerator_,numberNow*sizeof(int)) ; |
---|
3435 | delete [] whichGenerator_ ; |
---|
3436 | whichGenerator_ = temp ; |
---|
3437 | memset(whichGenerator_+numberNow,0,(maximumWhich_-numberNow)*sizeof(int)); |
---|
3438 | } |
---|
3439 | } |
---|
3440 | |
---|
3441 | /** Solve the model using cuts |
---|
3442 | |
---|
3443 | This version takes off redundant cuts from node. |
---|
3444 | Returns true if feasible. |
---|
3445 | |
---|
3446 | \todo |
---|
3447 | Why do I need to resolve the problem? What has been done between the last |
---|
3448 | relaxation and calling solveWithCuts? |
---|
3449 | |
---|
3450 | If numberTries == 0 then user did not want any cuts. |
---|
3451 | */ |
---|
3452 | |
---|
3453 | bool |
---|
3454 | CbcModel::solveWithCuts (OsiCuts &cuts, int numberTries, CbcNode *node) |
---|
3455 | /* |
---|
3456 | Parameters: |
---|
3457 | numberTries: (i) the maximum number of iterations for this round of cut |
---|
3458 | generation; if negative then we don't mind if drop is tiny. |
---|
3459 | |
---|
3460 | cuts: (o) all cuts generated in this round of cut generation |
---|
3461 | |
---|
3462 | node: (i) So we can update dynamic pseudo costs |
---|
3463 | */ |
---|
3464 | |
---|
3465 | |
---|
3466 | { |
---|
3467 | bool feasible = true ; |
---|
3468 | int lastNumberCuts = 0 ; |
---|
3469 | double lastObjective = -1.0e100 ; |
---|
3470 | int violated = 0 ; |
---|
3471 | int numberRowsAtStart = solver_->getNumRows() ; |
---|
3472 | //printf("solver had %d rows\n",numberRowsAtStart); |
---|
3473 | int numberColumns = solver_->getNumCols() ; |
---|
3474 | CoinBigIndex numberElementsAtStart = solver_->getNumElements(); |
---|
3475 | |
---|
3476 | numberOldActiveCuts_ = numberRowsAtStart-numberRowsAtContinuous_ ; |
---|
3477 | numberNewCuts_ = 0 ; |
---|
3478 | |
---|
3479 | bool onOptimalPath = false ; |
---|
3480 | const OsiRowCutDebugger *debugger = NULL; |
---|
3481 | if ((specialOptions_&1)!=0) { |
---|
3482 | /* |
---|
3483 | See OsiRowCutDebugger for details. In a nutshell, make sure that current |
---|
3484 | variable values do not conflict with a known optimal solution. (Obviously |
---|
3485 | this can be fooled when there are multiple solutions.) |
---|
3486 | */ |
---|
3487 | debugger = solver_->getRowCutDebugger() ; |
---|
3488 | if (debugger) |
---|
3489 | onOptimalPath = (debugger->onOptimalPath(*solver_)) ; |
---|
3490 | } |
---|
3491 | OsiCuts slackCuts; |
---|
3492 | /* |
---|
3493 | Resolve the problem. If we've lost feasibility, might as well bail out right |
---|
3494 | after the debug stuff. The resolve will also refresh cached copies of the |
---|
3495 | solver solution (cbcColLower_, ...) held by CbcModel. |
---|
3496 | */ |
---|
3497 | double objectiveValue = solver_->getObjValue()*solver_->getObjSense(); |
---|
3498 | if (node) |
---|
3499 | objectiveValue= node->objectiveValue(); |
---|
3500 | int returnCode = resolve(node ? node->nodeInfo() : NULL,1); |
---|
3501 | if (node&&!node->nodeInfo()->numberBranchesLeft()) |
---|
3502 | node->nodeInfo()->allBranchesGone(); // can clean up |
---|
3503 | feasible = returnCode != 0 ; |
---|
3504 | if (returnCode<0) |
---|
3505 | numberTries=0; |
---|
3506 | if (problemFeasibility_->feasible(this,0)<0) { |
---|
3507 | feasible=false; // pretend infeasible |
---|
3508 | } |
---|
3509 | |
---|
3510 | // Update branching information if wanted |
---|
3511 | if(node &&branchingMethod_) |
---|
3512 | branchingMethod_->updateInformation(solver_,node); |
---|
3513 | |
---|
3514 | #ifdef CBC_DEBUG |
---|
3515 | if (feasible) |
---|
3516 | { printf("Obj value %g (%s) %d rows\n",solver_->getObjValue(), |
---|
3517 | (solver_->isProvenOptimal())?"proven":"unproven", |
---|
3518 | solver_->getNumRows()) ; } |
---|
3519 | |
---|
3520 | else |
---|
3521 | { printf("Infeasible %d rows\n",solver_->getNumRows()) ; } |
---|
3522 | #endif |
---|
3523 | if ((specialOptions_&1)!=0) { |
---|
3524 | /* |
---|
3525 | If the RowCutDebugger said we were compatible with the optimal solution, |
---|
3526 | and now we're suddenly infeasible, we might be confused. Then again, we |
---|
3527 | may have fathomed by bound, heading for a rediscovery of an optimal solution. |
---|
3528 | */ |
---|
3529 | if (onOptimalPath && !solver_->isDualObjectiveLimitReached()) { |
---|
3530 | if (!feasible) { |
---|
3531 | solver_->writeMps("infeas"); |
---|
3532 | CoinWarmStartBasis *slack = |
---|
3533 | dynamic_cast<CoinWarmStartBasis *>(solver_->getEmptyWarmStart()) ; |
---|
3534 | solver_->setWarmStart(slack); |
---|
3535 | delete slack ; |
---|
3536 | solver_->setHintParam(OsiDoReducePrint,false,OsiHintDo,0) ; |
---|
3537 | solver_->initialSolve(); |
---|
3538 | } |
---|
3539 | assert(feasible) ; |
---|
3540 | } |
---|
3541 | } |
---|
3542 | |
---|
3543 | if (!feasible) { |
---|
3544 | numberInfeasibleNodes_++; |
---|
3545 | return (false) ; |
---|
3546 | } |
---|
3547 | sumChangeObjective1_ += solver_->getObjValue()*solver_->getObjSense() |
---|
3548 | - objectiveValue ; |
---|
3549 | if ( CoinCpuTime()-dblParam_[CbcStartSeconds] > dblParam_[CbcMaximumSeconds] ) |
---|
3550 | numberTries=0; // exit |
---|
3551 | //if ((numberNodes_%100)==0) |
---|
3552 | //printf("XXa sum obj changed by %g\n",sumChangeObjective1_); |
---|
3553 | objectiveValue = solver_->getObjValue()*solver_->getObjSense(); |
---|
3554 | // Return at once if numberTries zero |
---|
3555 | if (!numberTries) { |
---|
3556 | cuts=OsiCuts(); |
---|
3557 | numberNewCuts_=0; |
---|
3558 | return true; |
---|
3559 | } |
---|
3560 | /* |
---|
3561 | Do reduced cost fixing. |
---|
3562 | */ |
---|
3563 | reducedCostFix() ; |
---|
3564 | /* |
---|
3565 | Set up for at most numberTries rounds of cut generation. If numberTries is |
---|
3566 | negative, we'll ignore the minimumDrop_ cutoff and keep generating cuts for |
---|
3567 | the specified number of rounds. |
---|
3568 | */ |
---|
3569 | double minimumDrop = minimumDrop_ ; |
---|
3570 | if (numberTries<0) |
---|
3571 | { numberTries = -numberTries ; |
---|
3572 | minimumDrop = -1.0 ; } |
---|
3573 | /* |
---|
3574 | Is it time to scan the cuts in order to remove redundant cuts? If so, set |
---|
3575 | up to do it. |
---|
3576 | */ |
---|
3577 | # define SCANCUTS 100 |
---|
3578 | int *countColumnCuts = NULL ; |
---|
3579 | // Always accumulate row cut counts |
---|
3580 | int * countRowCuts =new int[numberCutGenerators_+numberHeuristics_] ; |
---|
3581 | memset(countRowCuts,0, |
---|
3582 | (numberCutGenerators_+numberHeuristics_)*sizeof(int)) ; |
---|
3583 | bool fullScan = false ; |
---|
3584 | if ((numberNodes_%SCANCUTS) == 0) |
---|
3585 | { fullScan = true ; |
---|
3586 | countColumnCuts = new int[numberCutGenerators_+numberHeuristics_] ; |
---|
3587 | memset(countColumnCuts,0, |
---|
3588 | (numberCutGenerators_+numberHeuristics_)*sizeof(int)) ; } |
---|
3589 | |
---|
3590 | double direction = solver_->getObjSense() ; |
---|
3591 | double startObjective = solver_->getObjValue()*direction ; |
---|
3592 | |
---|
3593 | currentPassNumber_ = 0 ; |
---|
3594 | double primalTolerance = 1.0e-7 ; |
---|
3595 | /* |
---|
3596 | Begin cut generation loop. Cuts generated during each iteration are |
---|
3597 | collected in theseCuts. The loop can be divided into four phases: |
---|
3598 | 1) Prep: Fix variables using reduced cost. In the first iteration only, |
---|
3599 | consider scanning globalCuts_ and activating any applicable cuts. |
---|
3600 | 2) Cut Generation: Call each generator and heuristic registered in the |
---|
3601 | generator_ and heuristic_ arrays. Newly generated global cuts are |
---|
3602 | copied to globalCuts_ at this time. |
---|
3603 | 3) Cut Installation and Reoptimisation: Install column and row cuts in |
---|
3604 | the solver. Copy row cuts to cuts (parameter). Reoptimise. |
---|
3605 | 4) Cut Purging: takeOffCuts() removes inactive cuts from the solver, and |
---|
3606 | does the necessary bookkeeping in the model. |
---|
3607 | */ |
---|
3608 | do |
---|
3609 | { currentPassNumber_++ ; |
---|
3610 | numberTries-- ; |
---|
3611 | OsiCuts theseCuts ; |
---|
3612 | /* |
---|
3613 | Scan previously generated global column and row cuts to see if any are |
---|
3614 | useful. |
---|
3615 | */ |
---|
3616 | int numberViolated=0; |
---|
3617 | if (currentPassNumber_ == 1 && howOftenGlobalScan_ > 0 && |
---|
3618 | (numberNodes_%howOftenGlobalScan_) == 0) |
---|
3619 | { int numberCuts = globalCuts_.sizeColCuts() ; |
---|
3620 | int i; |
---|
3621 | // possibly extend whichGenerator |
---|
3622 | resizeWhichGenerator(numberViolated, numberViolated+numberCuts); |
---|
3623 | for ( i = 0 ; i < numberCuts ; i++) |
---|
3624 | { const OsiColCut *thisCut = globalCuts_.colCutPtr(i) ; |
---|
3625 | if (thisCut->violated(cbcColSolution_)>primalTolerance) { |
---|
3626 | printf("Global cut added - violation %g\n", |
---|
3627 | thisCut->violated(cbcColSolution_)) ; |
---|
3628 | whichGenerator_[numberViolated++]=-1; |
---|
3629 | theseCuts.insert(*thisCut) ; |
---|
3630 | } |
---|
3631 | } |
---|
3632 | numberCuts = globalCuts_.sizeRowCuts() ; |
---|
3633 | // possibly extend whichGenerator |
---|
3634 | resizeWhichGenerator(numberViolated, numberViolated+numberCuts); |
---|
3635 | for ( i = 0;i<numberCuts;i++) { |
---|
3636 | const OsiRowCut * thisCut = globalCuts_.rowCutPtr(i) ; |
---|
3637 | if (thisCut->violated(cbcColSolution_)>primalTolerance) { |
---|
3638 | //printf("Global cut added - violation %g\n", |
---|
3639 | // thisCut->violated(cbcColSolution_)) ; |
---|
3640 | whichGenerator_[numberViolated++]=-1; |
---|
3641 | theseCuts.insert(*thisCut) ; |
---|
3642 | } |
---|
3643 | } |
---|
3644 | numberGlobalViolations_+=numberViolated; |
---|
3645 | } |
---|
3646 | /* |
---|
3647 | Generate new cuts (global and/or local) and/or apply heuristics. If |
---|
3648 | CglProbing is used, then it should be first as it can fix continuous |
---|
3649 | variables. |
---|
3650 | |
---|
3651 | At present, CglProbing is the only case where generateCuts will return |
---|
3652 | true. generateCuts actually modifies variable bounds in the solver when |
---|
3653 | CglProbing indicates that it can fix a variable. Reoptimisation is required |
---|
3654 | to take full advantage. |
---|
3655 | |
---|
3656 | The need to resolve here should only happen after a heuristic solution. |
---|
3657 | (Note default OSI implementation of optimalBasisIsAvailable always returns |
---|
3658 | false.) |
---|
3659 | */ |
---|
3660 | if (solverCharacteristics_->warmStart()&& |
---|
3661 | !solver_->optimalBasisIsAvailable()) { |
---|
3662 | //printf("XXXXYY no opt basis\n"); |
---|
3663 | resolve(node ? node->nodeInfo() : NULL,3); |
---|
3664 | } |
---|
3665 | if (nextRowCut_) { |
---|
3666 | // branch was a cut - add it |
---|
3667 | theseCuts.insert(*nextRowCut_); |
---|
3668 | if (handler_->logLevel()>1) |
---|
3669 | nextRowCut_->print(); |
---|
3670 | const OsiRowCut * cut=nextRowCut_; |
---|
3671 | double lb = cut->lb(); |
---|
3672 | double ub = cut->ub(); |
---|
3673 | int n=cut->row().getNumElements(); |
---|
3674 | const int * column = cut->row().getIndices(); |
---|
3675 | const double * element = cut->row().getElements(); |
---|
3676 | double sum=0.0; |
---|
3677 | for (int i=0;i<n;i++) { |
---|
3678 | int iColumn = column[i]; |
---|
3679 | double value = element[i]; |
---|
3680 | //if (cbcColSolution_[iColumn]>1.0e-7) |
---|
3681 | //printf("value of %d is %g\n",iColumn,cbcColSolution_[iColumn]); |
---|
3682 | sum += value * cbcColSolution_[iColumn]; |
---|
3683 | } |
---|
3684 | delete nextRowCut_; |
---|
3685 | nextRowCut_=NULL; |
---|
3686 | if (handler_->logLevel()>1) |
---|
3687 | printf("applying branch cut, sum is %g, bounds %g %g\n",sum,lb,ub); |
---|
3688 | // possibly extend whichGenerator |
---|
3689 | resizeWhichGenerator(numberViolated, numberViolated+1); |
---|
3690 | // set whichgenerator (also serves as marker to say don't delete0 |
---|
3691 | whichGenerator_[numberViolated++]=-2; |
---|
3692 | } |
---|
3693 | double * newSolution = new double [numberColumns] ; |
---|
3694 | double heuristicValue = getCutoff() ; |
---|
3695 | int found = -1; // no solution found |
---|
3696 | |
---|
3697 | for (int i = 0;i<numberCutGenerators_+numberHeuristics_;i++) { |
---|
3698 | int numberRowCutsBefore = theseCuts.sizeRowCuts() ; |
---|
3699 | int numberColumnCutsBefore = theseCuts.sizeColCuts() ; |
---|
3700 | if (i<numberCutGenerators_) { |
---|
3701 | bool generate = generator_[i]->normal(); |
---|
3702 | // skip if not optimal and should be (maybe a cut generator has fixed variables) |
---|
3703 | if (generator_[i]->needsOptimalBasis()&&!solver_->basisIsAvailable()) |
---|
3704 | generate=false; |
---|
3705 | if (generate) { |
---|
3706 | bool mustResolve = |
---|
3707 | generator_[i]->generateCuts(theseCuts,fullScan,node) ; |
---|
3708 | #ifdef CBC_DEBUG |
---|
3709 | { |
---|
3710 | int numberRowCutsAfter = theseCuts.sizeRowCuts() ; |
---|
3711 | int k ; |
---|
3712 | for (k = numberRowCutsBefore;k<numberRowCutsAfter;k++) { |
---|
3713 | OsiRowCut thisCut = theseCuts.rowCut(k) ; |
---|
3714 | /* check size of elements. |
---|
3715 | We can allow smaller but this helps debug generators as it |
---|
3716 | is unsafe to have small elements */ |
---|
3717 | int n=thisCut.row().getNumElements(); |
---|
3718 | const int * column = thisCut.row().getIndices(); |
---|
3719 | const double * element = thisCut.row().getElements(); |
---|
3720 | //assert (n); |
---|
3721 | for (int i=0;i<n;i++) { |
---|
3722 | double value = element[i]; |
---|
3723 | assert(fabs(value)>1.0e-12&&fabs(value)<1.0e20); |
---|
3724 | } |
---|
3725 | } |
---|
3726 | } |
---|
3727 | #endif |
---|
3728 | if (mustResolve) { |
---|
3729 | int returncode = resolve(node ? node->nodeInfo() : NULL,2); |
---|
3730 | feasible = returnCode != 0 ; |
---|
3731 | if (returncode<0) |
---|
3732 | numberTries=0; |
---|
3733 | if ((specialOptions_&1)!=0) { |
---|
3734 | debugger = solver_->getRowCutDebugger() ; |
---|
3735 | if (debugger) |
---|
3736 | onOptimalPath = (debugger->onOptimalPath(*solver_)) ; |
---|
3737 | else |
---|
3738 | onOptimalPath=false; |
---|
3739 | if (onOptimalPath && !solver_->isDualObjectiveLimitReached()) |
---|
3740 | assert(feasible) ; |
---|
3741 | } |
---|
3742 | if (!feasible) |
---|
3743 | break ; |
---|
3744 | } |
---|
3745 | } |
---|
3746 | } else { |
---|
3747 | // see if heuristic will do anything |
---|
3748 | double saveValue = heuristicValue ; |
---|
3749 | int ifSol = |
---|
3750 | heuristic_[i-numberCutGenerators_]->solution(heuristicValue, |
---|
3751 | newSolution, |
---|
3752 | theseCuts) ; |
---|
3753 | if (ifSol>0) { |
---|
3754 | // better solution found |
---|
3755 | found = i-numberCutGenerators_ ; |
---|
3756 | incrementUsed(newSolution); |
---|
3757 | } else if (ifSol<0) { |
---|
3758 | heuristicValue = saveValue ; |
---|
3759 | } |
---|
3760 | } |
---|
3761 | int numberRowCutsAfter = theseCuts.sizeRowCuts() ; |
---|
3762 | int numberColumnCutsAfter = theseCuts.sizeColCuts() ; |
---|
3763 | |
---|
3764 | if ((specialOptions_&1)!=0) { |
---|
3765 | if (onOptimalPath) { |
---|
3766 | int k ; |
---|
3767 | for (k = numberRowCutsBefore;k<numberRowCutsAfter;k++) { |
---|
3768 | OsiRowCut thisCut = theseCuts.rowCut(k) ; |
---|
3769 | if(debugger->invalidCut(thisCut)) { |
---|
3770 | solver_->writeMps("badCut"); |
---|
3771 | } |
---|
3772 | assert(!debugger->invalidCut(thisCut)) ; |
---|
3773 | } |
---|
3774 | } |
---|
3775 | } |
---|
3776 | /* |
---|
3777 | The cut generator/heuristic has done its thing, and maybe it generated some |
---|
3778 | cuts and/or a new solution. Do a bit of bookkeeping: load |
---|
3779 | whichGenerator[i] with the index of the generator responsible for a cut, |
---|
3780 | and place cuts flagged as global in the global cut pool for the model. |
---|
3781 | |
---|
3782 | lastNumberCuts is the sum of cuts added in previous iterations; it's the |
---|
3783 | offset to the proper starting position in whichGenerator. |
---|
3784 | */ |
---|
3785 | int numberBefore = |
---|
3786 | numberRowCutsBefore+numberColumnCutsBefore+lastNumberCuts ; |
---|
3787 | int numberAfter = |
---|
3788 | numberRowCutsAfter+numberColumnCutsAfter+lastNumberCuts ; |
---|
3789 | // possibly extend whichGenerator |
---|
3790 | resizeWhichGenerator(numberBefore, numberAfter); |
---|
3791 | int j ; |
---|
3792 | if (fullScan) { |
---|
3793 | // counts |
---|
3794 | countColumnCuts[i] += numberColumnCutsAfter-numberColumnCutsBefore ; |
---|
3795 | } |
---|
3796 | countRowCuts[i] += numberRowCutsAfter-numberRowCutsBefore ; |
---|
3797 | |
---|
3798 | for (j = numberRowCutsBefore;j<numberRowCutsAfter;j++) { |
---|
3799 | whichGenerator_[numberBefore++] = i ; |
---|
3800 | const OsiRowCut * thisCut = theseCuts.rowCutPtr(j) ; |
---|
3801 | if (thisCut->lb()>thisCut->ub()) |
---|
3802 | violated=-2; // sub-problem is infeasible |
---|
3803 | if (thisCut->globallyValid()) { |
---|
3804 | // add to global list |
---|
3805 | globalCuts_.insert(*thisCut) ; |
---|
3806 | } |
---|
3807 | } |
---|
3808 | for (j = numberColumnCutsBefore;j<numberColumnCutsAfter;j++) { |
---|
3809 | whichGenerator_[numberBefore++] = i ; |
---|
3810 | const OsiColCut * thisCut = theseCuts.colCutPtr(j) ; |
---|
3811 | if (thisCut->globallyValid()) { |
---|
3812 | // add to global list |
---|
3813 | globalCuts_.insert(*thisCut) ; |
---|
3814 | } |
---|
3815 | } |
---|
3816 | } |
---|
3817 | // If at root node and first pass do heuristics without cuts |
---|
3818 | if (!numberNodes_&¤tPassNumber_==1) { |
---|
3819 | // Save number solutions |
---|
3820 | int saveNumberSolutions = numberSolutions_; |
---|
3821 | int saveNumberHeuristicSolutions = numberHeuristicSolutions_; |
---|
3822 | for (int i = 0;i<numberHeuristics_;i++) { |
---|
3823 | // see if heuristic will do anything |
---|
3824 | double saveValue = heuristicValue ; |
---|
3825 | int ifSol = heuristic_[i]->solution(heuristicValue, |
---|
3826 | newSolution); |
---|
3827 | if (ifSol>0) { |
---|
3828 | // better solution found |
---|
3829 | found = i ; |
---|
3830 | incrementUsed(newSolution); |
---|
3831 | // increment number of solutions so other heuristics can test |
---|
3832 | numberSolutions_++; |
---|
3833 | numberHeuristicSolutions_++; |
---|
3834 | } else { |
---|
3835 | heuristicValue = saveValue ; |
---|
3836 | } |
---|
3837 | } |
---|
3838 | // Restore number solutions |
---|
3839 | numberSolutions_ = saveNumberSolutions; |
---|
3840 | numberHeuristicSolutions_ = saveNumberHeuristicSolutions; |
---|
3841 | } |
---|
3842 | // Add in any violated saved cuts |
---|
3843 | if (!theseCuts.sizeRowCuts()&&!theseCuts.sizeColCuts()) { |
---|
3844 | int numberOld = theseCuts.sizeRowCuts(); |
---|
3845 | int numberCuts = slackCuts.sizeRowCuts() ; |
---|
3846 | int i; |
---|
3847 | // possibly extend whichGenerator |
---|
3848 | resizeWhichGenerator(numberOld, numberOld+numberCuts); |
---|
3849 | for ( i = 0;i<numberCuts;i++) { |
---|
3850 | const OsiRowCut * thisCut = slackCuts.rowCutPtr(i) ; |
---|
3851 | if (thisCut->violated(cbcColSolution_)>100.0*primalTolerance) { |
---|
3852 | if (messageHandler()->logLevel()>2) |
---|
3853 | printf("Old cut added - violation %g\n", |
---|
3854 | thisCut->violated(cbcColSolution_)) ; |
---|
3855 | whichGenerator_[numberOld++]=-1; |
---|
3856 | theseCuts.insert(*thisCut) ; |
---|
3857 | } |
---|
3858 | } |
---|
3859 | } |
---|
3860 | /* |
---|
3861 | End of the loop to exercise each generator/heuristic. |
---|
3862 | |
---|
3863 | Did any of the heuristics turn up a new solution? Record it before we free |
---|
3864 | the vector. |
---|
3865 | */ |
---|
3866 | if (found >= 0) { |
---|
3867 | phase_=4; |
---|
3868 | incrementUsed(newSolution); |
---|
3869 | setBestSolution(CBC_ROUNDING,heuristicValue,newSolution) ; |
---|
3870 | lastHeuristic_ = heuristic_[found]; |
---|
3871 | CbcTreeLocal * tree |
---|
3872 | = dynamic_cast<CbcTreeLocal *> (tree_); |
---|
3873 | if (tree) |
---|
3874 | tree->passInSolution(bestSolution_,heuristicValue); |
---|
3875 | } |
---|
3876 | delete [] newSolution ; |
---|
3877 | |
---|
3878 | #if 0 |
---|
3879 | // switch on to get all cuts printed |
---|
3880 | theseCuts.printCuts() ; |
---|
3881 | #endif |
---|
3882 | int numberColumnCuts = theseCuts.sizeColCuts() ; |
---|
3883 | int numberRowCuts = theseCuts.sizeRowCuts() ; |
---|
3884 | if (violated>=0) |
---|
3885 | violated = numberRowCuts + numberColumnCuts ; |
---|
3886 | /* |
---|
3887 | Apply column cuts (aka bound tightening). This may be partially redundant |
---|
3888 | for column cuts returned by CglProbing, as generateCuts installs bounds |
---|
3889 | from CglProbing when it determines it can fix a variable. |
---|
3890 | |
---|
3891 | TODO: Looks like the use of violated has evolved. The value set above is |
---|
3892 | completely ignored. All that's left is violated == -1 indicates some |
---|
3893 | cut is violated, violated == -2 indicates infeasibility. Only |
---|
3894 | infeasibility warrants exceptional action. |
---|
3895 | |
---|
3896 | TODO: Strikes me that this code will fail to detect infeasibility, because |
---|
3897 | the breaks escape the inner loops but the outer loop keeps going. |
---|
3898 | Infeasibility in an early cut will be overwritten if a later cut is |
---|
3899 | merely violated. |
---|
3900 | */ |
---|
3901 | if (numberColumnCuts) { |
---|
3902 | |
---|
3903 | #ifdef CBC_DEBUG |
---|
3904 | double * oldLower = new double [numberColumns] ; |
---|
3905 | double * oldUpper = new double [numberColumns] ; |
---|
3906 | memcpy(oldLower,cbcColLower_,numberColumns*sizeof(double)) ; |
---|
3907 | memcpy(oldUpper,cbcColUpper_,numberColumns*sizeof(double)) ; |
---|
3908 | #endif |
---|
3909 | |
---|
3910 | double integerTolerance = getDblParam(CbcIntegerTolerance) ; |
---|
3911 | for (int i = 0;i<numberColumnCuts;i++) { |
---|
3912 | const OsiColCut * thisCut = theseCuts.colCutPtr(i) ; |
---|
3913 | const CoinPackedVector & lbs = thisCut->lbs() ; |
---|
3914 | const CoinPackedVector & ubs = thisCut->ubs() ; |
---|
3915 | int j ; |
---|
3916 | int n ; |
---|
3917 | const int * which ; |
---|
3918 | const double * values ; |
---|
3919 | n = lbs.getNumElements() ; |
---|
3920 | which = lbs.getIndices() ; |
---|
3921 | values = lbs.getElements() ; |
---|
3922 | for (j = 0;j<n;j++) { |
---|
3923 | int iColumn = which[j] ; |
---|
3924 | double value = cbcColSolution_[iColumn] ; |
---|
3925 | #if CBC_DEBUG>1 |
---|
3926 | printf("%d %g %g %g %g\n",iColumn,oldLower[iColumn], |
---|
3927 | cbcColSolution_[iColumn],oldUpper[iColumn],values[j]) ; |
---|
3928 | #endif |
---|
3929 | solver_->setColLower(iColumn,values[j]) ; |
---|
3930 | if (value<values[j]-integerTolerance) |
---|
3931 | violated = -1 ; |
---|
3932 | if (values[j]>cbcColUpper_[iColumn]+integerTolerance) { |
---|
3933 | // infeasible |
---|
3934 | violated = -2 ; |
---|
3935 | break ; |
---|
3936 | } |
---|
3937 | } |
---|
3938 | n = ubs.getNumElements() ; |
---|
3939 | which = ubs.getIndices() ; |
---|
3940 | values = ubs.getElements() ; |
---|
3941 | for (j = 0;j<n;j++) { |
---|
3942 | int iColumn = which[j] ; |
---|
3943 | double value = cbcColSolution_[iColumn] ; |
---|
3944 | #if CBC_DEBUG>1 |
---|
3945 | printf("%d %g %g %g %g\n",iColumn,oldLower[iColumn], |
---|
3946 | cbcColSolution_[iColumn],oldUpper[iColumn],values[j]) ; |
---|
3947 | #endif |
---|
3948 | solver_->setColUpper(iColumn,values[j]) ; |
---|
3949 | if (value>values[j]+integerTolerance) |
---|
3950 | violated = -1 ; |
---|
3951 | if (values[j]<cbcColLower_[iColumn]-integerTolerance) { |
---|
3952 | // infeasible |
---|
3953 | violated = -2 ; |
---|
3954 | break ; |
---|
3955 | } |
---|
3956 | } |
---|
3957 | } |
---|
3958 | #ifdef CBC_DEBUG |
---|
3959 | delete [] oldLower ; |
---|
3960 | delete [] oldUpper ; |
---|
3961 | #endif |
---|
3962 | } |
---|
3963 | /* |
---|
3964 | End installation of column cuts. The break here escapes the numberTries |
---|
3965 | loop. |
---|
3966 | */ |
---|
3967 | if (violated == -2||!feasible) { |
---|
3968 | // infeasible |
---|
3969 | feasible = false ; |
---|
3970 | violated = -2; |
---|
3971 | if (!numberNodes_) |
---|
3972 | messageHandler()->message(CBC_INFEAS, |
---|
3973 | messages()) |
---|
3974 | << CoinMessageEol ; |
---|
3975 | break ; |
---|
3976 | } |
---|
3977 | /* |
---|
3978 | Now apply the row (constraint) cuts. This is a bit more work because we need |
---|
3979 | to obtain and augment the current basis. |
---|
3980 | |
---|
3981 | TODO: Why do this work, if there are no row cuts? The current basis will do |
---|
3982 | just fine. |
---|
3983 | */ |
---|
3984 | int numberRowsNow = solver_->getNumRows() ; |
---|
3985 | assert(numberRowsNow == numberRowsAtStart+lastNumberCuts) ; |
---|
3986 | int numberToAdd = theseCuts.sizeRowCuts() ; |
---|
3987 | numberNewCuts_ = lastNumberCuts+numberToAdd ; |
---|
3988 | /* |
---|
3989 | Now actually add the row cuts and reoptimise. |
---|
3990 | |
---|
3991 | Install the cuts in the solver using applyRowCuts and |
---|
3992 | augment the basis with the corresponding slack. We also add each row cut to |
---|
3993 | the set of row cuts (cuts.insert()) supplied as a parameter. The new basis |
---|
3994 | must be set with setWarmStart(). |
---|
3995 | |
---|
3996 | TODO: Seems to me the original code could allocate addCuts with size 0, if |
---|
3997 | numberRowCuts was 0 and numberColumnCuts was nonzero. That might |
---|
3998 | explain the memory fault noted in the comment by AJK. Unfortunately, |
---|
3999 | just commenting out the delete[] results in massive memory leaks. Try |
---|
4000 | a revision to separate the row cut case. Why do we need addCuts at |
---|
4001 | all? A typing issue, apparently: OsiCut vs. OsiRowCut. |
---|
4002 | |
---|
4003 | TODO: It looks to me as if numberToAdd and numberRowCuts are identical at |
---|
4004 | this point. Confirm & get rid of one of them. |
---|
4005 | |
---|
4006 | TODO: Any reason why the three loops can't be consolidated? |
---|
4007 | */ |
---|
4008 | if (numberRowCuts > 0 || numberColumnCuts > 0) |
---|
4009 | { if (numberToAdd > 0) |
---|
4010 | { int i ; |
---|
4011 | // Faster to add all at once |
---|
4012 | const OsiRowCut ** addCuts = new const OsiRowCut * [numberToAdd] ; |
---|
4013 | for (i = 0 ; i < numberToAdd ; i++) |
---|
4014 | { addCuts[i] = &theseCuts.rowCut(i) ; } |
---|
4015 | solver_->applyRowCuts(numberToAdd,addCuts) ; |
---|
4016 | // AJK this caused a memory fault on Win32 |
---|
4017 | // may be okay now with ** form |
---|
4018 | delete [] addCuts ; |
---|
4019 | for (i = 0 ; i < numberToAdd ; i++) |
---|
4020 | { cuts.insert(theseCuts.rowCut(i)) ; } |
---|
4021 | CoinWarmStartBasis * basis = dynamic_cast<CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
4022 | assert(basis != NULL); // make sure not volume |
---|
4023 | /* dylp bug |
---|
4024 | |
---|
4025 | Consistent size used by OsiDylp as sanity check. Implicit resize seen |
---|
4026 | as an error. Hence this call to resize is necessary. |
---|
4027 | */ |
---|
4028 | basis->resize(numberRowsAtStart+numberNewCuts_,numberColumns) ; |
---|
4029 | for (i = 0 ; i < numberToAdd ; i++) |
---|
4030 | { basis->setArtifStatus(numberRowsNow+i, |
---|
4031 | CoinWarmStartBasis::basic) ; } |
---|
4032 | if (solver_->setWarmStart(basis) == false) |
---|
4033 | { throw CoinError("Fail setWarmStart() after cut installation.", |
---|
4034 | "solveWithCuts","CbcModel") ; } |
---|
4035 | delete basis; |
---|
4036 | } |
---|
4037 | feasible = resolve(node ? node->nodeInfo() : NULL,2) ; |
---|
4038 | if ( CoinCpuTime()-dblParam_[CbcStartSeconds] > dblParam_[CbcMaximumSeconds] ) |
---|
4039 | numberTries=0; // exit |
---|
4040 | # ifdef CBC_DEBUG |
---|
4041 | printf("Obj value after cuts %g %d rows\n",solver_->getObjValue(), |
---|
4042 | solver_->getNumRows()) ; |
---|
4043 | if (onOptimalPath && !solver_->isDualObjectiveLimitReached()) |
---|
4044 | assert(feasible) ; |
---|
4045 | # endif |
---|
4046 | } |
---|
4047 | /* |
---|
4048 | No cuts. Cut short the cut generation (numberTries) loop. |
---|
4049 | */ |
---|
4050 | else |
---|
4051 | { numberTries = 0 ; } |
---|
4052 | /* |
---|
4053 | If the problem is still feasible, first, call takeOffCuts() to remove cuts |
---|
4054 | that are now slack. takeOffCuts() will call the solver to reoptimise if |
---|
4055 | that's needed to restore a valid solution. |
---|
4056 | |
---|
4057 | Next, see if we should quit due to diminishing returns: |
---|
4058 | * we've tried three rounds of cut generation and we're getting |
---|
4059 | insufficient improvement in the objective; or |
---|
4060 | * we generated no cuts; or |
---|
4061 | * the solver declared optimality with 0 iterations after we added the |
---|
4062 | cuts generated in this round. |
---|
4063 | If we decide to keep going, prep for the next iteration. |
---|
4064 | |
---|
4065 | It just seems more safe to tell takeOffCuts() to call resolve(), even if |
---|
4066 | we're not continuing cut generation. Otherwise code executed between here |
---|
4067 | and final disposition of the node will need to be careful not to access the |
---|
4068 | lp solution. It can happen that we lose feasibility in takeOffCuts --- |
---|
4069 | numerical jitters when the cutoff bound is epsilon less than the current |
---|
4070 | best, and we're evaluating an alternative optimum. |
---|
4071 | |
---|
4072 | TODO: After successive rounds of code motion, there seems no need to |
---|
4073 | distinguish between the two checks for aborting the cut generation |
---|
4074 | loop. Confirm and clean up. |
---|
4075 | */ |
---|
4076 | if (feasible) |
---|
4077 | { int cutIterations = solver_->getIterationCount() ; |
---|
4078 | if (numberOldActiveCuts_+numberNewCuts_) { |
---|
4079 | OsiCuts * saveCuts = node ? NULL : &slackCuts; |
---|
4080 | takeOffCuts(cuts,resolveAfterTakeOffCuts_,saveCuts) ; |
---|
4081 | if (solver_->isDualObjectiveLimitReached()&&resolveAfterTakeOffCuts_) |
---|
4082 | { feasible = false ; |
---|
4083 | # ifdef CBC_DEBUG |
---|
4084 | double z = solver_->getObjValue() ; |
---|
4085 | double cut = getCutoff() ; |
---|
4086 | printf("Lost feasibility by %g in takeOffCuts; z = %g, cutoff = %g\n", |
---|
4087 | z-cut,z,cut) ; |
---|
4088 | # endif |
---|
4089 | } |
---|
4090 | } |
---|
4091 | if (feasible) { |
---|
4092 | numberRowsAtStart = numberOldActiveCuts_+numberRowsAtContinuous_ ; |
---|
4093 | lastNumberCuts = numberNewCuts_ ; |
---|
4094 | if (direction*solver_->getObjValue() < lastObjective+minimumDrop && |
---|
4095 | currentPassNumber_ >= 3) |
---|
4096 | { numberTries = 0 ; } |
---|
4097 | if (numberRowCuts+numberColumnCuts == 0 || cutIterations == 0) |
---|
4098 | { break ; } |
---|
4099 | if (numberTries > 0) { |
---|
4100 | reducedCostFix() ; |
---|
4101 | lastObjective = direction*solver_->getObjValue() ; |
---|
4102 | } |
---|
4103 | } |
---|
4104 | } |
---|
4105 | /* |
---|
4106 | We've lost feasibility --- this node won't be referencing the cuts we've |
---|
4107 | been collecting, so decrement the reference counts. |
---|
4108 | */ |
---|
4109 | if (!feasible) |
---|
4110 | { int i ; |
---|
4111 | for (i = 0;i<currentNumberCuts_;i++) { |
---|
4112 | // take off node |
---|
4113 | if (addedCuts_[i]) { |
---|
4114 | if (!addedCuts_[i]->decrement()) |
---|
4115 | delete addedCuts_[i] ; |
---|
4116 | addedCuts_[i] = NULL ; |
---|
4117 | } |
---|
4118 | } |
---|
4119 | numberTries = 0 ; |
---|
4120 | } |
---|
4121 | } while (numberTries>0) ; |
---|
4122 | //check feasibility. |
---|
4123 | //If solution seems to be integer feasible calling setBestSolution |
---|
4124 | //will eventually add extra global cuts which we need to install at |
---|
4125 | //the nodes |
---|
4126 | |
---|
4127 | |
---|
4128 | if(feasible&&solverCharacteristics_->solutionAddsCuts()) //check integer feasibility |
---|
4129 | { |
---|
4130 | bool integerFeasible = true; |
---|
4131 | const double * save = testSolution_; |
---|
4132 | testSolution_ = solver_->getColSolution(); |
---|
4133 | for (int i=0;i<numberObjects_ && integerFeasible;i++) |
---|
4134 | { |
---|
4135 | int preferredWay; |
---|
4136 | double infeasibility = object_[i]->infeasibility(preferredWay); |
---|
4137 | if(infeasibility) |
---|
4138 | integerFeasible = false; |
---|
4139 | } |
---|
4140 | testSolution_ = save; |
---|
4141 | if(integerFeasible)//update |
---|
4142 | { |
---|
4143 | double objValue = solver_->getObjValue(); |
---|
4144 | //int numberGlobalBefore = globalCuts_.sizeRowCuts(); |
---|
4145 | // SOLUTION2 so won't up cutoff or print message |
---|
4146 | setBestSolution(CBC_SOLUTION2, objValue, |
---|
4147 | solver_->getColSolution(),0); |
---|
4148 | #if 0 |
---|
4149 | int numberGlobalAfter = globalCuts_.sizeRowCuts(); |
---|
4150 | int numberToAdd = numberGlobalAfter - numberGlobalBefore; |
---|
4151 | if(numberToAdd > 0) |
---|
4152 | //We have added some cuts say they are tight at that node |
---|
4153 | //Basis and lp should already have been updated |
---|
4154 | { |
---|
4155 | feasible = (solver_->isProvenOptimal() && |
---|
4156 | !solver_->isDualObjectiveLimitReached()) ; |
---|
4157 | if(feasible) |
---|
4158 | { |
---|
4159 | int numberCuts = numberNewCuts_ = cuts.sizeRowCuts(); |
---|
4160 | // possibly extend whichGenerator |
---|
4161 | resizeWhichGenerator(numberCuts, numberToAdd+numberCuts); |
---|
4162 | |
---|
4163 | for (int i = numberGlobalBefore ; i < numberGlobalAfter ; i++) |
---|
4164 | { |
---|
4165 | whichGenerator_[numberNewCuts_++]=-1; |
---|
4166 | cuts.insert(globalCuts_.rowCut(i)) ; |
---|
4167 | } |
---|
4168 | numberNewCuts_ = lastNumberCuts+numberToAdd; |
---|
4169 | //now take off the cuts which are not tight anymore |
---|
4170 | takeOffCuts(cuts,resolveAfterTakeOffCuts_, NULL) ; |
---|
4171 | if (solver_->isDualObjectiveLimitReached()&&resolveAfterTakeOffCuts_) |
---|
4172 | { feasible = false ;} |
---|
4173 | } |
---|
4174 | if(!feasible) //node will be fathomed |
---|
4175 | { |
---|
4176 | for (int i = 0;i<currentNumberCuts_;i++) |
---|
4177 | { |
---|
4178 | // take off node |
---|
4179 | if (addedCuts_[i]) |
---|
4180 | { |
---|
4181 | if (!addedCuts_[i]->decrement()) |
---|
4182 | delete addedCuts_[i] ; |
---|
4183 | addedCuts_[i] = NULL ; |
---|
4184 | } |
---|
4185 | } |
---|
4186 | } |
---|
4187 | } |
---|
4188 | #endif |
---|
4189 | } |
---|
4190 | } |
---|
4191 | // Reduced cost fix at end |
---|
4192 | if (solver_->isProvenOptimal()) |
---|
4193 | reducedCostFix(); |
---|
4194 | // If at root node do heuristics |
---|
4195 | if (!numberNodes_) { |
---|
4196 | // First see if any cuts are slack |
---|
4197 | int numberRows = solver_->getNumRows(); |
---|
4198 | int numberAdded = numberRows-numberRowsAtContinuous_; |
---|
4199 | if (numberAdded) { |
---|
4200 | CoinWarmStartBasis * basis = dynamic_cast<CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
4201 | assert(basis != NULL); |
---|
4202 | int * added = new int[numberAdded]; |
---|
4203 | int nDelete = 0; |
---|
4204 | for (int j=numberRowsAtContinuous_;j<numberRows;j++) { |
---|
4205 | if (basis->getArtifStatus(j)==CoinWarmStartBasis::basic) { |
---|
4206 | //printf("%d slack!\n",j); |
---|
4207 | added[nDelete++]=j; |
---|
4208 | } |
---|
4209 | } |
---|
4210 | if (nDelete) { |
---|
4211 | solver_->deleteRows(nDelete,added); |
---|
4212 | } |
---|
4213 | delete [] added; |
---|
4214 | delete basis ; |
---|
4215 | } |
---|
4216 | // mark so heuristics can tell |
---|
4217 | int savePass=currentPassNumber_; |
---|
4218 | currentPassNumber_=999999; |
---|
4219 | double * newSolution = new double [numberColumns] ; |
---|
4220 | double heuristicValue = getCutoff() ; |
---|
4221 | int found = -1; // no solution found |
---|
4222 | for (int i = 0;i<numberHeuristics_;i++) { |
---|
4223 | // see if heuristic will do anything |
---|
4224 | double saveValue = heuristicValue ; |
---|
4225 | int ifSol = heuristic_[i]->solution(heuristicValue, |
---|
4226 | newSolution); |
---|
4227 | if (ifSol>0) { |
---|
4228 | // better solution found |
---|
4229 | found = i ; |
---|
4230 | incrementUsed(newSolution); |
---|
4231 | } else { |
---|
4232 | heuristicValue = saveValue ; |
---|
4233 | } |
---|
4234 | } |
---|
4235 | currentPassNumber_=savePass; |
---|
4236 | if (found >= 0) { |
---|
4237 | phase_=4; |
---|
4238 | incrementUsed(newSolution); |
---|
4239 | setBestSolution(CBC_ROUNDING,heuristicValue,newSolution) ; |
---|
4240 | lastHeuristic_ = heuristic_[found]; |
---|
4241 | } |
---|
4242 | delete [] newSolution ; |
---|
4243 | } |
---|
4244 | // Up change due to cuts |
---|
4245 | if (feasible) |
---|
4246 | sumChangeObjective2_ += solver_->getObjValue()*solver_->getObjSense() |
---|
4247 | - objectiveValue ; |
---|
4248 | //if ((numberNodes_%100)==0) |
---|
4249 | //printf("XXb sum obj changed by %g\n",sumChangeObjective2_); |
---|
4250 | /* |
---|
4251 | End of cut generation loop. |
---|
4252 | |
---|
4253 | Now, consider if we want to disable or adjust the frequency of use for any |
---|
4254 | of the cut generators. If the client specified a positive number for |
---|
4255 | howOften, it will never change. If the original value was negative, it'll |
---|
4256 | be converted to 1000000+|howOften|, and this value will be adjusted each |
---|
4257 | time fullScan is true. Actual cut generation is performed every |
---|
4258 | howOften%1000000 nodes; the 1000000 offset is just a convenient way to |
---|
4259 | specify that the frequency is adjustable. |
---|
4260 | |
---|
4261 | During cut generation, we recorded the number of cuts produced by each |
---|
4262 | generator for this node. For all cuts, whichGenerator records the generator |
---|
4263 | that produced a cut. |
---|
4264 | |
---|
4265 | TODO: All this should probably be hidden in a method of the CbcCutGenerator |
---|
4266 | class. |
---|
4267 | |
---|
4268 | TODO: Can the loop that scans over whichGenerator to accumulate per generator |
---|
4269 | counts be replaced by values in countRowCuts and countColumnCuts? |
---|
4270 | */ |
---|
4271 | #ifdef NODE_LOG |
---|
4272 | int fatherNum = (node == NULL) ? -1 : node->nodeNumber(); |
---|
4273 | double value = (node == NULL) ? -1 : node->branchingObject()->value(); |
---|
4274 | string bigOne = (solver_->getIterationCount() > 30) ? "*******" : ""; |
---|
4275 | string way = (node == NULL) ? "" : (node->branchingObject()->way())==1 ? "Down" : "Up"; |
---|
4276 | std::cout<<"Node "<<numberNodes_<<", father "<<fatherNum<<", #iterations "<<solver_->getIterationCount()<<", sol value : "<<solver_->getObjValue()<<std::endl; |
---|
4277 | #endif |
---|
4278 | if (fullScan&&numberCutGenerators_) { |
---|
4279 | /* If cuts just at root node then it will probably be faster to |
---|
4280 | update matrix and leave all in */ |
---|
4281 | int willBeCutsInTree=0; |
---|
4282 | double thisObjective = solver_->getObjValue()*direction ; |
---|
4283 | // get sizes |
---|
4284 | int numberRowsAdded = solver_->getNumRows()-numberRowsAtStart; |
---|
4285 | CoinBigIndex numberElementsAdded = solver_->getNumElements()-numberElementsAtStart ; |
---|
4286 | double densityOld = ((double) numberElementsAtStart)/((double) numberRowsAtStart); |
---|
4287 | double densityNew = numberRowsAdded ? ((double) (numberElementsAdded))/((double) numberRowsAdded) |
---|
4288 | : 0.0; |
---|
4289 | if (!numberNodes_) { |
---|
4290 | if (numberRowsAdded) |
---|
4291 | handler_->message(CBC_CUTS_STATS,messages_) |
---|
4292 | <<numberRowsAdded |
---|
4293 | <<densityNew |
---|
4294 | <<CoinMessageEol ; |
---|
4295 | if (thisObjective-startObjective<1.0e-5&&numberElementsAdded>0.2*numberElementsAtStart) |
---|
4296 | willBeCutsInTree=-1; |
---|
4297 | } |
---|
4298 | if ((numberRowsAdded>100+0.5*numberRowsAtStart |
---|
4299 | ||numberElementsAdded>0.5*numberElementsAtStart) |
---|
4300 | &&(densityNew>200.0&&numberRowsAdded>100&&densityNew>2.0*densityOld)) { |
---|
4301 | // much bigger |
---|
4302 | //if (thisObjective-startObjective<0.1*fabs(startObjective)+1.0e-5) |
---|
4303 | //willBeCutsInTree=-1; |
---|
4304 | //printf("Cuts will be taken off , %d rows added with density %g\n", |
---|
4305 | // numberRowsAdded,densityNew); |
---|
4306 | } |
---|
4307 | if (densityNew>100.0&&numberRowsAdded>2&&densityNew>2.0*densityOld) { |
---|
4308 | //if (thisObjective-startObjective<0.1*fabs(startObjective)+1.0e-5) |
---|
4309 | //willBeCutsInTree=-2; |
---|
4310 | //printf("Density says no cuts ? , %d rows added with density %g\n", |
---|
4311 | // numberRowsAdded,densityNew); |
---|
4312 | } |
---|
4313 | // Root node or every so often - see what to turn off |
---|
4314 | int i ; |
---|
4315 | for (i = 0;i<numberCutGenerators_;i++) { |
---|
4316 | int howOften = generator_[i]->howOften() ; |
---|
4317 | if (howOften>-90) |
---|
4318 | willBeCutsInTree=0; |
---|
4319 | } |
---|
4320 | if (!numberNodes_) |
---|
4321 | handler_->message(CBC_ROOT,messages_) |
---|
4322 | <<numberNewCuts_ |
---|
4323 | <<startObjective<<thisObjective |
---|
4324 | <<currentPassNumber_ |
---|
4325 | <<CoinMessageEol ; |
---|
4326 | int * count = new int[numberCutGenerators_] ; |
---|
4327 | memset(count,0,numberCutGenerators_*sizeof(int)) ; |
---|
4328 | int numberActiveGenerators=0; |
---|
4329 | for (i = 0;i<numberNewCuts_;i++) { |
---|
4330 | int iGenerator = whichGenerator_[i]; |
---|
4331 | if (iGenerator>=0&&iGenerator<numberCutGenerators_) |
---|
4332 | count[iGenerator]++ ; |
---|
4333 | } |
---|
4334 | double totalCuts = 0.0 ; |
---|
4335 | //#define JUST_ACTIVE |
---|
4336 | for (i = 0;i<numberCutGenerators_;i++) { |
---|
4337 | if (countRowCuts[i]||countColumnCuts[i]) |
---|
4338 | numberActiveGenerators++; |
---|
4339 | #ifdef JUST_ACTIVE |
---|
4340 | totalCuts += count[i] + 5.0*countColumnCuts[i] ; |
---|
4341 | #else |
---|
4342 | totalCuts += countRowCuts[i] + 5.0*countColumnCuts[i] ; |
---|
4343 | #endif |
---|
4344 | } |
---|
4345 | double small = (0.5* totalCuts) / |
---|
4346 | ((double) numberActiveGenerators) ; |
---|
4347 | for (i = 0;i<numberCutGenerators_;i++) { |
---|
4348 | int howOften = generator_[i]->howOften() ; |
---|
4349 | if (willBeCutsInTree<0&&howOften==-98) |
---|
4350 | howOften =-99; |
---|
4351 | if (howOften==-98&&generator_[i]->switchOffIfLessThan()<0) { |
---|
4352 | if (thisObjective-startObjective<0.005*fabs(startObjective)+1.0e-5) |
---|
4353 | howOften=-99; // switch off |
---|
4354 | if (thisObjective-startObjective<0.1*fabs(startObjective)+1.0e-5 |
---|
4355 | &&5*solver_->getNumRows()<solver_->getNumCols()) |
---|
4356 | howOften=-99; // switch off |
---|
4357 | } |
---|
4358 | if (howOften<-99) |
---|
4359 | continue ; |
---|
4360 | if (howOften<0||howOften >= 1000000) { |
---|
4361 | if( !numberNodes_) { |
---|
4362 | // If small number switch mostly off |
---|
4363 | #ifdef JUST_ACTIVE |
---|
4364 | double thisCuts = count[i] + 5.0*countColumnCuts[i] ; |
---|
4365 | #else |
---|
4366 | double thisCuts = countRowCuts[i] + 5.0*countColumnCuts[i] ; |
---|
4367 | #endif |
---|
4368 | if (!thisCuts||howOften == -99) { |
---|
4369 | if (howOften == -99||howOften == -98) |
---|
4370 | howOften = -100 ; |
---|
4371 | else |
---|
4372 | howOften = 1000000+SCANCUTS; // wait until next time |
---|
4373 | } else if (thisCuts<small) { |
---|
4374 | int k = (int) sqrt(small/thisCuts) ; |
---|
4375 | if (howOften!=-98) |
---|
4376 | howOften = k+1000000 ; |
---|
4377 | else |
---|
4378 | howOften=-100; |
---|
4379 | } else { |
---|
4380 | howOften = 1+1000000 ; |
---|
4381 | if (thisObjective-startObjective<0.1*fabs(startObjective)+1.0e-5) |
---|
4382 | generator_[i]->setWhatDepth(5); |
---|
4383 | } |
---|
4384 | } |
---|
4385 | // If cuts useless switch off |
---|
4386 | if (numberNodes_>=100000&&sumChangeObjective1_>2.0e2*(sumChangeObjective2_+1.0e-12)) { |
---|
4387 | howOften = 1000000+SCANCUTS; // wait until next time |
---|
4388 | //printf("switch off cut %d due to lack of use\n",i); |
---|
4389 | } |
---|
4390 | } |
---|
4391 | if (howOften>=0&&generator_[i]->generator()->mayGenerateRowCutsInTree()) |
---|
4392 | willBeCutsInTree=1; |
---|
4393 | |
---|
4394 | generator_[i]->setHowOften(howOften) ; |
---|
4395 | if (howOften>=1000000&&howOften<2000000&&0) { |
---|
4396 | // Go to depth |
---|
4397 | int bias=1; |
---|
4398 | if (howOften==1+1000000) |
---|
4399 | generator_[i]->setWhatDepth(bias+1); |
---|
4400 | else if (howOften<=10+1000000) |
---|
4401 | generator_[i]->setWhatDepth(bias+2); |
---|
4402 | else |
---|
4403 | generator_[i]->setWhatDepth(bias+1000); |
---|
4404 | } |
---|
4405 | int newFrequency = generator_[i]->howOften()%1000000 ; |
---|
4406 | // increment cut counts |
---|
4407 | generator_[i]->incrementNumberCutsInTotal(countRowCuts[i]); |
---|
4408 | generator_[i]->incrementNumberCutsActive(count[i]); |
---|
4409 | if (handler_->logLevel()>1||!numberNodes_) { |
---|
4410 | handler_->message(CBC_GENERATOR,messages_) |
---|
4411 | <<i |
---|
4412 | <<generator_[i]->cutGeneratorName() |
---|
4413 | <<countRowCuts[i]<<count[i] |
---|
4414 | <<countColumnCuts[i]; |
---|
4415 | handler_->printing(!numberNodes_&&generator_[i]->timing()) |
---|
4416 | <<generator_[i]->timeInCutGenerator(); |
---|
4417 | handler_->message() |
---|
4418 | <<newFrequency |
---|
4419 | <<CoinMessageEol ; |
---|
4420 | } |
---|
4421 | } |
---|
4422 | delete [] count ; |
---|
4423 | if( !numberNodes_) { |
---|
4424 | if (willBeCutsInTree==-2) |
---|
4425 | willBeCutsInTree=0; |
---|
4426 | if( willBeCutsInTree<=0) { |
---|
4427 | // Take off cuts |
---|
4428 | cuts = OsiCuts(); |
---|
4429 | numberNewCuts_=0; |
---|
4430 | if (!willBeCutsInTree) { |
---|
4431 | // update size of problem |
---|
4432 | numberRowsAtContinuous_ = solver_->getNumRows() ; |
---|
4433 | } else { |
---|
4434 | // take off cuts |
---|
4435 | int numberRows = solver_->getNumRows(); |
---|
4436 | int numberAdded = numberRows-numberRowsAtContinuous_; |
---|
4437 | if (numberAdded) { |
---|
4438 | int * added = new int[numberAdded]; |
---|
4439 | for (int i=0;i<numberAdded;i++) |
---|
4440 | added[i]=i+numberRowsAtContinuous_; |
---|
4441 | solver_->deleteRows(numberAdded,added); |
---|
4442 | delete [] added; |
---|
4443 | // resolve so optimal |
---|
4444 | solver_->resolve(); |
---|
4445 | } |
---|
4446 | } |
---|
4447 | #ifdef CBC_USE_CLP |
---|
4448 | OsiClpSolverInterface * clpSolver |
---|
4449 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
4450 | if (clpSolver) { |
---|
4451 | // Maybe solver might like to know only column bounds will change |
---|
4452 | //int options = clpSolver->specialOptions(); |
---|
4453 | //clpSolver->setSpecialOptions(options|128); |
---|
4454 | clpSolver->synchronizeModel(); |
---|
4455 | } |
---|
4456 | #endif |
---|
4457 | } else { |
---|
4458 | #ifdef CBC_USE_CLP |
---|
4459 | OsiClpSolverInterface * clpSolver |
---|
4460 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
4461 | if (clpSolver) { |
---|
4462 | // make sure factorization can't carry over |
---|
4463 | int options = clpSolver->specialOptions(); |
---|
4464 | clpSolver->setSpecialOptions(options&(~8)); |
---|
4465 | } |
---|
4466 | #endif |
---|
4467 | } |
---|
4468 | } |
---|
4469 | } else if (numberCutGenerators_) { |
---|
4470 | int i; |
---|
4471 | // add to counts anyway |
---|
4472 | for (i = 0;i<numberCutGenerators_;i++) |
---|
4473 | generator_[i]->incrementNumberCutsInTotal(countRowCuts[i]); |
---|
4474 | // What if not feasible as cuts may have helped |
---|
4475 | if (feasible) { |
---|
4476 | for (i = 0;i<numberNewCuts_;i++) { |
---|
4477 | int iGenerator = whichGenerator_[i]; |
---|
4478 | if (iGenerator>=0) |
---|
4479 | generator_[iGenerator]->incrementNumberCutsActive(); |
---|
4480 | } |
---|
4481 | } |
---|
4482 | } |
---|
4483 | |
---|
4484 | delete [] countRowCuts ; |
---|
4485 | delete [] countColumnCuts ; |
---|
4486 | |
---|
4487 | |
---|
4488 | #ifdef CHECK_CUT_COUNTS |
---|
4489 | if (feasible) |
---|
4490 | { |
---|
4491 | CoinWarmStartBasis * basis = dynamic_cast<CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
4492 | printf("solveWithCuts: Number of rows at end (only active cuts) %d\n", |
---|
4493 | numberRowsAtContinuous_+numberNewCuts_+numberOldActiveCuts_) ; |
---|
4494 | basis->print() ; delete basis;} |
---|
4495 | #endif |
---|
4496 | #ifdef CBC_DEBUG |
---|
4497 | if (onOptimalPath && !solver_->isDualObjectiveLimitReached()) |
---|
4498 | assert(feasible) ; |
---|
4499 | #endif |
---|
4500 | |
---|
4501 | return feasible ; } |
---|
4502 | |
---|
4503 | |
---|
4504 | /* |
---|
4505 | Remove slack cuts. We obtain a basis and scan it. Cuts with basic slacks |
---|
4506 | are purged. If any cuts are purged, resolve() is called to restore the |
---|
4507 | solution held in the solver. If resolve() pivots, there's the possibility |
---|
4508 | that a slack may be pivoted in (trust me :-), so the process iterates. |
---|
4509 | Setting allowResolve to false will suppress reoptimisation (but see note |
---|
4510 | below). |
---|
4511 | |
---|
4512 | At the level of the solver's constraint system, loose cuts are really |
---|
4513 | deleted. There's an implicit assumption that deleteRows will also update |
---|
4514 | the active basis in the solver. |
---|
4515 | |
---|
4516 | At the level of nodes and models, it's more complicated. |
---|
4517 | |
---|
4518 | New cuts exist only in the collection of cuts passed as a parameter. They |
---|
4519 | are deleted from the collection and that's the end of them. |
---|
4520 | |
---|
4521 | Older cuts have made it into addedCuts_. Two separate actions are needed. |
---|
4522 | The reference count for the CbcCountRowCut object is decremented. If this |
---|
4523 | count falls to 0, the node which owns the cut is located, the reference to |
---|
4524 | the cut is removed, and then the cut object is destroyed (courtesy of the |
---|
4525 | CbcCountRowCut destructor). We also need to set the addedCuts_ entry to |
---|
4526 | NULL. This is important so that when it comes time to generate basis edits |
---|
4527 | we can tell this cut was dropped from the basis during processing of the |
---|
4528 | node. |
---|
4529 | |
---|
4530 | NOTE: In general, it's necessary to call resolve() after purging slack |
---|
4531 | cuts. Deleting constraints constitutes a change in the problem, and |
---|
4532 | an OSI is not required to maintain a valid solution when the problem |
---|
4533 | is changed. But ... it's really useful to maintain the active basis, |
---|
4534 | and the OSI is supposed to do that. (Yes, it's splitting hairs.) In |
---|
4535 | some places, it's possible to know that the solution will never be |
---|
4536 | consulted after this call, only the basis. (E.g., this routine is |
---|
4537 | called as a last act before generating info to place the node in the |
---|
4538 | live set.) For such use, set allowResolve to false. |
---|
4539 | |
---|
4540 | TODO: No real harm would be done if we just ignored the rare occasion when |
---|
4541 | the call to resolve() pivoted a slack back into the basis. It's a |
---|
4542 | minor inefficiency, at worst. But it does break assertions which |
---|
4543 | check that there are no loose cuts in the basis. It might be better |
---|
4544 | to remove the assertions. |
---|
4545 | */ |
---|
4546 | |
---|
4547 | void |
---|
4548 | CbcModel::takeOffCuts (OsiCuts &newCuts, |
---|
4549 | bool allowResolve, OsiCuts * saveCuts) |
---|
4550 | |
---|
4551 | { // int resolveIterations = 0 ; |
---|
4552 | int firstOldCut = numberRowsAtContinuous_ ; |
---|
4553 | int totalNumberCuts = numberNewCuts_+numberOldActiveCuts_ ; |
---|
4554 | int *solverCutIndices = new int[totalNumberCuts] ; |
---|
4555 | int *newCutIndices = new int[numberNewCuts_] ; |
---|
4556 | const CoinWarmStartBasis* ws ; |
---|
4557 | CoinWarmStartBasis::Status status ; |
---|
4558 | bool needPurge = true ; |
---|
4559 | /* |
---|
4560 | The outer loop allows repetition of purge in the event that reoptimisation |
---|
4561 | changes the basis. To start an iteration, clear the deletion counts and grab |
---|
4562 | the current basis. |
---|
4563 | */ |
---|
4564 | while (needPurge) |
---|
4565 | { int numberNewToDelete = 0 ; |
---|
4566 | int numberOldToDelete = 0 ; |
---|
4567 | int i ; |
---|
4568 | ws = dynamic_cast<const CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
4569 | /* |
---|
4570 | Scan the basis entries of the old cuts generated prior to this round of cut |
---|
4571 | generation. Loose cuts are `removed' by decrementing their reference count |
---|
4572 | and setting the addedCuts_ entry to NULL. (If the reference count falls to |
---|
4573 | 0, they're really deleted. See CbcModel and CbcCountRowCut doc'n for |
---|
4574 | principles of cut handling.) |
---|
4575 | */ |
---|
4576 | int oldCutIndex = 0 ; |
---|
4577 | for (i = 0 ; i < numberOldActiveCuts_ ; i++) |
---|
4578 | { status = ws->getArtifStatus(i+firstOldCut) ; |
---|
4579 | while (!addedCuts_[oldCutIndex]) oldCutIndex++ ; |
---|
4580 | assert(oldCutIndex < currentNumberCuts_) ; |
---|
4581 | // always leave if from nextRowCut_ |
---|
4582 | if (status == CoinWarmStartBasis::basic&& |
---|
4583 | addedCuts_[oldCutIndex]->effectiveness()!=COIN_DBL_MAX) |
---|
4584 | { solverCutIndices[numberOldToDelete++] = i+firstOldCut ; |
---|
4585 | if (saveCuts) { |
---|
4586 | // send to cut pool |
---|
4587 | OsiRowCut * slackCut = addedCuts_[oldCutIndex]; |
---|
4588 | if (slackCut->effectiveness()!=-1.234) { |
---|
4589 | slackCut->setEffectiveness(-1.234); |
---|
4590 | saveCuts->insert(*slackCut); |
---|
4591 | } |
---|
4592 | } |
---|
4593 | if (addedCuts_[oldCutIndex]->decrement() == 0) |
---|
4594 | delete addedCuts_[oldCutIndex] ; |
---|
4595 | addedCuts_[oldCutIndex] = NULL ; |
---|
4596 | oldCutIndex++ ; } |
---|
4597 | else |
---|
4598 | { oldCutIndex++ ; } } |
---|
4599 | /* |
---|
4600 | Scan the basis entries of the new cuts generated with this round of cut |
---|
4601 | generation. At this point, newCuts is the only record of the new cuts, so |
---|
4602 | when we delete loose cuts from newCuts, they're really gone. newCuts is a |
---|
4603 | vector, so it's most efficient to compress it (eraseRowCut) from back to |
---|
4604 | front. |
---|
4605 | */ |
---|
4606 | int firstNewCut = firstOldCut+numberOldActiveCuts_ ; |
---|
4607 | int k = 0 ; |
---|
4608 | for (i = 0 ; i < numberNewCuts_ ; i++) |
---|
4609 | { status = ws->getArtifStatus(i+firstNewCut) ; |
---|
4610 | if (status == CoinWarmStartBasis::basic&&whichGenerator_[i]!=-2) |
---|
4611 | { solverCutIndices[numberNewToDelete+numberOldToDelete] = i+firstNewCut ; |
---|
4612 | newCutIndices[numberNewToDelete++] = i ; } |
---|
4613 | else |
---|
4614 | { // save which generator did it |
---|
4615 | whichGenerator_[k++] = whichGenerator_[i] ; } } |
---|
4616 | delete ws ; |
---|
4617 | for (i = numberNewToDelete-1 ; i >= 0 ; i--) |
---|
4618 | { int iCut = newCutIndices[i] ; |
---|
4619 | if (saveCuts) { |
---|
4620 | // send to cut pool |
---|
4621 | OsiRowCut * slackCut = newCuts.rowCutPtr(iCut); |
---|
4622 | if (slackCut->effectiveness()!=-1.234) { |
---|
4623 | slackCut->setEffectiveness(-1.234); |
---|
4624 | saveCuts->insert(*slackCut); |
---|
4625 | } |
---|
4626 | } |
---|
4627 | newCuts.eraseRowCut(iCut) ; } |
---|
4628 | /* |
---|
4629 | Did we delete anything? If so, delete the cuts from the constraint system |
---|
4630 | held in the solver and reoptimise unless we're forbidden to do so. If the |
---|
4631 | call to resolve() results in pivots, there's the possibility we again have |
---|
4632 | basic slacks. Repeat the purging loop. |
---|
4633 | */ |
---|
4634 | if (numberNewToDelete+numberOldToDelete > 0) |
---|
4635 | { solver_->deleteRows(numberNewToDelete+numberOldToDelete, |
---|
4636 | solverCutIndices) ; |
---|
4637 | numberNewCuts_ -= numberNewToDelete ; |
---|
4638 | numberOldActiveCuts_ -= numberOldToDelete ; |
---|
4639 | # ifdef CBC_DEBUG |
---|
4640 | printf("takeOffCuts: purged %d+%d cuts\n", numberOldToDelete, |
---|
4641 | numberNewToDelete ); |
---|
4642 | # endif |
---|
4643 | if (allowResolve) |
---|
4644 | { |
---|
4645 | phase_=3; |
---|
4646 | // can do quick optimality check |
---|
4647 | int easy=2; |
---|
4648 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,&easy) ; |
---|
4649 | solver_->resolve() ; |
---|
4650 | setPointers(solver_); |
---|
4651 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,NULL) ; |
---|
4652 | if (solver_->getIterationCount() == 0) |
---|
4653 | { needPurge = false ; } |
---|
4654 | # ifdef CBC_DEBUG |
---|
4655 | else |
---|
4656 | { printf( "Repeating purging loop. %d iters.\n", |
---|
4657 | solver_->getIterationCount()); } |
---|
4658 | # endif |
---|
4659 | } |
---|
4660 | else |
---|
4661 | { needPurge = false ; } } |
---|
4662 | else |
---|
4663 | { needPurge = false ; } } |
---|
4664 | /* |
---|
4665 | Clean up and return. |
---|
4666 | */ |
---|
4667 | delete [] solverCutIndices ; |
---|
4668 | delete [] newCutIndices ; |
---|
4669 | } |
---|
4670 | |
---|
4671 | |
---|
4672 | |
---|
4673 | int |
---|
4674 | CbcModel::resolve(CbcNodeInfo * parent, int whereFrom) |
---|
4675 | { |
---|
4676 | // We may have deliberately added in violated cuts - check to avoid message |
---|
4677 | int iRow; |
---|
4678 | int numberRows = solver_->getNumRows(); |
---|
4679 | const double * rowLower = solver_->getRowLower(); |
---|
4680 | const double * rowUpper = solver_->getRowUpper(); |
---|
4681 | bool feasible=true; |
---|
4682 | for (iRow= numberRowsAtContinuous_;iRow<numberRows;iRow++) { |
---|
4683 | if (rowLower[iRow]>rowUpper[iRow]+1.0e-8) |
---|
4684 | feasible=false; |
---|
4685 | } |
---|
4686 | // Can't happen if strong branching as would have been found before |
---|
4687 | if (!numberStrong_&&numberObjects_>numberIntegers_) { |
---|
4688 | int iColumn; |
---|
4689 | int numberColumns = solver_->getNumCols(); |
---|
4690 | const double * columnLower = solver_->getColLower(); |
---|
4691 | const double * columnUpper = solver_->getColUpper(); |
---|
4692 | for (iColumn= 0;iColumn<numberColumns;iColumn++) { |
---|
4693 | if (columnLower[iColumn]>columnUpper[iColumn]+1.0e-5) |
---|
4694 | feasible=false; |
---|
4695 | } |
---|
4696 | } |
---|
4697 | #if 0 |
---|
4698 | { |
---|
4699 | int iColumn; |
---|
4700 | int numberColumns = solver_->getNumCols(); |
---|
4701 | const double * columnLower = solver_->getColLower(); |
---|
4702 | const double * columnUpper = solver_->getColUpper(); |
---|
4703 | const double * sol = solver_->getColSolution(); |
---|
4704 | bool feasible=true; |
---|
4705 | double xx[37]; |
---|
4706 | memset(xx,0,sizeof(xx)); |
---|
4707 | xx[6]=xx[7]=xx[9]=xx[18]=xx[26]=xx[36]=1.0; |
---|
4708 | for (iColumn= 0;iColumn<numberColumns;iColumn++) { |
---|
4709 | if (solver_->isInteger(iColumn)) { |
---|
4710 | //printf("yy %d at %g (bounds %g, %g)\n",iColumn, |
---|
4711 | // sol[iColumn],columnLower[iColumn],columnUpper[iColumn]); |
---|
4712 | if (columnLower[iColumn]>xx[iColumn]|| |
---|
4713 | columnUpper[iColumn]<xx[iColumn]) |
---|
4714 | feasible=false; |
---|
4715 | //solver_->setColLower(iColumn,CoinMin(xx[iColumn],columnUpper[iColumn])); |
---|
4716 | //solver_->setColUpper(iColumn,CoinMax(xx[iColumn],columnLower[iColumn])); |
---|
4717 | } |
---|
4718 | } |
---|
4719 | if (feasible) |
---|
4720 | printf("On Feasible path\n"); |
---|
4721 | } |
---|
4722 | #endif |
---|
4723 | /* |
---|
4724 | Reoptimize. Consider the possibility that we should fathom on bounds. But be |
---|
4725 | careful --- where the objective takes on integral values, we may want to keep |
---|
4726 | a solution where the objective is right on the cutoff. |
---|
4727 | */ |
---|
4728 | if (feasible) |
---|
4729 | { |
---|
4730 | solver_->resolve() ; |
---|
4731 | numberIterations_ += solver_->getIterationCount() ; |
---|
4732 | feasible = (solver_->isProvenOptimal() && |
---|
4733 | !solver_->isDualObjectiveLimitReached()) ; |
---|
4734 | } |
---|
4735 | if (0&&feasible) { |
---|
4736 | const double * lb = solver_->getColLower(); |
---|
4737 | const double * ub = solver_->getColUpper(); |
---|
4738 | const double * x = solver_->getColSolution(); |
---|
4739 | const double * dj = solver_->getReducedCost(); |
---|
4740 | int numberColumns = solver_->getNumCols(); |
---|
4741 | for (int i=0;i<numberColumns;i++) { |
---|
4742 | if (dj[i]>1.0e-4&&ub[i]-lb[i]>1.0e-4&&x[i]>lb[i]+1.0e-4) |
---|
4743 | printf("error %d %g %g %g %g\n",i,dj[i],lb[i],x[i],ub[i]); |
---|
4744 | if (dj[i]<-1.0e-4&&ub[i]-lb[i]>1.0e-4&&x[i]<ub[i]-1.0e-4) |
---|
4745 | printf("error %d %g %g %g %g\n",i,dj[i],lb[i],x[i],ub[i]); |
---|
4746 | } |
---|
4747 | } |
---|
4748 | if (!feasible&& continuousObjective_ <-1.0e30) { |
---|
4749 | // at root node - double double check |
---|
4750 | bool saveTakeHint; |
---|
4751 | OsiHintStrength saveStrength; |
---|
4752 | solver_->getHintParam(OsiDoDualInResolve,saveTakeHint,saveStrength); |
---|
4753 | if (saveTakeHint||saveStrength==OsiHintIgnore) { |
---|
4754 | solver_->setHintParam(OsiDoDualInResolve,false,OsiHintDo) ; |
---|
4755 | solver_->resolve(); |
---|
4756 | solver_->setHintParam(OsiDoDualInResolve,saveTakeHint,saveStrength); |
---|
4757 | numberIterations_ += solver_->getIterationCount() ; |
---|
4758 | feasible = solver_->isProvenOptimal(); |
---|
4759 | // solver_->writeMps("infeas"); |
---|
4760 | } |
---|
4761 | } |
---|
4762 | setPointers(solver_); |
---|
4763 | int returnStatus = feasible ? 1 : 0; |
---|
4764 | if (strategy_) { |
---|
4765 | // user can play clever tricks here |
---|
4766 | int status = strategy_->status(this,parent,whereFrom); |
---|
4767 | if (status>=0) { |
---|
4768 | if (status==0) |
---|
4769 | returnStatus = 1; |
---|
4770 | else if(status==1) |
---|
4771 | returnStatus=-1; |
---|
4772 | else |
---|
4773 | returnStatus=0; |
---|
4774 | } |
---|
4775 | } |
---|
4776 | return returnStatus ; |
---|
4777 | } |
---|
4778 | |
---|
4779 | |
---|
4780 | /* Set up objects. Only do ones whose length is in range. |
---|
4781 | If makeEquality true then a new model may be returned if |
---|
4782 | modifications had to be made, otherwise "this" is returned. |
---|
4783 | |
---|
4784 | Could use Probing at continuous to extend objects |
---|
4785 | */ |
---|
4786 | CbcModel * |
---|
4787 | CbcModel::findCliques(bool makeEquality, |
---|
4788 | int atLeastThisMany, int lessThanThis, int defaultValue) |
---|
4789 | { |
---|
4790 | // No objects are allowed to exist |
---|
4791 | assert(numberObjects_==numberIntegers_||!numberObjects_); |
---|
4792 | CoinPackedMatrix matrixByRow(*solver_->getMatrixByRow()); |
---|
4793 | int numberRows = solver_->getNumRows(); |
---|
4794 | int numberColumns = solver_->getNumCols(); |
---|
4795 | |
---|
4796 | // We may want to add columns |
---|
4797 | int numberSlacks=0; |
---|
4798 | int * rows = new int[numberRows]; |
---|
4799 | double * element =new double[numberRows]; |
---|
4800 | |
---|
4801 | int iRow; |
---|
4802 | |
---|
4803 | findIntegers(true); |
---|
4804 | numberObjects_=numberIntegers_; |
---|
4805 | |
---|
4806 | int numberCliques=0; |
---|
4807 | CbcObject ** object = new CbcObject * [numberRows]; |
---|
4808 | int * which = new int[numberIntegers_]; |
---|
4809 | char * type = new char[numberIntegers_]; |
---|
4810 | int * lookup = new int[numberColumns]; |
---|
4811 | int i; |
---|
4812 | for (i=0;i<numberColumns;i++) |
---|
4813 | lookup[i]=-1; |
---|
4814 | for (i=0;i<numberIntegers_;i++) |
---|
4815 | lookup[integerVariable_[i]]=i; |
---|
4816 | |
---|
4817 | // Statistics |
---|
4818 | int totalP1=0,totalM1=0; |
---|
4819 | int numberBig=0,totalBig=0; |
---|
4820 | int numberFixed=0; |
---|
4821 | |
---|
4822 | // Row copy |
---|
4823 | const double * elementByRow = matrixByRow.getElements(); |
---|
4824 | const int * column = matrixByRow.getIndices(); |
---|
4825 | const CoinBigIndex * rowStart = matrixByRow.getVectorStarts(); |
---|
4826 | const int * rowLength = matrixByRow.getVectorLengths(); |
---|
4827 | |
---|
4828 | // Column lengths for slacks |
---|
4829 | const int * columnLength = solver_->getMatrixByCol()->getVectorLengths(); |
---|
4830 | |
---|
4831 | const double * lower = getColLower(); |
---|
4832 | const double * upper = getColUpper(); |
---|
4833 | const double * rowLower = getRowLower(); |
---|
4834 | const double * rowUpper = getRowUpper(); |
---|
4835 | |
---|
4836 | for (iRow=0;iRow<numberRows;iRow++) { |
---|
4837 | int numberP1=0, numberM1=0; |
---|
4838 | int j; |
---|
4839 | double upperValue=rowUpper[iRow]; |
---|
4840 | double lowerValue=rowLower[iRow]; |
---|
4841 | bool good=true; |
---|
4842 | int slack = -1; |
---|
4843 | for (j=rowStart[iRow];j<rowStart[iRow]+rowLength[iRow];j++) { |
---|
4844 | int iColumn = column[j]; |
---|
4845 | int iInteger=lookup[iColumn]; |
---|
4846 | if (upper[iColumn]-lower[iColumn]<1.0e-8) { |
---|
4847 | // fixed |
---|
4848 | upperValue -= lower[iColumn]*elementByRow[j]; |
---|
4849 | lowerValue -= lower[iColumn]*elementByRow[j]; |
---|
4850 | continue; |
---|
4851 | } else if (upper[iColumn]!=1.0||lower[iColumn]!=0.0) { |
---|
4852 | good = false; |
---|
4853 | break; |
---|
4854 | } else if (iInteger<0) { |
---|
4855 | good = false; |
---|
4856 | break; |
---|
4857 | } else { |
---|
4858 | if (columnLength[iColumn]==1) |
---|
4859 | slack = iInteger; |
---|
4860 | } |
---|
4861 | if (fabs(elementByRow[j])!=1.0) { |
---|
4862 | good=false; |
---|
4863 | break; |
---|
4864 | } else if (elementByRow[j]>0.0) { |
---|
4865 | which[numberP1++]=iInteger; |
---|
4866 | } else { |
---|
4867 | numberM1++; |
---|
4868 | which[numberIntegers_-numberM1]=iInteger; |
---|
4869 | } |
---|
4870 | } |
---|
4871 | int iUpper = (int) floor(upperValue+1.0e-5); |
---|
4872 | int iLower = (int) ceil(lowerValue-1.0e-5); |
---|
4873 | int state=0; |
---|
4874 | if (upperValue<1.0e6) { |
---|
4875 | if (iUpper==1-numberM1) |
---|
4876 | state=1; |
---|
4877 | else if (iUpper==-numberM1) |
---|
4878 | state=2; |
---|
4879 | else if (iUpper<-numberM1) |
---|
4880 | state=3; |
---|
4881 | } |
---|
4882 | if (!state&&lowerValue>-1.0e6) { |
---|
4883 | if (-iLower==1-numberP1) |
---|
4884 | state=-1; |
---|
4885 | else if (-iLower==-numberP1) |
---|
4886 | state=-2; |
---|
4887 | else if (-iLower<-numberP1) |
---|
4888 | state=-3; |
---|
4889 | } |
---|
4890 | if (good&&state) { |
---|
4891 | if (abs(state)==3) { |
---|
4892 | // infeasible |
---|
4893 | numberObjects_=-1; |
---|
4894 | break; |
---|
4895 | } else if (abs(state)==2) { |
---|
4896 | // we can fix all |
---|
4897 | numberFixed += numberP1+numberM1; |
---|
4898 | if (state>0) { |
---|
4899 | // fix all +1 at 0, -1 at 1 |
---|
4900 | for (i=0;i<numberP1;i++) |
---|
4901 | solver_->setColUpper(integerVariable_[which[i]],0.0); |
---|
4902 | for (i=0;i<numberM1;i++) |
---|
4903 | solver_->setColLower(integerVariable_[which[numberIntegers_-i-1]], |
---|
4904 | 1.0); |
---|
4905 | } else { |
---|
4906 | // fix all +1 at 1, -1 at 0 |
---|
4907 | for (i=0;i<numberP1;i++) |
---|
4908 | solver_->setColLower(integerVariable_[which[i]],1.0); |
---|
4909 | for (i=0;i<numberM1;i++) |
---|
4910 | solver_->setColUpper(integerVariable_[which[numberIntegers_-i-1]], |
---|
4911 | 0.0); |
---|
4912 | } |
---|
4913 | } else { |
---|
4914 | int length = numberP1+numberM1; |
---|
4915 | if (length >= atLeastThisMany&&length<lessThanThis) { |
---|
4916 | // create object |
---|
4917 | bool addOne=false; |
---|
4918 | int objectType; |
---|
4919 | if (iLower==iUpper) { |
---|
4920 | objectType=1; |
---|
4921 | } else { |
---|
4922 | if (makeEquality) { |
---|
4923 | objectType=1; |
---|
4924 | element[numberSlacks]=state; |
---|
4925 | rows[numberSlacks++]=iRow; |
---|
4926 | addOne=true; |
---|
4927 | } else { |
---|
4928 | objectType=0; |
---|
4929 | } |
---|
4930 | } |
---|
4931 | if (state>0) { |
---|
4932 | totalP1 += numberP1; |
---|
4933 | totalM1 += numberM1; |
---|
4934 | for (i=0;i<numberP1;i++) |
---|
4935 | type[i]=1; |
---|
4936 | for (i=0;i<numberM1;i++) { |
---|
4937 | which[numberP1]=which[numberIntegers_-i-1]; |
---|
4938 | type[numberP1++]=0; |
---|
4939 | } |
---|
4940 | } else { |
---|
4941 | totalP1 += numberM1; |
---|
4942 | totalM1 += numberP1; |
---|
4943 | for (i=0;i<numberP1;i++) |
---|
4944 | type[i]=0; |
---|
4945 | for (i=0;i<numberM1;i++) { |
---|
4946 | which[numberP1]=which[numberIntegers_-i-1]; |
---|
4947 | type[numberP1++]=1; |
---|
4948 | } |
---|
4949 | } |
---|
4950 | if (addOne) { |
---|
4951 | // add in slack |
---|
4952 | which[numberP1]=numberIntegers_+numberSlacks-1; |
---|
4953 | slack = numberP1; |
---|
4954 | type[numberP1++]=1; |
---|
4955 | } else if (slack >= 0) { |
---|
4956 | for (i=0;i<numberP1;i++) { |
---|
4957 | if (which[i]==slack) { |
---|
4958 | slack=i; |
---|
4959 | } |
---|
4960 | } |
---|
4961 | } |
---|
4962 | object[numberCliques] = new CbcClique(this,objectType,numberP1, |
---|
4963 | which,type, |
---|
4964 | 1000000+numberCliques,slack); |
---|
4965 | numberCliques++; |
---|
4966 | } else if (numberP1+numberM1 >= lessThanThis) { |
---|
4967 | // too big |
---|
4968 | numberBig++; |
---|
4969 | totalBig += numberP1+numberM1; |
---|
4970 | } |
---|
4971 | } |
---|
4972 | } |
---|
4973 | } |
---|
4974 | delete [] which; |
---|
4975 | delete [] type; |
---|
4976 | delete [] lookup; |
---|
4977 | if (numberCliques<0) { |
---|
4978 | printf("*** Problem infeasible\n"); |
---|
4979 | } else { |
---|
4980 | if (numberCliques) |
---|
4981 | printf("%d cliques of average size %g found, %d P1, %d M1\n", |
---|
4982 | numberCliques, |
---|
4983 | ((double)(totalP1+totalM1))/((double) numberCliques), |
---|
4984 | totalP1,totalM1); |
---|
4985 | else |
---|
4986 | printf("No cliques found\n"); |
---|
4987 | if (numberBig) |
---|
4988 | printf("%d large cliques ( >= %d) found, total %d\n", |
---|
4989 | numberBig,lessThanThis,totalBig); |
---|
4990 | if (numberFixed) |
---|
4991 | printf("%d variables fixed\n",numberFixed); |
---|
4992 | } |
---|
4993 | if (numberCliques>0&&numberSlacks&&makeEquality) { |
---|
4994 | printf("adding %d integer slacks\n",numberSlacks); |
---|
4995 | // add variables to make equality rows |
---|
4996 | int * temp = new int[numberIntegers_+numberSlacks]; |
---|
4997 | memcpy(temp,integerVariable_,numberIntegers_*sizeof(int)); |
---|
4998 | // Get new model |
---|
4999 | CbcModel * newModel = new CbcModel(*this); |
---|
5000 | OsiSolverInterface * newSolver = newModel->solver(); |
---|
5001 | for (i=0;i<numberSlacks;i++) { |
---|
5002 | temp[i+numberIntegers_]=i+numberColumns; |
---|
5003 | int iRow = rows[i]; |
---|
5004 | double value = element[i]; |
---|
5005 | double lowerValue = 0.0; |
---|
5006 | double upperValue = 1.0; |
---|
5007 | double objValue = 0.0; |
---|
5008 | CoinPackedVector column(1,&iRow,&value); |
---|
5009 | newSolver->addCol(column,lowerValue,upperValue,objValue); |
---|
5010 | // set integer |
---|
5011 | newSolver->setInteger(numberColumns+i); |
---|
5012 | if (value >0) |
---|
5013 | newSolver->setRowLower(iRow,rowUpper[iRow]); |
---|
5014 | else |
---|
5015 | newSolver->setRowUpper(iRow,rowLower[iRow]); |
---|
5016 | } |
---|
5017 | // replace list of integers |
---|
5018 | for (i=0;i<newModel->numberObjects_;i++) |
---|
5019 | delete newModel->object_[i]; |
---|
5020 | newModel->numberObjects_ = 0; |
---|
5021 | delete [] newModel->object_; |
---|
5022 | newModel->object_=NULL; |
---|
5023 | newModel->findIntegers(true); //Set up all integer objects |
---|
5024 | for (i=0;i<numberIntegers_;i++) { |
---|
5025 | newModel->modifiableObject(i)->setPriority(object_[i]->priority()); |
---|
5026 | } |
---|
5027 | if (originalColumns_) { |
---|
5028 | // old model had originalColumns |
---|
5029 | delete [] newModel->originalColumns_; |
---|
5030 | newModel->originalColumns_ = new int[numberColumns+numberSlacks]; |
---|
5031 | memcpy(newModel->originalColumns_,originalColumns_,numberColumns*sizeof(int)); |
---|
5032 | // mark as not in previous model |
---|
5033 | for (i=numberColumns;i<numberColumns+numberSlacks;i++) |
---|
5034 | newModel->originalColumns_[i]=-1; |
---|
5035 | } |
---|
5036 | delete [] rows; |
---|
5037 | delete [] element; |
---|
5038 | newModel->addObjects(numberCliques,object); |
---|
5039 | for (;i<numberCliques;i++) |
---|
5040 | delete object[i]; |
---|
5041 | delete [] object; |
---|
5042 | newModel->synchronizeModel(); |
---|
5043 | return newModel; |
---|
5044 | } else { |
---|
5045 | if (numberCliques>0) { |
---|
5046 | addObjects(numberCliques,object); |
---|
5047 | for (;i<numberCliques;i++) |
---|
5048 | delete object[i]; |
---|
5049 | synchronizeModel(); |
---|
5050 | } |
---|
5051 | delete [] object; |
---|
5052 | delete [] rows; |
---|
5053 | delete [] element; |
---|
5054 | return this; |
---|
5055 | } |
---|
5056 | } |
---|
5057 | // Fill in useful estimates |
---|
5058 | void |
---|
5059 | CbcModel::pseudoShadow(double * down, double * up) |
---|
5060 | { |
---|
5061 | // Column copy of matrix |
---|
5062 | const double * element = solver_->getMatrixByCol()->getElements(); |
---|
5063 | const int * row = solver_->getMatrixByCol()->getIndices(); |
---|
5064 | const CoinBigIndex * columnStart = solver_->getMatrixByCol()->getVectorStarts(); |
---|
5065 | const int * columnLength = solver_->getMatrixByCol()->getVectorLengths(); |
---|
5066 | const double *objective = solver_->getObjCoefficients() ; |
---|
5067 | int numberColumns = solver_->getNumCols() ; |
---|
5068 | double direction = solver_->getObjSense(); |
---|
5069 | int iColumn; |
---|
5070 | const double * dual = cbcRowPrice_; |
---|
5071 | down = new double[numberColumns]; |
---|
5072 | up = new double[numberColumns]; |
---|
5073 | double upSum=1.0e-20; |
---|
5074 | double downSum = 1.0e-20; |
---|
5075 | int numberIntegers=0; |
---|
5076 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
5077 | CoinBigIndex start = columnStart[iColumn]; |
---|
5078 | CoinBigIndex end = start + columnLength[iColumn]; |
---|
5079 | double upValue = 0.0; |
---|
5080 | double downValue = 0.0; |
---|
5081 | double value = direction*objective[iColumn]; |
---|
5082 | if (value) { |
---|
5083 | if (value>0.0) |
---|
5084 | upValue += value; |
---|
5085 | else |
---|
5086 | downValue -= value; |
---|
5087 | } |
---|
5088 | for (CoinBigIndex j=start;j<end;j++) { |
---|
5089 | int iRow = row[j]; |
---|
5090 | value = -dual[iRow]; |
---|
5091 | if (value) { |
---|
5092 | value *= element[j]; |
---|
5093 | if (value>0.0) |
---|
5094 | upValue += value; |
---|
5095 | else |
---|
5096 | downValue -= value; |
---|
5097 | } |
---|
5098 | } |
---|
5099 | // use dj if bigger |
---|
5100 | double dj = cbcReducedCost_[iColumn]; |
---|
5101 | upValue = CoinMax(upValue,dj); |
---|
5102 | downValue = CoinMax(downValue,-dj); |
---|
5103 | up[iColumn]=upValue; |
---|
5104 | down[iColumn]=downValue; |
---|
5105 | if (solver_->isInteger(iColumn)) { |
---|
5106 | if (!numberNodes_) |
---|
5107 | printf("%d - dj %g up %g down %g cost %g\n", |
---|
5108 | iColumn,dj,upValue,downValue,objective[iColumn]); |
---|
5109 | upSum += upValue; |
---|
5110 | downSum += downValue; |
---|
5111 | numberIntegers++; |
---|
5112 | } |
---|
5113 | } |
---|
5114 | if (numberIntegers) { |
---|
5115 | double smallDown = 0.01*(downSum/((double) numberIntegers)); |
---|
5116 | double smallUp = 0.01*(upSum/((double) numberIntegers)); |
---|
5117 | for (int i=0;i<numberObjects_;i++) { |
---|
5118 | CbcSimpleIntegerDynamicPseudoCost * obj1 = |
---|
5119 | dynamic_cast <CbcSimpleIntegerDynamicPseudoCost *>(object_[i]) ; |
---|
5120 | if (obj1) { |
---|
5121 | iColumn = obj1->columnNumber(); |
---|
5122 | double upPseudoCost = obj1->upDynamicPseudoCost(); |
---|
5123 | double saveUp = upPseudoCost; |
---|
5124 | upPseudoCost = CoinMax(upPseudoCost,smallUp); |
---|
5125 | upPseudoCost = CoinMax(upPseudoCost,up[iColumn]); |
---|
5126 | upPseudoCost = CoinMax(upPseudoCost,0.1*down[iColumn]); |
---|
5127 | obj1->setUpDynamicPseudoCost(upPseudoCost); |
---|
5128 | if (upPseudoCost>saveUp&&!numberNodes_) |
---|
5129 | printf("For %d up went from %g to %g\n", |
---|
5130 | iColumn,saveUp,upPseudoCost); |
---|
5131 | double downPseudoCost = obj1->downDynamicPseudoCost(); |
---|
5132 | double saveDown = downPseudoCost; |
---|
5133 | downPseudoCost = CoinMax(downPseudoCost,smallDown); |
---|
5134 | downPseudoCost = CoinMax(downPseudoCost,down[iColumn]); |
---|
5135 | downPseudoCost = CoinMax(downPseudoCost,0.1*down[iColumn]); |
---|
5136 | obj1->setDownDynamicPseudoCost(downPseudoCost); |
---|
5137 | if (downPseudoCost>saveDown&&!numberNodes_) |
---|
5138 | printf("For %d down went from %g to %g\n", |
---|
5139 | iColumn,saveDown,downPseudoCost); |
---|
5140 | } |
---|
5141 | } |
---|
5142 | } |
---|
5143 | delete [] down; |
---|
5144 | delete [] up; |
---|
5145 | } |
---|
5146 | |
---|
5147 | /* |
---|
5148 | Set branching priorities. |
---|
5149 | |
---|
5150 | Setting integer priorities looks pretty robust; the call to findIntegers |
---|
5151 | makes sure that SimpleInteger objects are in place. Setting priorities for |
---|
5152 | other objects is entirely dependent on their existence, and the routine may |
---|
5153 | quietly fail in several directions. |
---|
5154 | */ |
---|
5155 | |
---|
5156 | void |
---|
5157 | CbcModel::passInPriorities (const int * priorities, |
---|
5158 | bool ifObject) |
---|
5159 | { |
---|
5160 | findIntegers(false); |
---|
5161 | int i; |
---|
5162 | if (priorities) { |
---|
5163 | int i0=0; |
---|
5164 | int i1=numberObjects_-1; |
---|
5165 | if (ifObject) { |
---|
5166 | for (i=numberIntegers_;i<numberObjects_;i++) { |
---|
5167 | object_[i]->setPriority(priorities[i-numberIntegers_]); |
---|
5168 | } |
---|
5169 | i0=numberIntegers_; |
---|
5170 | } else { |
---|
5171 | for (i=0;i<numberIntegers_;i++) { |
---|
5172 | object_[i]->setPriority(priorities[i]); |
---|
5173 | } |
---|
5174 | i1=numberIntegers_-1; |
---|
5175 | } |
---|
5176 | messageHandler()->message(CBC_PRIORITY, |
---|
5177 | messages()) |
---|
5178 | << i0<<i1<<numberObjects_ << CoinMessageEol ; |
---|
5179 | } |
---|
5180 | } |
---|
5181 | |
---|
5182 | // Delete all object information |
---|
5183 | void |
---|
5184 | CbcModel::deleteObjects() |
---|
5185 | { |
---|
5186 | int i; |
---|
5187 | for (i=0;i<numberObjects_;i++) |
---|
5188 | delete object_[i]; |
---|
5189 | delete [] object_; |
---|
5190 | object_ = NULL; |
---|
5191 | numberObjects_=0; |
---|
5192 | findIntegers(true); |
---|
5193 | } |
---|
5194 | |
---|
5195 | /*! |
---|
5196 | Ensure all attached objects (CbcObjects, heuristics, and cut |
---|
5197 | generators) point to this model. |
---|
5198 | */ |
---|
5199 | void CbcModel::synchronizeModel() |
---|
5200 | { |
---|
5201 | int i; |
---|
5202 | for (i=0;i<numberHeuristics_;i++) |
---|
5203 | heuristic_[i]->setModel(this); |
---|
5204 | for (i=0;i<numberObjects_;i++) |
---|
5205 | object_[i]->setModel(this); |
---|
5206 | for (i=0;i<numberCutGenerators_;i++) |
---|
5207 | generator_[i]->refreshModel(this); |
---|
5208 | } |
---|
5209 | |
---|
5210 | // Fill in integers and create objects |
---|
5211 | |
---|
5212 | /** |
---|
5213 | The routine first does a scan to count the number of integer variables. |
---|
5214 | It then creates an array, integerVariable_, to store the indices of the |
---|
5215 | integer variables, and an array of `objects', one for each variable. |
---|
5216 | |
---|
5217 | The scan is repeated, this time recording the index of each integer |
---|
5218 | variable in integerVariable_, and creating an CbcSimpleInteger object that |
---|
5219 | contains information about the integer variable. Initially, this is just |
---|
5220 | the index and upper & lower bounds. |
---|
5221 | |
---|
5222 | \todo |
---|
5223 | Note the assumption in cbc that the first numberIntegers_ objects are |
---|
5224 | CbcSimpleInteger. In particular, the code which handles the startAgain |
---|
5225 | case assumes that if the object_ array exists it can simply replace the first |
---|
5226 | numberInteger_ objects. This is arguably unsafe. |
---|
5227 | |
---|
5228 | I am going to re-order if necessary |
---|
5229 | */ |
---|
5230 | |
---|
5231 | void |
---|
5232 | CbcModel::findIntegers(bool startAgain,int type) |
---|
5233 | { |
---|
5234 | assert(solver_); |
---|
5235 | /* |
---|
5236 | No need to do this if we have previous information, unless forced to start |
---|
5237 | over. |
---|
5238 | */ |
---|
5239 | if (numberIntegers_&&!startAgain&&object_) |
---|
5240 | return; |
---|
5241 | /* |
---|
5242 | Clear out the old integer variable list, then count the number of integer |
---|
5243 | variables. |
---|
5244 | */ |
---|
5245 | delete [] integerVariable_; |
---|
5246 | numberIntegers_=0; |
---|
5247 | int numberColumns = getNumCols(); |
---|
5248 | int iColumn; |
---|
5249 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
5250 | if (isInteger(iColumn)) |
---|
5251 | numberIntegers_++; |
---|
5252 | } |
---|
5253 | // Find out how many old non-integer objects there are |
---|
5254 | int nObjects=0; |
---|
5255 | CbcObject ** oldObject = object_; |
---|
5256 | int iObject; |
---|
5257 | for (iObject = 0;iObject<numberObjects_;iObject++) { |
---|
5258 | CbcSimpleInteger * obj = |
---|
5259 | dynamic_cast <CbcSimpleInteger *>(oldObject[iObject]) ; |
---|
5260 | if (obj) |
---|
5261 | delete oldObject[iObject]; |
---|
5262 | else |
---|
5263 | oldObject[nObjects++]=oldObject[iObject]; |
---|
5264 | } |
---|
5265 | |
---|
5266 | /* |
---|
5267 | Found any? Allocate an array to hold the indices of the integer variables. |
---|
5268 | Make a large enough array for all objects |
---|
5269 | */ |
---|
5270 | object_ = new CbcObject * [numberIntegers_+nObjects]; |
---|
5271 | numberObjects_=numberIntegers_+nObjects;; |
---|
5272 | integerVariable_ = new int [numberIntegers_]; |
---|
5273 | /* |
---|
5274 | Walk the variables again, filling in the indices and creating objects for |
---|
5275 | the integer variables. Initially, the objects hold the index and upper & |
---|
5276 | lower bounds. |
---|
5277 | */ |
---|
5278 | numberIntegers_=0; |
---|
5279 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
5280 | if(isInteger(iColumn)) { |
---|
5281 | if (!type) |
---|
5282 | object_[numberIntegers_] = |
---|
5283 | new CbcSimpleInteger(this,numberIntegers_,iColumn); |
---|
5284 | else if (type==1) |
---|
5285 | object_[numberIntegers_] = |
---|
5286 | new CbcSimpleIntegerPseudoCost(this,numberIntegers_,iColumn,0.3); |
---|
5287 | integerVariable_[numberIntegers_++]=iColumn; |
---|
5288 | } |
---|
5289 | } |
---|
5290 | // Now append other objects |
---|
5291 | memcpy(object_+numberIntegers_,oldObject,nObjects*sizeof(CbcObject *)); |
---|
5292 | // Delete old array (just array) |
---|
5293 | delete [] oldObject; |
---|
5294 | |
---|
5295 | if (!numberObjects_) |
---|
5296 | handler_->message(CBC_NOINT,messages_) << CoinMessageEol ; |
---|
5297 | } |
---|
5298 | /* If numberBeforeTrust >0 then we are going to use CbcBranchDynamic. |
---|
5299 | Scan and convert CbcSimpleInteger objects |
---|
5300 | */ |
---|
5301 | void |
---|
5302 | CbcModel::convertToDynamic() |
---|
5303 | { |
---|
5304 | int iObject; |
---|
5305 | const double * cost = solver_->getObjCoefficients(); |
---|
5306 | for (iObject = 0;iObject<numberObjects_;iObject++) { |
---|
5307 | CbcSimpleInteger * obj1 = |
---|
5308 | dynamic_cast <CbcSimpleInteger *>(object_[iObject]) ; |
---|
5309 | CbcSimpleIntegerPseudoCost * obj1a = |
---|
5310 | dynamic_cast <CbcSimpleIntegerPseudoCost *>(object_[iObject]) ; |
---|
5311 | CbcSimpleIntegerDynamicPseudoCost * obj2 = |
---|
5312 | dynamic_cast <CbcSimpleIntegerDynamicPseudoCost *>(object_[iObject]) ; |
---|
5313 | if (obj1&&!obj2) { |
---|
5314 | // replace |
---|
5315 | int iColumn = obj1->columnNumber(); |
---|
5316 | int priority = obj1->priority(); |
---|
5317 | int preferredWay = obj1->preferredWay(); |
---|
5318 | delete object_[iObject]; |
---|
5319 | double costValue = CoinMax(1.0e-5,fabs(cost[iColumn])); |
---|
5320 | // treat as if will cost what it says up |
---|
5321 | double upCost=costValue; |
---|
5322 | // and balance at breakeven of 0.3 |
---|
5323 | double downCost=(0.7*upCost)/0.3; |
---|
5324 | if (obj1a) { |
---|
5325 | upCost=obj1a->upPseudoCost(); |
---|
5326 | downCost=obj1a->downPseudoCost(); |
---|
5327 | } |
---|
5328 | CbcSimpleIntegerDynamicPseudoCost * newObject = |
---|
5329 | new CbcSimpleIntegerDynamicPseudoCost(this,iObject,iColumn,downCost,upCost); |
---|
5330 | newObject->setNumberBeforeTrust(numberBeforeTrust_); |
---|
5331 | newObject->setPriority(priority); |
---|
5332 | newObject->setPreferredWay(preferredWay); |
---|
5333 | object_[iObject] = newObject; |
---|
5334 | } |
---|
5335 | } |
---|
5336 | if (branchingMethod_&&(branchingMethod_->whichMethod()&1)==0) { |
---|
5337 | // Need a method which can do better |
---|
5338 | branchingMethod_=NULL; |
---|
5339 | } |
---|
5340 | } |
---|
5341 | |
---|
5342 | /* Add in any object information (objects are cloned - owner can delete |
---|
5343 | originals */ |
---|
5344 | void |
---|
5345 | CbcModel::addObjects(int numberObjects, CbcObject ** objects) |
---|
5346 | { |
---|
5347 | // If integers but not enough objects fudge |
---|
5348 | if (numberIntegers_>numberObjects_) |
---|
5349 | findIntegers(true); |
---|
5350 | /* But if incoming objects inherit from simple integer we just want |
---|
5351 | to replace */ |
---|
5352 | int numberColumns = solver_->getNumCols(); |
---|
5353 | /** mark is -1 if not integer, >=0 if using existing simple integer and |
---|
5354 | >=numberColumns if using new integer */ |
---|
5355 | int * mark = new int[numberColumns]; |
---|
5356 | int i; |
---|
5357 | for (i=0;i<numberColumns;i++) |
---|
5358 | mark[i]=-1; |
---|
5359 | int newNumberObjects = numberObjects; |
---|
5360 | int newIntegers=0; |
---|
5361 | for (i=0;i<numberObjects;i++) { |
---|
5362 | CbcSimpleInteger * obj = |
---|
5363 | dynamic_cast <CbcSimpleInteger *>(objects[i]) ; |
---|
5364 | if (obj) { |
---|
5365 | int iColumn = obj->columnNumber(); |
---|
5366 | mark[iColumn]=i+numberColumns; |
---|
5367 | newIntegers++; |
---|
5368 | } |
---|
5369 | } |
---|
5370 | // and existing |
---|
5371 | for (i=0;i<numberObjects_;i++) { |
---|
5372 | CbcSimpleInteger * obj = |
---|
5373 | dynamic_cast <CbcSimpleInteger *>(object_[i]) ; |
---|
5374 | if (obj) { |
---|
5375 | int iColumn = obj->columnNumber(); |
---|
5376 | if (mark[iColumn]<0) { |
---|
5377 | newIntegers++; |
---|
5378 | newNumberObjects++; |
---|
5379 | mark[iColumn]=i; |
---|
5380 | } |
---|
5381 | } |
---|
5382 | } |
---|
5383 | delete [] integerVariable_; |
---|
5384 | integerVariable_=NULL; |
---|
5385 | if (newIntegers!=numberIntegers_) |
---|
5386 | printf("changing number of integers from %d to %d\n", |
---|
5387 | numberIntegers_,newIntegers); |
---|
5388 | numberIntegers_ = newIntegers; |
---|
5389 | integerVariable_ = new int [numberIntegers_]; |
---|
5390 | CbcObject ** temp = new CbcObject * [newNumberObjects]; |
---|
5391 | // Put integers first |
---|
5392 | newIntegers=0; |
---|
5393 | numberIntegers_=0; |
---|
5394 | for (i=0;i<numberColumns;i++) { |
---|
5395 | int which = mark[i]; |
---|
5396 | if (which>=0) { |
---|
5397 | if (!isInteger(i)) { |
---|
5398 | newIntegers++; |
---|
5399 | solver_->setInteger(i); |
---|
5400 | } |
---|
5401 | if (which<numberColumns) { |
---|
5402 | temp[numberIntegers_]=object_[which]; |
---|
5403 | object_[which]=NULL; |
---|
5404 | } else { |
---|
5405 | temp[numberIntegers_]=objects[which-numberColumns]->clone(); |
---|
5406 | temp[numberIntegers_]->setModel(this); |
---|
5407 | } |
---|
5408 | integerVariable_[numberIntegers_++]=i; |
---|
5409 | } |
---|
5410 | } |
---|
5411 | if (newIntegers) |
---|
5412 | printf("%d variables were declared integer\n",newIntegers); |
---|
5413 | int n=numberIntegers_; |
---|
5414 | // Now rest of old |
---|
5415 | for (i=0;i<numberObjects_;i++) { |
---|
5416 | if (object_[i]) { |
---|
5417 | CbcSimpleInteger * obj = |
---|
5418 | dynamic_cast <CbcSimpleInteger *>(object_[i]) ; |
---|
5419 | if (obj) { |
---|
5420 | delete object_[i]; |
---|
5421 | } else { |
---|
5422 | temp[n++]=object_[i]; |
---|
5423 | } |
---|
5424 | } |
---|
5425 | } |
---|
5426 | // and rest of new |
---|
5427 | for (i=0;i<numberObjects;i++) { |
---|
5428 | CbcSimpleInteger * obj = |
---|
5429 | dynamic_cast <CbcSimpleInteger *>(objects[i]) ; |
---|
5430 | if (!obj) { |
---|
5431 | temp[n]=objects[i]->clone(); |
---|
5432 | temp[n++]->setModel(this); |
---|
5433 | } |
---|
5434 | } |
---|
5435 | delete [] mark; |
---|
5436 | delete [] object_; |
---|
5437 | object_ = temp; |
---|
5438 | assert (n==newNumberObjects); |
---|
5439 | numberObjects_ = newNumberObjects; |
---|
5440 | } |
---|
5441 | |
---|
5442 | /** |
---|
5443 | This routine sets the objective cutoff value used for fathoming and |
---|
5444 | determining monotonic variables. |
---|
5445 | |
---|
5446 | If the fathoming discipline is strict, a small tolerance is added to the |
---|
5447 | new cutoff. This avoids problems due to roundoff when the target value |
---|
5448 | is exact. The common example would be an IP with only integer variables in |
---|
5449 | the objective. If the target is set to the exact value z of the optimum, |
---|
5450 | it's possible to end up fathoming an ancestor of the solution because the |
---|
5451 | solver returns z+epsilon. |
---|
5452 | |
---|
5453 | Determining if strict fathoming is needed is best done by analysis. |
---|
5454 | In cbc, that's analyseObjective. The default is false. |
---|
5455 | |
---|
5456 | In cbc we always minimize so add epsilon |
---|
5457 | */ |
---|
5458 | |
---|
5459 | void CbcModel::setCutoff (double value) |
---|
5460 | |
---|
5461 | { |
---|
5462 | #if 0 |
---|
5463 | double tol = 0 ; |
---|
5464 | int fathomStrict = getIntParam(CbcFathomDiscipline) ; |
---|
5465 | if (fathomStrict == 1) |
---|
5466 | { solver_->getDblParam(OsiDualTolerance,tol) ; |
---|
5467 | tol = tol*(1+fabs(value)) ; |
---|
5468 | |
---|
5469 | value += tol ; } |
---|
5470 | #endif |
---|
5471 | dblParam_[CbcCurrentCutoff]=value; |
---|
5472 | // Solvers know about direction |
---|
5473 | double direction = solver_->getObjSense(); |
---|
5474 | solver_->setDblParam(OsiDualObjectiveLimit,value*direction); } |
---|
5475 | |
---|
5476 | |
---|
5477 | |
---|
5478 | /* |
---|
5479 | Call this to really test if a valid solution can be feasible. The cutoff is |
---|
5480 | passed in as a parameter so that we don't need to worry here after swapping |
---|
5481 | solvers. The solution is assumed to be numberColumns in size. If |
---|
5482 | fixVariables is true then the bounds of the continuous solver are updated. |
---|
5483 | The routine returns the objective value determined by reoptimizing from |
---|
5484 | scratch. If the solution is rejected, this will be worse than the cutoff. |
---|
5485 | |
---|
5486 | TODO: There's an issue with getting the correct cutoff value: We update the |
---|
5487 | cutoff in the regular solver, but not in continuousSolver_. But our only |
---|
5488 | use for continuousSolver_ is verifying candidate solutions. Would it |
---|
5489 | make sense to update the cutoff? Then we wouldn't need to step around |
---|
5490 | isDualObjectiveLimitReached(). |
---|
5491 | */ |
---|
5492 | double |
---|
5493 | CbcModel::checkSolution (double cutoff, const double *solution, |
---|
5494 | bool fixVariables) |
---|
5495 | |
---|
5496 | { |
---|
5497 | if (!solverCharacteristics_->solutionAddsCuts()) { |
---|
5498 | // Can trust solution |
---|
5499 | int numberColumns = solver_->getNumCols(); |
---|
5500 | |
---|
5501 | /* |
---|
5502 | Grab the continuous solver (the pristine copy of the problem, made before |
---|
5503 | starting to work on the root node). Save the bounds on the variables. |
---|
5504 | Install the solution passed as a parameter, and copy it to the model's |
---|
5505 | currentSolution_. |
---|
5506 | |
---|
5507 | TODO: This is a belt-and-suspenders approach. Once the code has settled |
---|
5508 | a bit, we can cast a critical eye here. |
---|
5509 | */ |
---|
5510 | OsiSolverInterface * saveSolver = solver_; |
---|
5511 | if (continuousSolver_) |
---|
5512 | solver_ = continuousSolver_; |
---|
5513 | // move solution to continuous copy |
---|
5514 | solver_->setColSolution(solution); |
---|
5515 | // Put current solution in safe place |
---|
5516 | // Point to current solution |
---|
5517 | const double * save = testSolution_; |
---|
5518 | // Safe as will be const inside infeasibility() |
---|
5519 | testSolution_ = solver_->getColSolution(); |
---|
5520 | //memcpy(currentSolution_,solver_->getColSolution(), |
---|
5521 | // numberColumns*sizeof(double)); |
---|
5522 | //solver_->messageHandler()->setLogLevel(4); |
---|
5523 | |
---|
5524 | // save original bounds |
---|
5525 | double * saveUpper = new double[numberColumns]; |
---|
5526 | double * saveLower = new double[numberColumns]; |
---|
5527 | memcpy(saveUpper,getColUpper(),numberColumns*sizeof(double)); |
---|
5528 | memcpy(saveLower,getColLower(),numberColumns*sizeof(double)); |
---|
5529 | |
---|
5530 | /* |
---|
5531 | Run through the objects and use feasibleRegion() to set variable bounds |
---|
5532 | so as to fix the variables specified in the objects at their value in this |
---|
5533 | solution. Since the object list contains (at least) one object for every |
---|
5534 | integer variable, this has the effect of fixing all integer variables. |
---|
5535 | */ |
---|
5536 | int i; |
---|
5537 | for (i=0;i<numberObjects_;i++) |
---|
5538 | object_[i]->feasibleRegion(); |
---|
5539 | // We can switch off check |
---|
5540 | if ((specialOptions_&4)==0) { |
---|
5541 | if ((specialOptions_&2)==0&&solverCharacteristics_->warmStart()) { |
---|
5542 | /* |
---|
5543 | Remove any existing warm start information to be sure there is no |
---|
5544 | residual influence on initialSolve(). |
---|
5545 | */ |
---|
5546 | CoinWarmStartBasis *slack = |
---|
5547 | dynamic_cast<CoinWarmStartBasis *>(solver_->getEmptyWarmStart()) ; |
---|
5548 | solver_->setWarmStart(slack); |
---|
5549 | delete slack ; |
---|
5550 | } |
---|
5551 | // Give a hint to do dual |
---|
5552 | bool saveTakeHint; |
---|
5553 | OsiHintStrength saveStrength; |
---|
5554 | bool gotHint = (solver_->getHintParam(OsiDoDualInInitial,saveTakeHint,saveStrength)); |
---|
5555 | assert (gotHint); |
---|
5556 | solver_->setHintParam(OsiDoDualInInitial,true,OsiHintTry); |
---|
5557 | solver_->initialSolve(); |
---|
5558 | if (!solver_->isProvenOptimal()) |
---|
5559 | { //printf("checkSolution infeas! Retrying wihout scaling.\n"); |
---|
5560 | bool saveTakeHint; |
---|
5561 | OsiHintStrength saveStrength; |
---|
5562 | bool savePrintHint; |
---|
5563 | //solver_->writeMps("infeas"); |
---|
5564 | bool gotHint = (solver_->getHintParam(OsiDoReducePrint,savePrintHint,saveStrength)); |
---|
5565 | gotHint = (solver_->getHintParam(OsiDoScale,saveTakeHint,saveStrength)); |
---|
5566 | solver_->setHintParam(OsiDoScale,false,OsiHintTry); |
---|
5567 | //solver_->setHintParam(OsiDoReducePrint,false,OsiHintTry) ; |
---|
5568 | solver_->setHintParam(OsiDoDualInInitial,false,OsiHintTry); |
---|
5569 | solver_->initialSolve(); |
---|
5570 | solver_->setHintParam(OsiDoScale,saveTakeHint,saveStrength); |
---|
5571 | //solver_->setHintParam(OsiDoReducePrint,savePrintHint,OsiHintTry) ; |
---|
5572 | } |
---|
5573 | //assert(solver_->isProvenOptimal()); |
---|
5574 | solver_->setHintParam(OsiDoDualInInitial,saveTakeHint,saveStrength); |
---|
5575 | } |
---|
5576 | double objectiveValue = solver_->getObjValue()*solver_->getObjSense(); |
---|
5577 | |
---|
5578 | /* |
---|
5579 | Check that the solution still beats the objective cutoff. |
---|
5580 | |
---|
5581 | If it passes, make a copy of the primal variable values and do some |
---|
5582 | cleanup and checks: |
---|
5583 | + Values of all variables are are within original bounds and values of |
---|
5584 | all integer variables are within tolerance of integral. |
---|
5585 | + There are no constraint violations. |
---|
5586 | There really should be no need for the check against original bounds. |
---|
5587 | Perhaps an opportunity for a sanity check? |
---|
5588 | */ |
---|
5589 | if ((solver_->isProvenOptimal()||(specialOptions_&4)!=0) && objectiveValue <= cutoff) { |
---|
5590 | double * solution = new double[numberColumns]; |
---|
5591 | memcpy(solution ,solver_->getColSolution(),numberColumns*sizeof(double)) ; |
---|
5592 | |
---|
5593 | int iColumn; |
---|
5594 | #ifndef NDEBUG |
---|
5595 | double integerTolerance = getIntegerTolerance() ; |
---|
5596 | #endif |
---|
5597 | for (iColumn = 0 ; iColumn < numberColumns ; iColumn++) { |
---|
5598 | double value = solution[iColumn] ; |
---|
5599 | value = CoinMax(value, saveLower[iColumn]) ; |
---|
5600 | value = CoinMin(value, saveUpper[iColumn]) ; |
---|
5601 | if (solver_->isInteger(iColumn)) |
---|
5602 | assert(fabs(value-solution[iColumn]) <= integerTolerance) ; |
---|
5603 | solution[iColumn] = value ; |
---|
5604 | } |
---|
5605 | if ((specialOptions_&16)==0) { |
---|
5606 | const double * rowLower = solver_->getRowLower() ; |
---|
5607 | const double * rowUpper = solver_->getRowUpper() ; |
---|
5608 | int numberRows = solver_->getNumRows() ; |
---|
5609 | double *rowActivity = new double[numberRows] ; |
---|
5610 | memset(rowActivity,0,numberRows*sizeof(double)) ; |
---|
5611 | solver_->getMatrixByCol()->times(solution,rowActivity) ; |
---|
5612 | double primalTolerance ; |
---|
5613 | solver_->getDblParam(OsiPrimalTolerance,primalTolerance) ; |
---|
5614 | double largestInfeasibility =0.0; |
---|
5615 | for (i=0 ; i < numberRows ; i++) { |
---|
5616 | largestInfeasibility = CoinMax(largestInfeasibility, |
---|
5617 | rowLower[i]-rowActivity[i]); |
---|
5618 | largestInfeasibility = CoinMax(largestInfeasibility, |
---|
5619 | rowActivity[i]-rowUpper[i]); |
---|
5620 | } |
---|
5621 | if (largestInfeasibility>100.0*primalTolerance) { |
---|
5622 | handler_->message(CBC_NOTFEAS3, messages_) |
---|
5623 | << largestInfeasibility << CoinMessageEol ; |
---|
5624 | objectiveValue=1.0e50 ; |
---|
5625 | } |
---|
5626 | delete [] rowActivity ; |
---|
5627 | } |
---|
5628 | delete [] solution; |
---|
5629 | } else { |
---|
5630 | objectiveValue=1.0e50 ; |
---|
5631 | } |
---|
5632 | /* |
---|
5633 | Regardless of what we think of the solution, we may need to restore the |
---|
5634 | original bounds of the continuous solver. Unfortunately, const'ness |
---|
5635 | prevents us from simply reversing the memcpy used to make these snapshots. |
---|
5636 | */ |
---|
5637 | if (!fixVariables) |
---|
5638 | { for (int iColumn = 0 ; iColumn < numberColumns ; iColumn++) |
---|
5639 | { solver_->setColLower(iColumn,saveLower[iColumn]) ; |
---|
5640 | solver_->setColUpper(iColumn,saveUpper[iColumn]) ; } } |
---|
5641 | delete [] saveLower; |
---|
5642 | delete [] saveUpper; |
---|
5643 | |
---|
5644 | /* |
---|
5645 | Restore the usual solver. |
---|
5646 | */ |
---|
5647 | solver_=saveSolver; |
---|
5648 | testSolution_ = save; |
---|
5649 | return objectiveValue; |
---|
5650 | } else { |
---|
5651 | // Outer approximation or similar |
---|
5652 | //If this is true then the solution comes from the nlp we don't need to resolve the same nlp with ipopt |
---|
5653 | //solverCharacteristics_->setSolver(solver_); |
---|
5654 | bool solutionComesFromNlp = solverCharacteristics_->bestObjectiveValue()<cutoff; |
---|
5655 | double objectiveValue; |
---|
5656 | int numberColumns = solver_->getNumCols(); |
---|
5657 | double *saveLower = NULL; |
---|
5658 | double * saveUpper = NULL; |
---|
5659 | |
---|
5660 | if(! solutionComesFromNlp)//Otherwise solution already comes from ipopt and cuts are known |
---|
5661 | { |
---|
5662 | if(fixVariables)//Will temporarily fix all integer valued var |
---|
5663 | { |
---|
5664 | // save original bounds |
---|
5665 | saveUpper = new double[numberColumns]; |
---|
5666 | saveLower = new double[numberColumns]; |
---|
5667 | memcpy(saveUpper,solver_->getColUpper(),numberColumns*sizeof(double)); |
---|
5668 | memcpy(saveLower,solver_->getColLower(),numberColumns*sizeof(double)); |
---|
5669 | //in any case solution should be already loaded into solver_ |
---|
5670 | /* |
---|
5671 | Run through the objects and use feasibleRegion() to set variable bounds |
---|
5672 | so as to fix the variables specified in the objects at their value in this |
---|
5673 | solution. Since the object list contains (at least) one object for every |
---|
5674 | integer variable, this has the effect of fixing all integer variables. |
---|
5675 | */ |
---|
5676 | const double * save = testSolution_; |
---|
5677 | testSolution_ = solution; |
---|
5678 | for (int i=0;i<numberObjects_;i++) |
---|
5679 | object_[i]->feasibleRegion(); |
---|
5680 | testSolution_ = save; |
---|
5681 | solver_->resolve(); |
---|
5682 | } |
---|
5683 | |
---|
5684 | /* |
---|
5685 | Now step through the cut generators and see if any of them are flagged to |
---|
5686 | run when a new solution is discovered. Only global cuts are useful. |
---|
5687 | (The solution being evaluated may not correspond to the current location in the |
---|
5688 | search tree --- discovered by heuristic, for example.) |
---|
5689 | */ |
---|
5690 | OsiCuts theseCuts; |
---|
5691 | int i; |
---|
5692 | int lastNumberCuts=0; |
---|
5693 | for (i=0;i<numberCutGenerators_;i++) |
---|
5694 | { |
---|
5695 | if (generator_[i]->atSolution()) |
---|
5696 | { |
---|
5697 | generator_[i]->generateCuts(theseCuts,true,NULL); |
---|
5698 | int numberCuts = theseCuts.sizeRowCuts(); |
---|
5699 | for (int j=lastNumberCuts;j<numberCuts;j++) |
---|
5700 | { |
---|
5701 | const OsiRowCut * thisCut = theseCuts.rowCutPtr(j); |
---|
5702 | if (thisCut->globallyValid()) |
---|
5703 | { |
---|
5704 | // if ((specialOptions_&1)!=0) |
---|
5705 | // { |
---|
5706 | // /* As these are global cuts - |
---|
5707 | // a) Always get debugger object |
---|
5708 | // b) Not fatal error to cutoff optimal (if we have just got optimal) |
---|
5709 | // */ |
---|
5710 | // const OsiRowCutDebugger *debugger = solver_->getRowCutDebuggerAlways() ; |
---|
5711 | // if (debugger) |
---|
5712 | // { |
---|
5713 | // if(debugger->invalidCut(*thisCut)) |
---|
5714 | // printf("ZZZZ Global cut - cuts off optimal solution!\n"); |
---|
5715 | // } |
---|
5716 | // } |
---|
5717 | // add to global list |
---|
5718 | globalCuts_.insert(*thisCut); |
---|
5719 | generator_[i]->incrementNumberCutsInTotal(); |
---|
5720 | } |
---|
5721 | } |
---|
5722 | } |
---|
5723 | } |
---|
5724 | // int numberCuts = theseCuts.sizeColCuts(); |
---|
5725 | // for (i=0;i<numberCuts;i++) { |
---|
5726 | // const OsiColCut * thisCut = theseCuts.colCutPtr(i); |
---|
5727 | // if (thisCut->globallyValid()) { |
---|
5728 | // // add to global list |
---|
5729 | // globalCuts_.insert(*thisCut); |
---|
5730 | // } |
---|
5731 | // } |
---|
5732 | //have to retrieve the solution and its value from the nlp |
---|
5733 | } |
---|
5734 | double newObjectiveValue=cutoff; |
---|
5735 | if(solverCharacteristics_->solution(newObjectiveValue, |
---|
5736 | const_cast<double *> (solution), |
---|
5737 | numberColumns)) |
---|
5738 | { |
---|
5739 | objectiveValue = newObjectiveValue; |
---|
5740 | } |
---|
5741 | else |
---|
5742 | { |
---|
5743 | objectiveValue = 2e50; |
---|
5744 | } |
---|
5745 | if (!solutionComesFromNlp && fixVariables) |
---|
5746 | { |
---|
5747 | for (int iColumn = 0 ; iColumn < numberColumns ; iColumn++) |
---|
5748 | { |
---|
5749 | solver_->setColLower(iColumn,saveLower[iColumn]) ; |
---|
5750 | solver_->setColUpper(iColumn,saveUpper[iColumn]) ; |
---|
5751 | } |
---|
5752 | delete [] saveLower; |
---|
5753 | delete [] saveUpper; |
---|
5754 | } |
---|
5755 | return objectiveValue; |
---|
5756 | } |
---|
5757 | } |
---|
5758 | |
---|
5759 | /* |
---|
5760 | Call this routine from anywhere when a solution is found. The solution |
---|
5761 | vector is assumed to contain one value for each structural variable. |
---|
5762 | |
---|
5763 | The first action is to run checkSolution() to confirm the objective and |
---|
5764 | feasibility. If this check causes the solution to be rejected, we're done. |
---|
5765 | If fixVariables = true, the variable bounds held by the continuous solver |
---|
5766 | will be left fixed to the values in the solution; otherwise they are |
---|
5767 | restored to the original values. |
---|
5768 | |
---|
5769 | If the solution is accepted, install it as the best solution. |
---|
5770 | |
---|
5771 | The routine also contains a hook to run any cut generators that are flagged |
---|
5772 | to run when a new solution is discovered. There's a potential hazard because |
---|
5773 | the cut generators see the continuous solver >after< possible restoration of |
---|
5774 | original bounds (which may well invalidate the solution). |
---|
5775 | */ |
---|
5776 | |
---|
5777 | void |
---|
5778 | CbcModel::setBestSolution (CBC_Message how, |
---|
5779 | double & objectiveValue, const double *solution, |
---|
5780 | bool fixVariables) |
---|
5781 | |
---|
5782 | { |
---|
5783 | if (!solverCharacteristics_->solutionAddsCuts()) { |
---|
5784 | // Can trust solution |
---|
5785 | double cutoff = CoinMin(getCutoff(),bestObjective_) ; |
---|
5786 | |
---|
5787 | /* |
---|
5788 | Double check the solution to catch pretenders. |
---|
5789 | */ |
---|
5790 | objectiveValue = checkSolution(cutoff,solution,fixVariables); |
---|
5791 | if (objectiveValue > cutoff) { |
---|
5792 | if (objectiveValue>1.0e30) |
---|
5793 | handler_->message(CBC_NOTFEAS1, messages_) << CoinMessageEol ; |
---|
5794 | else |
---|
5795 | handler_->message(CBC_NOTFEAS2, messages_) |
---|
5796 | << objectiveValue << cutoff << CoinMessageEol ; |
---|
5797 | } else { |
---|
5798 | /* |
---|
5799 | We have a winner. Install it as the new incumbent. |
---|
5800 | Bump the objective cutoff value and solution counts. Give the user the |
---|
5801 | good news. |
---|
5802 | */ |
---|
5803 | bestObjective_ = objectiveValue; |
---|
5804 | int numberColumns = solver_->getNumCols(); |
---|
5805 | if (!bestSolution_) |
---|
5806 | bestSolution_ = new double[numberColumns]; |
---|
5807 | CoinCopyN(solution,numberColumns,bestSolution_); |
---|
5808 | |
---|
5809 | cutoff = bestObjective_-dblParam_[CbcCutoffIncrement]; |
---|
5810 | // This is not correct - that way cutoff can go up if maximization |
---|
5811 | //double direction = solver_->getObjSense(); |
---|
5812 | //setCutoff(cutoff*direction); |
---|
5813 | setCutoff(cutoff); |
---|
5814 | |
---|
5815 | if (how==CBC_ROUNDING) |
---|
5816 | numberHeuristicSolutions_++; |
---|
5817 | numberSolutions_++; |
---|
5818 | if (numberHeuristicSolutions_==numberSolutions_) |
---|
5819 | stateOfSearch_ = 1; |
---|
5820 | else |
---|
5821 | stateOfSearch_ = 2; |
---|
5822 | |
---|
5823 | handler_->message(how,messages_) |
---|
5824 | <<bestObjective_<<numberIterations_ |
---|
5825 | <<numberNodes_<<getCurrentSeconds() |
---|
5826 | <<CoinMessageEol; |
---|
5827 | /* |
---|
5828 | Now step through the cut generators and see if any of them are flagged to |
---|
5829 | run when a new solution is discovered. Only global cuts are useful. (The |
---|
5830 | solution being evaluated may not correspond to the current location in the |
---|
5831 | search tree --- discovered by heuristic, for example.) |
---|
5832 | */ |
---|
5833 | OsiCuts theseCuts; |
---|
5834 | int i; |
---|
5835 | int lastNumberCuts=0; |
---|
5836 | for (i=0;i<numberCutGenerators_;i++) { |
---|
5837 | bool generate = generator_[i]->atSolution(); |
---|
5838 | // skip if not optimal and should be (maybe a cut generator has fixed variables) |
---|
5839 | if (generator_[i]->needsOptimalBasis()&&!solver_->basisIsAvailable()) |
---|
5840 | generate=false; |
---|
5841 | if (generate) { |
---|
5842 | generator_[i]->generateCuts(theseCuts,true,NULL); |
---|
5843 | int numberCuts = theseCuts.sizeRowCuts(); |
---|
5844 | for (int j=lastNumberCuts;j<numberCuts;j++) { |
---|
5845 | const OsiRowCut * thisCut = theseCuts.rowCutPtr(j); |
---|
5846 | if (thisCut->globallyValid()) { |
---|
5847 | if ((specialOptions_&1)!=0) { |
---|
5848 | /* As these are global cuts - |
---|
5849 | a) Always get debugger object |
---|
5850 | b) Not fatal error to cutoff optimal (if we have just got optimal) |
---|
5851 | */ |
---|
5852 | const OsiRowCutDebugger *debugger = solver_->getRowCutDebuggerAlways() ; |
---|
5853 | if (debugger) { |
---|
5854 | if(debugger->invalidCut(*thisCut)) |
---|
5855 | printf("ZZZZ Global cut - cuts off optimal solution!\n"); |
---|
5856 | } |
---|
5857 | } |
---|
5858 | // add to global list |
---|
5859 | globalCuts_.insert(*thisCut); |
---|
5860 | generator_[i]->incrementNumberCutsInTotal(); |
---|
5861 | } |
---|
5862 | } |
---|
5863 | } |
---|
5864 | } |
---|
5865 | int numberCuts = theseCuts.sizeColCuts(); |
---|
5866 | for (i=0;i<numberCuts;i++) { |
---|
5867 | const OsiColCut * thisCut = theseCuts.colCutPtr< |
---|