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 | #include <cassert> |
---|
12 | #include <cmath> |
---|
13 | #include <cfloat> |
---|
14 | |
---|
15 | #include "OsiSolverInterface.hpp" |
---|
16 | #include "CoinWarmStartBasis.hpp" |
---|
17 | #include "CoinPackedMatrix.hpp" |
---|
18 | #include "CoinHelperFunctions.hpp" |
---|
19 | #include "CbcBranchActual.hpp" |
---|
20 | #include "CbcBranchDynamic.hpp" |
---|
21 | #include "CbcHeuristic.hpp" |
---|
22 | #include "CbcModel.hpp" |
---|
23 | #include "CbcStatistics.hpp" |
---|
24 | #include "CbcStrategy.hpp" |
---|
25 | #include "CbcMessage.hpp" |
---|
26 | #include "OsiRowCut.hpp" |
---|
27 | #include "OsiColCut.hpp" |
---|
28 | #include "OsiRowCutDebugger.hpp" |
---|
29 | #include "OsiCuts.hpp" |
---|
30 | #include "CbcCountRowCut.hpp" |
---|
31 | #include "CbcCutGenerator.hpp" |
---|
32 | #include "CbcFeasibilityBase.hpp" |
---|
33 | // include Probing |
---|
34 | #include "CglProbing.hpp" |
---|
35 | |
---|
36 | #define COIN_USE_CLP |
---|
37 | #ifdef COIN_USE_CLP |
---|
38 | // include Presolve from Clp |
---|
39 | #include "ClpPresolve.hpp" |
---|
40 | #include "OsiClpSolverInterface.hpp" |
---|
41 | #include "ClpEventHandler.hpp" |
---|
42 | #endif |
---|
43 | |
---|
44 | #include "CoinTime.hpp" |
---|
45 | |
---|
46 | #include "CbcCompareActual.hpp" |
---|
47 | #include "CbcTree.hpp" |
---|
48 | /* Various functions local to CbcModel.cpp */ |
---|
49 | |
---|
50 | namespace { |
---|
51 | |
---|
52 | //------------------------------------------------------------------- |
---|
53 | // Returns the greatest common denominator of two |
---|
54 | // positive integers, a and b, found using Euclid's algorithm |
---|
55 | //------------------------------------------------------------------- |
---|
56 | static int gcd(int a, int b) |
---|
57 | { |
---|
58 | int remainder = -1; |
---|
59 | // make sure a<=b (will always remain so) |
---|
60 | if(a > b) { |
---|
61 | // Swap a and b |
---|
62 | int temp = a; |
---|
63 | a = b; |
---|
64 | b = temp; |
---|
65 | } |
---|
66 | // if zero then gcd is nonzero (zero may occur in rhs of packed) |
---|
67 | if (!a) { |
---|
68 | if (b) { |
---|
69 | return b; |
---|
70 | } else { |
---|
71 | printf("**** gcd given two zeros!!\n"); |
---|
72 | abort(); |
---|
73 | } |
---|
74 | } |
---|
75 | while (remainder) { |
---|
76 | remainder = b % a; |
---|
77 | b = a; |
---|
78 | a = remainder; |
---|
79 | } |
---|
80 | return b; |
---|
81 | } |
---|
82 | |
---|
83 | |
---|
84 | |
---|
85 | #ifdef CHECK_NODE_FULL |
---|
86 | |
---|
87 | /* |
---|
88 | Routine to verify that tree linkage is correct. The invariant that is tested |
---|
89 | is |
---|
90 | |
---|
91 | reference count = (number of actual references) + (number of branches left) |
---|
92 | |
---|
93 | The routine builds a set of paired arrays, info and count, by traversing the |
---|
94 | tree. Each CbcNodeInfo is recorded in info, and the number of times it is |
---|
95 | referenced (via the parent field) is recorded in count. Then a final check is |
---|
96 | made to see if the numberPointingToThis_ field agrees. |
---|
97 | */ |
---|
98 | |
---|
99 | void verifyTreeNodes (const CbcTree * branchingTree, const CbcModel &model) |
---|
100 | |
---|
101 | { printf("*** CHECKING tree after %d nodes\n",model.getNodeCount()) ; |
---|
102 | |
---|
103 | int j ; |
---|
104 | int nNodes = branchingTree->size() ; |
---|
105 | # define MAXINFO 1000 |
---|
106 | int *count = new int [MAXINFO] ; |
---|
107 | CbcNodeInfo **info = new CbcNodeInfo*[MAXINFO] ; |
---|
108 | int nInfo = 0 ; |
---|
109 | /* |
---|
110 | Collect all CbcNodeInfo objects in info, by starting from each live node and |
---|
111 | traversing back to the root. Nodes in the live set should have unexplored |
---|
112 | branches remaining. |
---|
113 | |
---|
114 | TODO: The `while (nodeInfo)' loop could be made to break on reaching a |
---|
115 | common ancester (nodeInfo is found in info[k]). Alternatively, the |
---|
116 | check could change to signal an error if nodeInfo is not found above a |
---|
117 | common ancestor. |
---|
118 | */ |
---|
119 | for (j = 0 ; j < nNodes ; j++) |
---|
120 | { CbcNode *node = branchingTree->nodePointer(j) ; |
---|
121 | CbcNodeInfo *nodeInfo = node->nodeInfo() ; |
---|
122 | int change = node->nodeInfo()->numberBranchesLeft() ; |
---|
123 | assert(change) ; |
---|
124 | while (nodeInfo) |
---|
125 | { int k ; |
---|
126 | for (k = 0 ; k < nInfo ; k++) |
---|
127 | { if (nodeInfo == info[k]) break ; } |
---|
128 | if (k == nInfo) |
---|
129 | { assert(nInfo < MAXINFO) ; |
---|
130 | nInfo++ ; |
---|
131 | info[k] = nodeInfo ; |
---|
132 | count[k] = 0 ; } |
---|
133 | nodeInfo = nodeInfo->parent() ; } } |
---|
134 | /* |
---|
135 | Walk the info array. For each nodeInfo, look up its parent in info and |
---|
136 | increment the corresponding count. |
---|
137 | */ |
---|
138 | for (j = 0 ; j < nInfo ; j++) |
---|
139 | { CbcNodeInfo *nodeInfo = info[j] ; |
---|
140 | nodeInfo = nodeInfo->parent() ; |
---|
141 | if (nodeInfo) |
---|
142 | { int k ; |
---|
143 | for (k = 0 ; k < nInfo ; k++) |
---|
144 | { if (nodeInfo == info[k]) break ; } |
---|
145 | assert (k < nInfo) ; |
---|
146 | count[k]++ ; } } |
---|
147 | /* |
---|
148 | Walk the info array one more time and check that the invariant holds. The |
---|
149 | number of references (numberPointingToThis()) should equal the sum of the |
---|
150 | number of actual references (held in count[]) plus the number of potential |
---|
151 | references (unexplored branches, numberBranchesLeft()). |
---|
152 | */ |
---|
153 | for (j = 0;j < nInfo;j++) { |
---|
154 | CbcNodeInfo * nodeInfo = info[j] ; |
---|
155 | if (nodeInfo) { |
---|
156 | int k ; |
---|
157 | for (k = 0;k < nInfo;k++) |
---|
158 | if (nodeInfo == info[k]) |
---|
159 | break ; |
---|
160 | printf("Nodeinfo %x - %d left, %d count\n", |
---|
161 | nodeInfo, |
---|
162 | nodeInfo->numberBranchesLeft(), |
---|
163 | nodeInfo->numberPointingToThis()) ; |
---|
164 | assert(nodeInfo->numberPointingToThis() == |
---|
165 | count[k]+nodeInfo->numberBranchesLeft()) ; } } |
---|
166 | |
---|
167 | delete [] count ; |
---|
168 | delete [] info ; |
---|
169 | |
---|
170 | return ; } |
---|
171 | |
---|
172 | #endif /* CHECK_NODE_FULL */ |
---|
173 | |
---|
174 | |
---|
175 | |
---|
176 | #ifdef CHECK_CUT_COUNTS |
---|
177 | |
---|
178 | /* |
---|
179 | Routine to verify that cut reference counts are correct. |
---|
180 | */ |
---|
181 | void verifyCutCounts (const CbcTree * branchingTree, CbcModel &model) |
---|
182 | |
---|
183 | { printf("*** CHECKING cuts after %d nodes\n",model.getNodeCount()) ; |
---|
184 | |
---|
185 | int j ; |
---|
186 | int nNodes = branchingTree->size() ; |
---|
187 | |
---|
188 | /* |
---|
189 | cut.tempNumber_ exists for the purpose of doing this verification. Clear it |
---|
190 | in all cuts. We traverse the tree by starting from each live node and working |
---|
191 | back to the root. At each CbcNodeInfo, check for cuts. |
---|
192 | */ |
---|
193 | for (j = 0 ; j < nNodes ; j++) |
---|
194 | { CbcNode *node = branchingTree->nodePointer(j) ; |
---|
195 | CbcNodeInfo * nodeInfo = node->nodeInfo() ; |
---|
196 | assert (node->nodeInfo()->numberBranchesLeft()) ; |
---|
197 | while (nodeInfo) |
---|
198 | { int k ; |
---|
199 | for (k = 0 ; k < nodeInfo->numberCuts() ; k++) |
---|
200 | { CbcCountRowCut *cut = nodeInfo->cuts()[k] ; |
---|
201 | if (cut) cut->tempNumber_ = 0; } |
---|
202 | nodeInfo = nodeInfo->parent() ; } } |
---|
203 | /* |
---|
204 | Walk the live set again, this time collecting the list of cuts in use at each |
---|
205 | node. addCuts1 will collect the cuts in model.addedCuts_. Take into account |
---|
206 | that when we recreate the basis for a node, we compress out the slack cuts. |
---|
207 | */ |
---|
208 | for (j = 0 ; j < nNodes ; j++) |
---|
209 | { CoinWarmStartBasis *debugws = model.getEmptyBasis() ; |
---|
210 | CbcNode *node = branchingTree->nodePointer(j) ; |
---|
211 | CbcNodeInfo *nodeInfo = node->nodeInfo(); |
---|
212 | int change = node->nodeInfo()->numberBranchesLeft() ; |
---|
213 | printf("Node %d %x (info %x) var %d way %d obj %g",j,node, |
---|
214 | node->nodeInfo(),node->variable(),node->way(), |
---|
215 | node->objectiveValue()) ; |
---|
216 | |
---|
217 | model.addCuts1(node,debugws) ; |
---|
218 | |
---|
219 | int i ; |
---|
220 | int numberRowsAtContinuous = model.numberRowsAtContinuous() ; |
---|
221 | CbcCountRowCut **addedCuts = model.addedCuts() ; |
---|
222 | for (i = 0 ; i < model.currentNumberCuts() ; i++) |
---|
223 | { CoinWarmStartBasis::Status status = |
---|
224 | debugws->getArtifStatus(i+numberRowsAtContinuous) ; |
---|
225 | if (status != CoinWarmStartBasis::basic && addedCuts[i]) |
---|
226 | { addedCuts[i]->tempNumber_ += change ; } } |
---|
227 | |
---|
228 | while (nodeInfo) |
---|
229 | { nodeInfo = nodeInfo->parent() ; |
---|
230 | if (nodeInfo) printf(" -> %x",nodeInfo); } |
---|
231 | printf("\n") ; |
---|
232 | delete debugws ; } |
---|
233 | /* |
---|
234 | The moment of truth: We've tallied up the references by direct scan of the |
---|
235 | search tree. Check for agreement with the count in the cut. |
---|
236 | |
---|
237 | TODO: Rewrite to check and print mismatch only when tempNumber_ == 0? |
---|
238 | */ |
---|
239 | for (j = 0 ; j < nNodes ; j++) |
---|
240 | { CbcNode *node = branchingTree->nodePointer(j) ; |
---|
241 | CbcNodeInfo *nodeInfo = node->nodeInfo(); |
---|
242 | while (nodeInfo) |
---|
243 | { int k ; |
---|
244 | for (k = 0 ; k < nodeInfo->numberCuts() ; k++) |
---|
245 | { CbcCountRowCut *cut = nodeInfo->cuts()[k] ; |
---|
246 | if (cut && cut->tempNumber_ >= 0) |
---|
247 | { if (cut->tempNumber_ != cut->numberPointingToThis()) |
---|
248 | printf("mismatch %x %d %x %d %d\n",nodeInfo,k, |
---|
249 | cut,cut->tempNumber_,cut->numberPointingToThis()) ; |
---|
250 | else |
---|
251 | printf(" match %x %d %x %d %d\n", nodeInfo,k, |
---|
252 | cut,cut->tempNumber_,cut->numberPointingToThis()) ; |
---|
253 | cut->tempNumber_ = -1 ; } } |
---|
254 | nodeInfo = nodeInfo->parent() ; } } |
---|
255 | |
---|
256 | return ; } |
---|
257 | |
---|
258 | #endif /* CHECK_CUT_COUNTS */ |
---|
259 | |
---|
260 | } |
---|
261 | |
---|
262 | /* End unnamed namespace for CbcModel.cpp */ |
---|
263 | |
---|
264 | |
---|
265 | |
---|
266 | void |
---|
267 | CbcModel::analyzeObjective () |
---|
268 | /* |
---|
269 | Try to find a minimum change in the objective function. The first scan |
---|
270 | checks that there are no continuous variables with non-zero coefficients, |
---|
271 | and grabs the largest objective coefficient associated with an unfixed |
---|
272 | integer variable. The second scan attempts to scale up the objective |
---|
273 | coefficients to a point where they are sufficiently close to integer that |
---|
274 | we can pretend they are integer, and calculate a gcd over the coefficients |
---|
275 | of interest. This will be the minimum increment for the scaled coefficients. |
---|
276 | The final action is to scale the increment back for the original coefficients |
---|
277 | and install it, if it's better than the existing value. |
---|
278 | |
---|
279 | John's note: We could do better than this. |
---|
280 | |
---|
281 | John's second note - apologies for changing s to z |
---|
282 | */ |
---|
283 | { const double *objective = getObjCoefficients() ; |
---|
284 | const double *lower = getColLower() ; |
---|
285 | const double *upper = getColUpper() ; |
---|
286 | /* |
---|
287 | Take a first scan to see if there are unfixed continuous variables in the |
---|
288 | objective. If so, the minimum objective change could be arbitrarily small. |
---|
289 | Also pick off the maximum coefficient of an unfixed integer variable. |
---|
290 | |
---|
291 | If the objective is found to contain only integer variables, set the |
---|
292 | fathoming discipline to strict. |
---|
293 | */ |
---|
294 | double maximumCost = 0.0 ; |
---|
295 | bool possibleMultiple = true ; |
---|
296 | int iColumn ; |
---|
297 | int numberColumns = getNumCols() ; |
---|
298 | for (iColumn = 0 ; iColumn < numberColumns ; iColumn++) |
---|
299 | { if (upper[iColumn] > lower[iColumn]+1.0e-8) |
---|
300 | { if (isInteger(iColumn)) |
---|
301 | maximumCost = CoinMax(maximumCost,fabs(objective[iColumn])) ; |
---|
302 | else if (objective[iColumn]) |
---|
303 | possibleMultiple = false ; } } |
---|
304 | setIntParam(CbcModel::CbcFathomDiscipline,possibleMultiple) ; |
---|
305 | /* |
---|
306 | If a nontrivial increment is possible, try and figure it out. We're looking |
---|
307 | for gcd(c<j>) for all c<j> that are coefficients of unfixed integer |
---|
308 | variables. Since the c<j> might not be integers, try and inflate them |
---|
309 | sufficiently that they look like integers (and we'll deflate the gcd |
---|
310 | later). |
---|
311 | |
---|
312 | 2520.0 is used as it is a nice multiple of 2,3,5,7 |
---|
313 | */ |
---|
314 | if (possibleMultiple&&maximumCost) |
---|
315 | { int increment = 0 ; |
---|
316 | double multiplier = 2520.0 ; |
---|
317 | while (10.0*multiplier*maximumCost < 1.0e8) |
---|
318 | multiplier *= 10.0 ; |
---|
319 | |
---|
320 | for (iColumn = 0 ; iColumn < numberColumns ; iColumn++) |
---|
321 | { if (upper[iColumn] > lower[iColumn]+1.0e-8) |
---|
322 | { if (isInteger(iColumn)&&objective[iColumn]) |
---|
323 | { double value = fabs(objective[iColumn])*multiplier ; |
---|
324 | int nearest = (int) floor(value+0.5) ; |
---|
325 | if (fabs(value-floor(value+0.5)) > 1.0e-8) |
---|
326 | { increment = 0 ; |
---|
327 | break ; } |
---|
328 | else if (!increment) |
---|
329 | { increment = nearest ; } |
---|
330 | else |
---|
331 | { increment = gcd(increment,nearest) ; } } } } |
---|
332 | /* |
---|
333 | If the increment beats the current value for objective change, install it. |
---|
334 | */ |
---|
335 | if (increment) |
---|
336 | { double value = increment ; |
---|
337 | double cutoff = getDblParam(CbcModel::CbcCutoffIncrement) ; |
---|
338 | value /= multiplier ; |
---|
339 | if (value*0.999 > cutoff) |
---|
340 | { messageHandler()->message(CBC_INTEGERINCREMENT, |
---|
341 | messages()) |
---|
342 | << value << CoinMessageEol ; |
---|
343 | setDblParam(CbcModel::CbcCutoffIncrement,value*0.999) ; } } } |
---|
344 | |
---|
345 | return ; |
---|
346 | } |
---|
347 | |
---|
348 | |
---|
349 | |
---|
350 | /** |
---|
351 | \todo |
---|
352 | Normally, it looks like we enter here from command dispatch in the main |
---|
353 | routine, after calling the solver for an initial solution |
---|
354 | (CbcModel::initialSolve, which simply calls the solver's initialSolve |
---|
355 | routine.) The first thing we do is call resolve. Presumably there are |
---|
356 | circumstances where this is nontrivial? There's also a call from |
---|
357 | CbcModel::originalModel (tied up with integer presolve), which should be |
---|
358 | checked. |
---|
359 | |
---|
360 | */ |
---|
361 | |
---|
362 | /* |
---|
363 | The overall flow can be divided into three stages: |
---|
364 | * Prep: Check that the lp relaxation remains feasible at the root. If so, |
---|
365 | do all the setup for B&C. |
---|
366 | * Process the root node: Generate cuts, apply heuristics, and in general do |
---|
367 | the best we can to resolve the problem without B&C. |
---|
368 | * Do B&C search until we hit a limit or exhaust the search tree. |
---|
369 | |
---|
370 | Keep in mind that in general there is no node in the search tree that |
---|
371 | corresponds to the active subproblem. The active subproblem is represented |
---|
372 | by the current state of the model, of the solver, and of the constraint |
---|
373 | system held by the solver. |
---|
374 | */ |
---|
375 | |
---|
376 | void CbcModel::branchAndBound(int doStatistics) |
---|
377 | |
---|
378 | { |
---|
379 | // Set up strategies |
---|
380 | if (strategy_) { |
---|
381 | strategy_->setupCutGenerators(*this); |
---|
382 | strategy_->setupHeuristics(*this); |
---|
383 | // Set strategy print level to models |
---|
384 | strategy_->setupPrinting(*this,handler_->logLevel()); |
---|
385 | strategy_->setupOther(*this); |
---|
386 | } |
---|
387 | bool eventHappened=false; |
---|
388 | ClpEventHandler * eventHandler=NULL; |
---|
389 | #ifdef COIN_USE_CLP |
---|
390 | { |
---|
391 | OsiClpSolverInterface * clpSolver |
---|
392 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
393 | if (clpSolver) { |
---|
394 | ClpSimplex * clpSimplex = clpSolver->getModelPtr(); |
---|
395 | eventHandler = clpSimplex->eventHandler(); |
---|
396 | } |
---|
397 | } |
---|
398 | #endif |
---|
399 | if (!nodeCompare_) |
---|
400 | nodeCompare_=new CbcCompareDefault();; |
---|
401 | if (!problemFeasibility_) |
---|
402 | problemFeasibility_=new CbcFeasibilityBase(); |
---|
403 | # ifdef CBC_DEBUG |
---|
404 | std::string problemName ; |
---|
405 | solver_->getStrParam(OsiProbName,problemName) ; |
---|
406 | printf("Problem name - %s\n",problemName.c_str()) ; |
---|
407 | solver_->setHintParam(OsiDoReducePrint,false,OsiHintDo,0) ; |
---|
408 | # endif |
---|
409 | /* |
---|
410 | Assume we're done, and see if we're proven wrong. |
---|
411 | */ |
---|
412 | status_ = 0 ; |
---|
413 | secondaryStatus_ = 0; |
---|
414 | phase_=0; |
---|
415 | /* |
---|
416 | Scan the variables, noting the integer variables. Create an |
---|
417 | CbcSimpleInteger object for each integer variable. |
---|
418 | */ |
---|
419 | findIntegers(false) ; |
---|
420 | // If dynamic pseudo costs then do |
---|
421 | if (numberBeforeTrust_>0) |
---|
422 | convertToDynamic(); |
---|
423 | |
---|
424 | /* |
---|
425 | Ensure that objects on the lists of CbcObjects, heuristics, and cut |
---|
426 | generators attached to this model all refer to this model. |
---|
427 | */ |
---|
428 | synchronizeModel() ; |
---|
429 | /* |
---|
430 | Capture a time stamp before we start. |
---|
431 | */ |
---|
432 | dblParam_[CbcStartSeconds] = CoinCpuTime(); |
---|
433 | // Set so we can tell we are in initial phase in resolve |
---|
434 | continuousObjective_ = -COIN_DBL_MAX ; |
---|
435 | /* |
---|
436 | Solve the relaxation. |
---|
437 | |
---|
438 | Apparently there are circumstances where this will be non-trivial --- i.e., |
---|
439 | we've done something since initialSolve that's trashed the solution to the |
---|
440 | continuous relaxation. |
---|
441 | */ |
---|
442 | bool feasible = resolve() ; |
---|
443 | if (problemFeasibility_->feasible(this,0)<0) { |
---|
444 | feasible=false; // pretend infeasible |
---|
445 | } |
---|
446 | /* |
---|
447 | If the linear relaxation of the root is infeasible, bail out now. Otherwise, |
---|
448 | continue with processing the root node. |
---|
449 | */ |
---|
450 | if (!feasible) |
---|
451 | { handler_->message(CBC_INFEAS,messages_)<< CoinMessageEol ; |
---|
452 | status_ = 0 ; |
---|
453 | secondaryStatus_ = 1; |
---|
454 | originalContinuousObjective_ = COIN_DBL_MAX; |
---|
455 | return ; } |
---|
456 | // Save objective (just so user can access it) |
---|
457 | originalContinuousObjective_ = solver_->getObjValue(); |
---|
458 | bestPossibleObjective_=originalContinuousObjective_; |
---|
459 | sumChangeObjective1_=0.0; |
---|
460 | sumChangeObjective2_=0.0; |
---|
461 | /* |
---|
462 | OsiRowCutDebugger knows an optimal answer for a subset of MIP problems. |
---|
463 | Assuming it recognises the problem, when called upon it will check a cut to |
---|
464 | see if it cuts off the optimal answer. |
---|
465 | */ |
---|
466 | // If debugger exists set specialOptions_ bit |
---|
467 | if (solver_->getRowCutDebuggerAlways()) |
---|
468 | specialOptions_ |= 1; |
---|
469 | |
---|
470 | # ifdef CBC_DEBUG |
---|
471 | if ((specialOptions_&1)==0) |
---|
472 | solver_->activateRowCutDebugger(problemName.c_str()) ; |
---|
473 | if (solver_->getRowCutDebuggerAlways()) |
---|
474 | specialOptions_ |= 1; |
---|
475 | # endif |
---|
476 | |
---|
477 | /* |
---|
478 | Begin setup to process a feasible root node. |
---|
479 | */ |
---|
480 | bestObjective_ = CoinMin(bestObjective_,1.0e50) ; |
---|
481 | numberSolutions_ = 0 ; |
---|
482 | stateOfSearch_=0; |
---|
483 | numberHeuristicSolutions_ = 0 ; |
---|
484 | // Everything is minimization |
---|
485 | double cutoff=getCutoff() ; |
---|
486 | double direction = solver_->getObjSense() ; |
---|
487 | if (cutoff < 1.0e20&&direction<0.0) |
---|
488 | messageHandler()->message(CBC_CUTOFF_WARNING1, |
---|
489 | messages()) |
---|
490 | << cutoff << -cutoff << CoinMessageEol ; |
---|
491 | if (cutoff > bestObjective_) |
---|
492 | cutoff = bestObjective_ ; |
---|
493 | setCutoff(cutoff) ; |
---|
494 | /* |
---|
495 | We probably already have a current solution, but just in case ... |
---|
496 | */ |
---|
497 | int numberColumns = getNumCols() ; |
---|
498 | if (!currentSolution_) |
---|
499 | currentSolution_ = new double[numberColumns] ; |
---|
500 | testSolution_ = currentSolution_; |
---|
501 | /* |
---|
502 | Create a copy of the solver, thus capturing the original (root node) |
---|
503 | constraint system (aka the continuous system). |
---|
504 | */ |
---|
505 | continuousSolver_ = solver_->clone() ; |
---|
506 | #ifdef COIN_USE_CLP |
---|
507 | { |
---|
508 | OsiClpSolverInterface * clpSolver |
---|
509 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
510 | if (clpSolver) { |
---|
511 | ClpSimplex * clpSimplex = clpSolver->getModelPtr(); |
---|
512 | // take off names |
---|
513 | clpSimplex->dropNames(); |
---|
514 | } |
---|
515 | } |
---|
516 | #endif |
---|
517 | |
---|
518 | numberRowsAtContinuous_ = getNumRows() ; |
---|
519 | /* |
---|
520 | Check the objective to see if we can deduce a nontrivial increment. If |
---|
521 | it's better than the current value for CbcCutoffIncrement, it'll be |
---|
522 | installed. |
---|
523 | */ |
---|
524 | analyzeObjective() ; |
---|
525 | /* |
---|
526 | Set up for cut generation. addedCuts_ holds the cuts which are relevant for |
---|
527 | the active subproblem. whichGenerator will be used to record the generator |
---|
528 | that produced a given cut. |
---|
529 | */ |
---|
530 | int maximumWhich = 1000 ; |
---|
531 | int * whichGenerator = new int[maximumWhich] ; |
---|
532 | int currentNumberCuts = 0 ; |
---|
533 | maximumNumberCuts_ = 0 ; |
---|
534 | currentNumberCuts_ = 0 ; |
---|
535 | delete [] addedCuts_ ; |
---|
536 | addedCuts_ = NULL ; |
---|
537 | /* |
---|
538 | Set up an empty heap and associated data structures to hold the live set |
---|
539 | (problems which require further exploration). |
---|
540 | */ |
---|
541 | tree_->setComparison(*nodeCompare_) ; |
---|
542 | /* |
---|
543 | Used to record the path from a node to the root of the search tree, so that |
---|
544 | we can then traverse from the root to the node when restoring a subproblem. |
---|
545 | */ |
---|
546 | maximumDepth_ = 10 ; |
---|
547 | delete [] walkback_ ; |
---|
548 | walkback_ = new CbcNodeInfo * [maximumDepth_] ; |
---|
549 | /* |
---|
550 | Used to generate bound edits for CbcPartialNodeInfo. |
---|
551 | */ |
---|
552 | double * lowerBefore = new double [numberColumns] ; |
---|
553 | double * upperBefore = new double [numberColumns] ; |
---|
554 | /* |
---|
555 | |
---|
556 | Generate cuts at the root node and reoptimise. solveWithCuts does the heavy |
---|
557 | lifting. It will iterate a generate/reoptimise loop (including reduced cost |
---|
558 | fixing) until no cuts are generated, the change in objective falls off, or |
---|
559 | the limit on the number of rounds of cut generation is exceeded. |
---|
560 | |
---|
561 | At the end of all this, any cuts will be recorded in cuts and also |
---|
562 | installed in the solver's constraint system. We'll have reoptimised, and |
---|
563 | removed any slack cuts (numberOldActiveCuts and numberNewCuts have been |
---|
564 | adjusted accordingly). |
---|
565 | |
---|
566 | Tell cut generators they can be a bit more aggressive at root node |
---|
567 | |
---|
568 | TODO: Why don't we make a copy of the solution after solveWithCuts? |
---|
569 | TODO: If numberUnsatisfied == 0, don't we have a solution? |
---|
570 | */ |
---|
571 | phase_=1; |
---|
572 | int iCutGenerator; |
---|
573 | for (iCutGenerator = 0;iCutGenerator<numberCutGenerators_;iCutGenerator++) { |
---|
574 | CglCutGenerator * generator = generator_[iCutGenerator]->generator(); |
---|
575 | generator->setAggressiveness(generator->getAggressiveness()+100); |
---|
576 | } |
---|
577 | OsiCuts cuts ; |
---|
578 | int anyAction = -1 ; |
---|
579 | int numberOldActiveCuts = 0 ; |
---|
580 | int numberNewCuts = 0 ; |
---|
581 | // Array to mark solution |
---|
582 | delete [] usedInSolution_; |
---|
583 | usedInSolution_ = new int[numberColumns]; |
---|
584 | CoinZeroN(usedInSolution_,numberColumns); |
---|
585 | /* |
---|
586 | For printing totals and for CbcNode (numberNodes_) |
---|
587 | */ |
---|
588 | numberIterations_ = 0 ; |
---|
589 | numberNodes_ = 0 ; |
---|
590 | numberNodes2_ = 0 ; |
---|
591 | int maximumStatistics=0; |
---|
592 | CbcStatistics ** statistics = NULL; |
---|
593 | // Do on switch |
---|
594 | if (doStatistics) { |
---|
595 | maximumStatistics=10000; |
---|
596 | statistics = new CbcStatistics * [maximumStatistics]; |
---|
597 | memset(statistics,0,maximumStatistics*sizeof(CbcStatistics *)); |
---|
598 | } |
---|
599 | |
---|
600 | { int iObject ; |
---|
601 | int preferredWay ; |
---|
602 | int numberUnsatisfied = 0 ; |
---|
603 | memcpy(currentSolution_,solver_->getColSolution(), |
---|
604 | numberColumns*sizeof(double)) ; |
---|
605 | |
---|
606 | for (iObject = 0 ; iObject < numberObjects_ ; iObject++) |
---|
607 | { double infeasibility = |
---|
608 | object_[iObject]->infeasibility(preferredWay) ; |
---|
609 | if (infeasibility ) numberUnsatisfied++ ; } |
---|
610 | if (numberUnsatisfied) { |
---|
611 | feasible = solveWithCuts(cuts,maximumCutPassesAtRoot_, |
---|
612 | NULL,numberOldActiveCuts,numberNewCuts, |
---|
613 | maximumWhich, whichGenerator) ; |
---|
614 | } |
---|
615 | } |
---|
616 | // make cut generators less aggressive |
---|
617 | for (iCutGenerator = 0;iCutGenerator<numberCutGenerators_;iCutGenerator++) { |
---|
618 | CglCutGenerator * generator = generator_[iCutGenerator]->generator(); |
---|
619 | generator->setAggressiveness(generator->getAggressiveness()-100); |
---|
620 | } |
---|
621 | currentNumberCuts_ = numberNewCuts ; |
---|
622 | /* |
---|
623 | We've taken the continuous relaxation as far as we can. Time to branch. |
---|
624 | The first order of business is to actually create a node. chooseBranch |
---|
625 | currently uses strong branching to evaluate branch object candidates, |
---|
626 | unless forced back to simple branching. If chooseBranch concludes that a |
---|
627 | branching candidate is monotone (anyAction == -1) or infeasible (anyAction |
---|
628 | == -2) when forced to integer values, it returns here immediately. |
---|
629 | |
---|
630 | Monotone variables trigger a call to resolve(). If the problem remains |
---|
631 | feasible, try again to choose a branching variable. At the end of the loop, |
---|
632 | resolved == true indicates that some variables were fixed. |
---|
633 | |
---|
634 | Loss of feasibility will result in the deletion of newNode. |
---|
635 | */ |
---|
636 | |
---|
637 | bool resolved = false ; |
---|
638 | CbcNode *newNode = NULL ; |
---|
639 | if (feasible) |
---|
640 | { newNode = new CbcNode ; |
---|
641 | newNode->setObjectiveValue(direction*solver_->getObjValue()) ; |
---|
642 | anyAction = -1 ; |
---|
643 | // To make depth available we may need a fake node |
---|
644 | CbcNode fakeNode; |
---|
645 | if (!currentNode_) { |
---|
646 | // Not true if sub trees assert (!numberNodes_); |
---|
647 | currentNode_=&fakeNode; |
---|
648 | } |
---|
649 | phase_=3; |
---|
650 | // only allow twenty passes |
---|
651 | int numberPassesLeft=20; |
---|
652 | while (anyAction == -1) |
---|
653 | { |
---|
654 | if (numberBeforeTrust_<=0 ) { |
---|
655 | anyAction = newNode->chooseBranch(this,NULL,numberPassesLeft) ; |
---|
656 | } else { |
---|
657 | anyAction = newNode->chooseDynamicBranch(this,NULL,numberPassesLeft) ; |
---|
658 | if (anyAction==-3) |
---|
659 | anyAction = newNode->chooseBranch(this,NULL,numberPassesLeft) ; // dynamic did nothing |
---|
660 | } |
---|
661 | numberPassesLeft--; |
---|
662 | if (anyAction == -1) |
---|
663 | { feasible = resolve() ; |
---|
664 | if (problemFeasibility_->feasible(this,0)<0) { |
---|
665 | feasible=false; // pretend infeasible |
---|
666 | } |
---|
667 | resolved = true ; |
---|
668 | # ifdef CBC_DEBUG |
---|
669 | printf("Resolve (root) as something fixed, Obj value %g %d rows\n", |
---|
670 | solver_->getObjValue(), |
---|
671 | solver_->getNumRows()) ; |
---|
672 | # endif |
---|
673 | if (!feasible) anyAction = -2 ; } |
---|
674 | if (anyAction == -2||newNode->objectiveValue() >= cutoff) |
---|
675 | { delete newNode ; |
---|
676 | newNode = NULL ; |
---|
677 | feasible = false ; } } } |
---|
678 | /* |
---|
679 | At this point, the root subproblem is infeasible or fathomed by bound |
---|
680 | (newNode == NULL), or we're live with an objective value that satisfies the |
---|
681 | current objective cutoff. |
---|
682 | */ |
---|
683 | assert (!newNode || newNode->objectiveValue() <= cutoff) ; |
---|
684 | // Save address of root node as we don't want to delete it |
---|
685 | CbcNode * rootNode = newNode; |
---|
686 | /* |
---|
687 | The common case is that the lp relaxation is feasible but doesn't satisfy |
---|
688 | integrality (i.e., newNode->variable() >= 0, indicating we've been able to |
---|
689 | select a branching variable). Remove any cuts that have gone slack due to |
---|
690 | forcing monotone variables. Then tack on an CbcFullNodeInfo object and full |
---|
691 | basis (via createInfo()) and stash the new cuts in the nodeInfo (via |
---|
692 | addCuts()). If, by some miracle, we have an integral solution at the root |
---|
693 | (newNode->variable() < 0), takeOffCuts() will ensure that the solver holds |
---|
694 | a valid solution for use by setBestSolution(). |
---|
695 | */ |
---|
696 | CoinWarmStartBasis *lastws = 0 ; |
---|
697 | if (feasible && newNode->variable() >= 0) |
---|
698 | { if (resolved) |
---|
699 | { bool needValidSolution = (newNode->variable() < 0) ; |
---|
700 | takeOffCuts(cuts,whichGenerator,numberOldActiveCuts,numberNewCuts, |
---|
701 | needValidSolution) ; |
---|
702 | # ifdef CHECK_CUT_COUNTS |
---|
703 | { printf("Number of rows after chooseBranch fix (root)" |
---|
704 | "(active only) %d\n", |
---|
705 | numberRowsAtContinuous_+numberNewCuts+numberOldActiveCuts) ; |
---|
706 | const CoinWarmStartBasis* debugws = |
---|
707 | dynamic_cast <const CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
708 | debugws->print() ; |
---|
709 | delete debugws ; } |
---|
710 | # endif |
---|
711 | } |
---|
712 | newNode->createInfo(this,NULL,NULL,NULL,NULL,0,0) ; |
---|
713 | newNode->nodeInfo()->addCuts(cuts, |
---|
714 | newNode->numberBranches(),whichGenerator) ; |
---|
715 | /* |
---|
716 | Courtesy of createInfo, there's now a full basis stashed in |
---|
717 | newNode->nodeInfo_->basis_. We're about to make two more copies, lastws and |
---|
718 | model.basis_. |
---|
719 | |
---|
720 | (jf) With some thought I should be able to get rid of lastws and use |
---|
721 | basis_. |
---|
722 | (lh) I agree, but haven't pursued it to the end. |
---|
723 | */ |
---|
724 | if (basis_) delete basis_ ; |
---|
725 | basis_ = dynamic_cast<CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
726 | if (lastws) delete lastws ; |
---|
727 | lastws = dynamic_cast<CoinWarmStartBasis*>(basis_->clone()) ; } |
---|
728 | /* |
---|
729 | Continuous data to be used later |
---|
730 | */ |
---|
731 | continuousObjective_ = solver_->getObjValue()*solver_->getObjSense(); |
---|
732 | continuousInfeasibilities_ = 0 ; |
---|
733 | if (newNode) |
---|
734 | { continuousObjective_ = newNode->objectiveValue() ; |
---|
735 | delete [] continuousSolution_; |
---|
736 | continuousSolution_ = CoinCopyOfArray(solver_->getColSolution(), |
---|
737 | numberColumns); |
---|
738 | continuousInfeasibilities_ = newNode->numberUnsatisfied() ; } |
---|
739 | /* |
---|
740 | Bound may have changed so reset in objects |
---|
741 | */ |
---|
742 | { int i ; |
---|
743 | for (i = 0;i < numberObjects_;i++) |
---|
744 | object_[i]->resetBounds() ; } |
---|
745 | bool stoppedOnGap = false ; |
---|
746 | /* |
---|
747 | Feasible? Then we should have either a live node prepped for future |
---|
748 | expansion (indicated by variable() >= 0), or (miracle of miracles) an |
---|
749 | integral solution at the root node. |
---|
750 | |
---|
751 | initializeInfo sets the reference counts in the nodeInfo object. Since |
---|
752 | this node is still live, push it onto the heap that holds the live set. |
---|
753 | */ |
---|
754 | double bestValue = 0.0 ; |
---|
755 | if (newNode) { |
---|
756 | bestValue = newNode->objectiveValue(); |
---|
757 | if (newNode->variable() >= 0) { |
---|
758 | newNode->initializeInfo() ; |
---|
759 | tree_->push(newNode) ; |
---|
760 | if (statistics) { |
---|
761 | if (numberNodes2_==maximumStatistics) { |
---|
762 | maximumStatistics = 2*maximumStatistics; |
---|
763 | CbcStatistics ** temp = new CbcStatistics * [maximumStatistics]; |
---|
764 | memset(temp,0,maximumStatistics*sizeof(CbcStatistics *)); |
---|
765 | memcpy(temp,statistics,numberNodes2_*sizeof(CbcStatistics *)); |
---|
766 | delete [] statistics; |
---|
767 | statistics=temp; |
---|
768 | } |
---|
769 | assert (!statistics[numberNodes2_]); |
---|
770 | statistics[numberNodes2_]=new CbcStatistics(newNode); |
---|
771 | } |
---|
772 | numberNodes2_++; |
---|
773 | # ifdef CHECK_NODE |
---|
774 | printf("Node %x on tree\n",newNode) ; |
---|
775 | # endif |
---|
776 | } else { |
---|
777 | // continuous is integer |
---|
778 | double objectiveValue = newNode->objectiveValue(); |
---|
779 | setBestSolution(CBC_SOLUTION,objectiveValue, |
---|
780 | solver_->getColSolution()) ; |
---|
781 | delete newNode ; |
---|
782 | newNode = NULL ; |
---|
783 | } |
---|
784 | } |
---|
785 | |
---|
786 | if (printFrequency_ <= 0) { |
---|
787 | printFrequency_ = 1000 ; |
---|
788 | if (getNumCols() > 2000) |
---|
789 | printFrequency_ = 100 ; |
---|
790 | } |
---|
791 | /* Tell solver we are in Branch and Cut |
---|
792 | Could use last parameter for subtle differences */ |
---|
793 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,NULL) ; |
---|
794 | /* |
---|
795 | It is possible that strong branching fixes one variable and then the code goes round |
---|
796 | again and again. This can take too long. So we need to warn user - just once. |
---|
797 | */ |
---|
798 | int numberLongStrong=0; |
---|
799 | /* |
---|
800 | At last, the actual branch-and-cut search loop, which will iterate until |
---|
801 | the live set is empty or we hit some limit (integrality gap, time, node |
---|
802 | count, etc.). The overall flow is to rebuild a subproblem, reoptimise using |
---|
803 | solveWithCuts(), choose a branching pattern with chooseBranch(), and finally |
---|
804 | add the node to the live set. |
---|
805 | |
---|
806 | The first action is to winnow the live set to remove nodes which are worse |
---|
807 | than the current objective cutoff. |
---|
808 | */ |
---|
809 | while (!tree_->empty()) |
---|
810 | { if (cutoff > getCutoff()) { |
---|
811 | if (eventHandler) { |
---|
812 | if (!eventHandler->event(ClpEventHandler::solution)) { |
---|
813 | eventHappened=true; // exit |
---|
814 | } |
---|
815 | } |
---|
816 | // Do from deepest |
---|
817 | tree_->cleanTree(this, getCutoff(),bestPossibleObjective_) ; |
---|
818 | nodeCompare_->newSolution(this) ; |
---|
819 | nodeCompare_->newSolution(this,continuousObjective_, |
---|
820 | continuousInfeasibilities_) ; |
---|
821 | tree_->setComparison(*nodeCompare_) ; |
---|
822 | if (tree_->empty()) |
---|
823 | break; // finished |
---|
824 | } |
---|
825 | cutoff = getCutoff() ; |
---|
826 | /* |
---|
827 | Periodic activities: Opportunities to |
---|
828 | + tweak the nodeCompare criteria, |
---|
829 | + check if we've closed the integrality gap enough to quit, |
---|
830 | + print a summary line to let the user know we're working |
---|
831 | */ |
---|
832 | if ((numberNodes_%1000) == 0) { |
---|
833 | bool redoTree=nodeCompare_->every1000Nodes(this, numberNodes_) ; |
---|
834 | // redo tree if wanted |
---|
835 | if (redoTree) |
---|
836 | tree_->setComparison(*nodeCompare_) ; |
---|
837 | } |
---|
838 | if ((numberNodes_%printFrequency_) == 0) { |
---|
839 | int j ; |
---|
840 | int nNodes = tree_->size() ; |
---|
841 | bestPossibleObjective_ = 1.0e100 ; |
---|
842 | for (j = 0;j < nNodes;j++) { |
---|
843 | CbcNode * node = tree_->nodePointer(j) ; |
---|
844 | if (node&&node->objectiveValue() < bestPossibleObjective_) |
---|
845 | bestPossibleObjective_ = node->objectiveValue() ; |
---|
846 | } |
---|
847 | messageHandler()->message(CBC_STATUS,messages()) |
---|
848 | << numberNodes_<< nNodes<< bestObjective_<< bestPossibleObjective_ |
---|
849 | << CoinMessageEol ; |
---|
850 | if (eventHandler) { |
---|
851 | if (!eventHandler->event(ClpEventHandler::treeStatus)) { |
---|
852 | eventHappened=true; // exit |
---|
853 | } |
---|
854 | } |
---|
855 | } |
---|
856 | // If no solution but many nodes - signal change in strategy |
---|
857 | if (numberNodes_>2*numberObjects_+1000&&stateOfSearch_!=2) |
---|
858 | stateOfSearch_=3; |
---|
859 | // See if can stop on gap |
---|
860 | double testGap = CoinMax(dblParam_[CbcAllowableGap], |
---|
861 | CoinMax(fabs(bestObjective_),fabs(bestPossibleObjective_)) |
---|
862 | *dblParam_[CbcAllowableFractionGap]); |
---|
863 | if (bestObjective_-bestPossibleObjective_ < testGap) { |
---|
864 | stoppedOnGap = true ; |
---|
865 | } |
---|
866 | |
---|
867 | # ifdef CHECK_NODE_FULL |
---|
868 | verifyTreeNodes(tree_,*this) ; |
---|
869 | # endif |
---|
870 | # ifdef CHECK_CUT_COUNTS |
---|
871 | verifyCutCounts(tree_,*this) ; |
---|
872 | # endif |
---|
873 | |
---|
874 | /* |
---|
875 | Now we come to the meat of the loop. To create the active subproblem, we'll |
---|
876 | pop the most promising node in the live set, rebuild the subproblem it |
---|
877 | represents, and then execute the current arm of the branch to create the |
---|
878 | active subproblem. |
---|
879 | */ |
---|
880 | CbcNode *node = tree_->bestNode(cutoff) ; |
---|
881 | // Possible one on tree worse than cutoff |
---|
882 | if (!node) |
---|
883 | continue; |
---|
884 | currentNode_=node; // so can be accessed elsewhere |
---|
885 | #ifdef CBC_DEBUG |
---|
886 | printf("%d unsat, way %d, obj %g est %g\n", |
---|
887 | node->numberUnsatisfied(),node->way(),node->objectiveValue(), |
---|
888 | node->guessedObjectiveValue()); |
---|
889 | #endif |
---|
890 | // Save clone in branching decision |
---|
891 | if(branchingMethod_) |
---|
892 | branchingMethod_->saveBranchingObject(node->modifiableBranchingObject()); |
---|
893 | bool nodeOnTree=false; // Node has been popped |
---|
894 | // Say not on optimal path |
---|
895 | bool onOptimalPath=false; |
---|
896 | # ifdef CHECK_NODE |
---|
897 | /* |
---|
898 | WARNING: The use of integerVariable_[*] here will break as soon as the |
---|
899 | branching object is something other than an integer variable. |
---|
900 | This needs some thought. |
---|
901 | */ |
---|
902 | printf("Node %x popped from tree - %d left, %d count\n",node, |
---|
903 | node->nodeInfo()->numberBranchesLeft(), |
---|
904 | node->nodeInfo()->numberPointingToThis()) ; |
---|
905 | printf("\tdepth = %d, z = %g, unsat = %d, var = %d.\n", |
---|
906 | node->depth(),node->objectiveValue(), |
---|
907 | node->numberUnsatisfied(), |
---|
908 | integerVariable_[node->variable()]) ; |
---|
909 | # endif |
---|
910 | |
---|
911 | /* |
---|
912 | Rebuild the subproblem for this node: Call addCuts() to adjust the model |
---|
913 | to recreate the subproblem for this node (set proper variable bounds, add |
---|
914 | cuts, create a basis). This may result in the problem being fathomed by |
---|
915 | bound or infeasibility. Returns 1 if node is fathomed. |
---|
916 | Execute the current arm of the branch: If the problem survives, save the |
---|
917 | resulting variable bounds and call branch() to modify variable bounds |
---|
918 | according to the current arm of the branching object. If we're processing |
---|
919 | the final arm of the branching object, flag the node for removal from the |
---|
920 | live set. |
---|
921 | */ |
---|
922 | CbcNodeInfo * nodeInfo = node->nodeInfo() ; |
---|
923 | newNode = NULL ; |
---|
924 | if (!addCuts(node,lastws)) |
---|
925 | { int i ; |
---|
926 | const double * lower = getColLower() ; |
---|
927 | const double * upper = getColUpper() ; |
---|
928 | for (i = 0 ; i < numberColumns ; i++) |
---|
929 | { lowerBefore[i]= lower[i] ; |
---|
930 | upperBefore[i]= upper[i] ; } |
---|
931 | bool deleteNode ; |
---|
932 | if (node->branch()) |
---|
933 | { |
---|
934 | // set nodenumber correctly |
---|
935 | node->nodeInfo()->setNodeNumber(numberNodes2_); |
---|
936 | tree_->push(node) ; |
---|
937 | if (statistics) { |
---|
938 | if (numberNodes2_==maximumStatistics) { |
---|
939 | maximumStatistics = 2*maximumStatistics; |
---|
940 | CbcStatistics ** temp = new CbcStatistics * [maximumStatistics]; |
---|
941 | memset(temp,0,maximumStatistics*sizeof(CbcStatistics *)); |
---|
942 | memcpy(temp,statistics,numberNodes2_*sizeof(CbcStatistics *)); |
---|
943 | delete [] statistics; |
---|
944 | statistics=temp; |
---|
945 | } |
---|
946 | assert (!statistics[numberNodes2_]); |
---|
947 | statistics[numberNodes2_]=new CbcStatistics(node); |
---|
948 | } |
---|
949 | numberNodes2_++; |
---|
950 | nodeOnTree=true; // back on tree |
---|
951 | deleteNode = false ; |
---|
952 | # ifdef CHECK_NODE |
---|
953 | printf("Node %x pushed back on tree - %d left, %d count\n",node, |
---|
954 | nodeInfo->numberBranchesLeft(), |
---|
955 | nodeInfo->numberPointingToThis()) ; |
---|
956 | # endif |
---|
957 | } |
---|
958 | else |
---|
959 | { deleteNode = true ; } |
---|
960 | |
---|
961 | if ((specialOptions_&1)!=0) { |
---|
962 | /* |
---|
963 | This doesn't work as intended --- getRowCutDebugger will return null |
---|
964 | unless the current feasible solution region includes the optimal solution |
---|
965 | that RowCutDebugger knows. There's no way to tell inactive from off the |
---|
966 | optimal path. |
---|
967 | */ |
---|
968 | const OsiRowCutDebugger *debugger = solver_->getRowCutDebugger() ; |
---|
969 | if (debugger) |
---|
970 | { if(debugger->onOptimalPath(*solver_)) { |
---|
971 | onOptimalPath=true; |
---|
972 | printf("On optimal path\n") ; |
---|
973 | } else { |
---|
974 | printf("Not on optimal path\n") ; } |
---|
975 | } |
---|
976 | } |
---|
977 | /* |
---|
978 | Reoptimize, possibly generating cuts and/or using heuristics to find |
---|
979 | solutions. Cut reference counts are unaffected unless we lose feasibility, |
---|
980 | in which case solveWithCuts() will make the adjustment. |
---|
981 | */ |
---|
982 | phase_=2; |
---|
983 | cuts = OsiCuts() ; |
---|
984 | currentNumberCuts = solver_->getNumRows()-numberRowsAtContinuous_ ; |
---|
985 | int saveNumber = numberIterations_; |
---|
986 | feasible = solveWithCuts(cuts,maximumCutPasses_,node, |
---|
987 | numberOldActiveCuts,numberNewCuts, |
---|
988 | maximumWhich,whichGenerator) ; |
---|
989 | if (statistics) { |
---|
990 | assert (numberNodes2_); |
---|
991 | assert (statistics[numberNodes2_-1]); |
---|
992 | assert (statistics[numberNodes2_-1]->node()==numberNodes2_-1); |
---|
993 | statistics[numberNodes2_-1]->endOfBranch(numberIterations_-saveNumber, |
---|
994 | feasible ? solver_->getObjValue() |
---|
995 | : COIN_DBL_MAX); |
---|
996 | } |
---|
997 | /* |
---|
998 | Check for abort on limits: node count, solution count, time, integrality gap. |
---|
999 | */ |
---|
1000 | double totalTime = CoinCpuTime()-dblParam_[CbcStartSeconds] ; |
---|
1001 | if (numberNodes_ < intParam_[CbcMaxNumNode] && |
---|
1002 | numberSolutions_ < intParam_[CbcMaxNumSol] && |
---|
1003 | totalTime < dblParam_[CbcMaximumSeconds] && |
---|
1004 | !stoppedOnGap&&!eventHappened) |
---|
1005 | { |
---|
1006 | /* |
---|
1007 | Are we still feasible? If so, create a node and do the work to attach a |
---|
1008 | branching object, reoptimising as needed if chooseBranch() identifies |
---|
1009 | monotone objects. |
---|
1010 | |
---|
1011 | Finally, attach a partial nodeInfo object and store away any cuts that we |
---|
1012 | created back in solveWithCuts. addCuts() will also deal with the cut |
---|
1013 | reference counts. |
---|
1014 | |
---|
1015 | TODO: (lh) I'm confused. We create a nodeInfo without checking whether we |
---|
1016 | have a solution or not. Then we use numberUnsatisfied() to decide |
---|
1017 | whether to stash the cuts and bump reference counts. Other places we |
---|
1018 | use variable() (i.e., presence of a branching variable). Equivalent? |
---|
1019 | */ |
---|
1020 | if (onOptimalPath) |
---|
1021 | assert (feasible); |
---|
1022 | if (feasible) |
---|
1023 | { newNode = new CbcNode ; |
---|
1024 | newNode->setObjectiveValue(direction*solver_->getObjValue()) ; |
---|
1025 | if (newNode->objectiveValue() >= getCutoff()) |
---|
1026 | anyAction=-2; |
---|
1027 | anyAction =-1 ; |
---|
1028 | resolved = false ; |
---|
1029 | if (newNode->objectiveValue() >= getCutoff()) |
---|
1030 | anyAction=-2; |
---|
1031 | // only allow twenty passes |
---|
1032 | int numberPassesLeft=20; |
---|
1033 | while (anyAction == -1) |
---|
1034 | { |
---|
1035 | if (numberBeforeTrust_<=0 ) { |
---|
1036 | anyAction = newNode->chooseBranch(this,node,numberPassesLeft) ; |
---|
1037 | } else { |
---|
1038 | anyAction = newNode->chooseDynamicBranch(this,node,numberPassesLeft) ; |
---|
1039 | if (anyAction==-3) |
---|
1040 | anyAction = newNode->chooseBranch(this,node,numberPassesLeft) ; // dynamic did nothing |
---|
1041 | } |
---|
1042 | if (onOptimalPath) |
---|
1043 | assert (anyAction!=-2); // can be useful but gives false positives on strong |
---|
1044 | numberPassesLeft--; |
---|
1045 | if (numberPassesLeft<=-1) { |
---|
1046 | if (!numberLongStrong) |
---|
1047 | messageHandler()->message(CBC_WARNING_STRONG, |
---|
1048 | messages()) << CoinMessageEol ; |
---|
1049 | numberLongStrong++; |
---|
1050 | } |
---|
1051 | if (anyAction == -1) |
---|
1052 | { |
---|
1053 | // can do quick optimality check |
---|
1054 | int easy=2; |
---|
1055 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,&easy) ; |
---|
1056 | feasible = resolve() ; |
---|
1057 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,NULL) ; |
---|
1058 | resolved = true ; |
---|
1059 | if (problemFeasibility_->feasible(this,0)<0) { |
---|
1060 | feasible=false; // pretend infeasible |
---|
1061 | } |
---|
1062 | if (feasible) |
---|
1063 | { newNode->setObjectiveValue(direction* |
---|
1064 | solver_->getObjValue()) ; |
---|
1065 | if (newNode->objectiveValue() >= getCutoff()) |
---|
1066 | anyAction=-2; |
---|
1067 | } |
---|
1068 | else |
---|
1069 | { anyAction = -2 ; } } } |
---|
1070 | if (anyAction >= 0) |
---|
1071 | { if (resolved) |
---|
1072 | { bool needValidSolution = (newNode->variable() < 0) ; |
---|
1073 | takeOffCuts(cuts,whichGenerator,numberOldActiveCuts, |
---|
1074 | numberNewCuts,needValidSolution) ; |
---|
1075 | # ifdef CHECK_CUT_COUNTS |
---|
1076 | { printf("Number of rows after chooseBranch fix (node)" |
---|
1077 | "(active only) %d\n", |
---|
1078 | numberRowsAtContinuous_+numberNewCuts+ |
---|
1079 | numberOldActiveCuts) ; |
---|
1080 | const CoinWarmStartBasis* debugws = |
---|
1081 | dynamic_cast<const CoinWarmStartBasis*> |
---|
1082 | (solver_->getWarmStart()) ; |
---|
1083 | debugws->print() ; |
---|
1084 | delete debugws ; } |
---|
1085 | # endif |
---|
1086 | } |
---|
1087 | newNode->createInfo(this,node,lastws,lowerBefore,upperBefore, |
---|
1088 | numberOldActiveCuts,numberNewCuts) ; |
---|
1089 | if (newNode->numberUnsatisfied()) |
---|
1090 | newNode->nodeInfo()->addCuts(cuts,newNode->numberBranches(), |
---|
1091 | whichGenerator) ; } } |
---|
1092 | else |
---|
1093 | { anyAction = -2 ; } |
---|
1094 | // May have slipped through i.e. anyAction == 0 and objective above cutoff |
---|
1095 | if ( anyAction >=0 ) { |
---|
1096 | assert (newNode); |
---|
1097 | if (newNode->objectiveValue() >= getCutoff()) |
---|
1098 | anyAction = -2; // say bad after all |
---|
1099 | } |
---|
1100 | /* |
---|
1101 | If we end up infeasible, we can delete the new node immediately. Since this |
---|
1102 | node won't be needing the cuts we collected, decrement the reference counts. |
---|
1103 | If we are feasible, then we'll be placing this node into the live set, so |
---|
1104 | increment the reference count in the current (parent) nodeInfo. |
---|
1105 | */ |
---|
1106 | if (anyAction == -2) |
---|
1107 | { delete newNode ; |
---|
1108 | newNode = NULL ; |
---|
1109 | // switch off any hot start |
---|
1110 | hotstartStrategy_=0; |
---|
1111 | for (i = 0 ; i < currentNumberCuts_ ; i++) |
---|
1112 | { if (addedCuts_[i]) |
---|
1113 | { if (!addedCuts_[i]->decrement(1)) |
---|
1114 | delete addedCuts_[i] ; } } } |
---|
1115 | else |
---|
1116 | { nodeInfo->increment() ; } |
---|
1117 | /* |
---|
1118 | At this point, there are three possibilities: |
---|
1119 | * We have a live node (variable() >= 0) which will require further |
---|
1120 | branching to resolve. Before we push it onto the search tree, try for |
---|
1121 | a heuristic solution. |
---|
1122 | * We have a solution, in which case newNode is non-null but we have no |
---|
1123 | branching variable. Decrement the cut counts and save the solution. |
---|
1124 | * The node was found to be infeasible, in which case it's already been |
---|
1125 | deleted, and newNode is null. |
---|
1126 | |
---|
1127 | TODO: (lh) Now I'm more confused. I thought that the call to addCuts() above |
---|
1128 | took care of incrementing the reference counts for cuts at newNode. |
---|
1129 | Clearly I need to look more carefully. |
---|
1130 | */ |
---|
1131 | if (eventHandler) { |
---|
1132 | if (!eventHandler->event(ClpEventHandler::node)) { |
---|
1133 | eventHappened=true; // exit |
---|
1134 | } |
---|
1135 | } |
---|
1136 | assert (!newNode || newNode->objectiveValue() <= getCutoff()) ; |
---|
1137 | if (statistics) { |
---|
1138 | assert (numberNodes2_); |
---|
1139 | assert (statistics[numberNodes2_-1]); |
---|
1140 | assert (statistics[numberNodes2_-1]->node()==numberNodes2_-1); |
---|
1141 | if (newNode) |
---|
1142 | statistics[numberNodes2_-1]->updateInfeasibility(newNode->numberUnsatisfied()); |
---|
1143 | else |
---|
1144 | statistics[numberNodes2_-1]->sayInfeasible(); |
---|
1145 | } |
---|
1146 | if (newNode) |
---|
1147 | { if (newNode->variable() >= 0) |
---|
1148 | { handler_->message(CBC_BRANCH,messages_) |
---|
1149 | << numberNodes_<< newNode->objectiveValue() |
---|
1150 | << newNode->numberUnsatisfied()<< newNode->depth() |
---|
1151 | << CoinMessageEol ; |
---|
1152 | // Increment cut counts (taking off current) |
---|
1153 | int numberLeft = newNode->numberBranches() ; |
---|
1154 | for (i = 0;i < currentNumberCuts_;i++) |
---|
1155 | { if (addedCuts_[i]) |
---|
1156 | { |
---|
1157 | # ifdef CHECK_CUT_COUNTS |
---|
1158 | printf("Count on cut %x increased by %d\n",addedCuts_[i], |
---|
1159 | numberLeft-1) ; |
---|
1160 | # endif |
---|
1161 | addedCuts_[i]->increment(numberLeft-1) ; } } |
---|
1162 | |
---|
1163 | double estValue = newNode->guessedObjectiveValue() ; |
---|
1164 | int found = -1 ; |
---|
1165 | // no - overhead on small problems solver_->resolve() ; // double check current optimal |
---|
1166 | // assert (!solver_->getIterationCount()); |
---|
1167 | double * newSolution = new double [numberColumns] ; |
---|
1168 | double heurValue = getCutoff() ; |
---|
1169 | int iHeur ; |
---|
1170 | for (iHeur = 0 ; iHeur < numberHeuristics_ ; iHeur++) |
---|
1171 | { double saveValue = heurValue ; |
---|
1172 | int ifSol = heuristic_[iHeur]->solution(heurValue,newSolution) ; |
---|
1173 | if (ifSol > 0) { |
---|
1174 | // new solution found |
---|
1175 | found = iHeur ; |
---|
1176 | incrementUsed(newSolution); |
---|
1177 | } |
---|
1178 | else |
---|
1179 | if (ifSol < 0) // just returning an estimate |
---|
1180 | { estValue = CoinMin(heurValue,estValue) ; |
---|
1181 | heurValue = saveValue ; } } |
---|
1182 | if (found >= 0) { |
---|
1183 | setBestSolution(CBC_ROUNDING,heurValue,newSolution) ; |
---|
1184 | lastHeuristic_ = heuristic_[found]; |
---|
1185 | } |
---|
1186 | delete [] newSolution ; |
---|
1187 | newNode->setGuessedObjectiveValue(estValue) ; |
---|
1188 | tree_->push(newNode) ; |
---|
1189 | if (statistics) { |
---|
1190 | if (numberNodes2_==maximumStatistics) { |
---|
1191 | maximumStatistics = 2*maximumStatistics; |
---|
1192 | CbcStatistics ** temp = new CbcStatistics * [maximumStatistics]; |
---|
1193 | memset(temp,0,maximumStatistics*sizeof(CbcStatistics *)); |
---|
1194 | memcpy(temp,statistics,numberNodes2_*sizeof(CbcStatistics *)); |
---|
1195 | delete [] statistics; |
---|
1196 | statistics=temp; |
---|
1197 | } |
---|
1198 | assert (!statistics[numberNodes2_]); |
---|
1199 | statistics[numberNodes2_]=new CbcStatistics(newNode); |
---|
1200 | } |
---|
1201 | numberNodes2_++; |
---|
1202 | # ifdef CHECK_NODE |
---|
1203 | printf("Node %x pushed on tree c\n",newNode) ; |
---|
1204 | # endif |
---|
1205 | } |
---|
1206 | else |
---|
1207 | { for (i = 0 ; i < currentNumberCuts_ ; i++) |
---|
1208 | { if (addedCuts_[i]) |
---|
1209 | { if (!addedCuts_[i]->decrement(1)) |
---|
1210 | delete addedCuts_[i] ; } } |
---|
1211 | double objectiveValue = newNode->objectiveValue(); |
---|
1212 | setBestSolution(CBC_SOLUTION,objectiveValue, |
---|
1213 | solver_->getColSolution()) ; |
---|
1214 | lastHeuristic_ = NULL; |
---|
1215 | incrementUsed(solver_->getColSolution()); |
---|
1216 | assert(nodeInfo->numberPointingToThis() <= 2) ; |
---|
1217 | // avoid accidental pruning, if newNode was final branch arm |
---|
1218 | nodeInfo->increment(); |
---|
1219 | delete newNode ; |
---|
1220 | nodeInfo->decrement() ; } } |
---|
1221 | /* |
---|
1222 | This node has been completely expanded and can be removed from the live |
---|
1223 | set. |
---|
1224 | */ |
---|
1225 | if (deleteNode) |
---|
1226 | delete node ; } |
---|
1227 | /* |
---|
1228 | End of the non-abort actions. The next block of code is executed if we've |
---|
1229 | aborted because we hit one of the limits. Clean up by deleting the live set |
---|
1230 | and break out of the node processing loop. |
---|
1231 | */ |
---|
1232 | else |
---|
1233 | { |
---|
1234 | tree_->cleanTree(this,-COIN_DBL_MAX,bestPossibleObjective_) ; |
---|
1235 | delete nextRowCut_; |
---|
1236 | // We need to get rid of node if is has already been popped from tree |
---|
1237 | if (!nodeOnTree&&!stoppedOnGap&&node!=rootNode) |
---|
1238 | delete node; |
---|
1239 | if (stoppedOnGap) |
---|
1240 | { messageHandler()->message(CBC_GAP,messages()) |
---|
1241 | << bestObjective_-bestPossibleObjective_ |
---|
1242 | << dblParam_[CbcAllowableGap] |
---|
1243 | << dblParam_[CbcAllowableFractionGap]*100.0 |
---|
1244 | << CoinMessageEol ; |
---|
1245 | secondaryStatus_ = 2; |
---|
1246 | status_ = 0 ; } |
---|
1247 | else |
---|
1248 | if (isNodeLimitReached()) |
---|
1249 | { handler_->message(CBC_MAXNODES,messages_) << CoinMessageEol ; |
---|
1250 | secondaryStatus_ = 3; |
---|
1251 | status_ = 1 ; } |
---|
1252 | else |
---|
1253 | if (totalTime >= dblParam_[CbcMaximumSeconds]) |
---|
1254 | { handler_->message(CBC_MAXTIME,messages_) << CoinMessageEol ; |
---|
1255 | secondaryStatus_ = 4; |
---|
1256 | status_ = 1 ; } |
---|
1257 | else |
---|
1258 | if (eventHappened) |
---|
1259 | { handler_->message(CBC_EVENT,messages_) << CoinMessageEol ; |
---|
1260 | secondaryStatus_ = 5; |
---|
1261 | status_ = 5 ; } |
---|
1262 | else |
---|
1263 | { handler_->message(CBC_MAXSOLS,messages_) << CoinMessageEol ; |
---|
1264 | secondaryStatus_ = 6; |
---|
1265 | status_ = 1 ; } |
---|
1266 | break ; } |
---|
1267 | /* |
---|
1268 | Delete cuts to get back to the original system. |
---|
1269 | |
---|
1270 | I'm thinking this is redundant --- the call to addCuts that conditions entry |
---|
1271 | to this code block also performs this action. |
---|
1272 | */ |
---|
1273 | int numberToDelete = getNumRows()-numberRowsAtContinuous_ ; |
---|
1274 | if (numberToDelete) |
---|
1275 | { int * delRows = new int[numberToDelete] ; |
---|
1276 | int i ; |
---|
1277 | for (i = 0 ; i < numberToDelete ; i++) |
---|
1278 | { delRows[i] = i+numberRowsAtContinuous_ ; } |
---|
1279 | solver_->deleteRows(numberToDelete,delRows) ; |
---|
1280 | delete [] delRows ; } } |
---|
1281 | /* |
---|
1282 | This node fathomed when addCuts atttempted to revive it. Toss it. |
---|
1283 | */ |
---|
1284 | else |
---|
1285 | { delete node ; } } |
---|
1286 | /* |
---|
1287 | That's it, we've exhausted the search tree, or broken out of the loop because |
---|
1288 | we hit some limit on evaluation. |
---|
1289 | |
---|
1290 | We may have got an intelligent tree so give it one more chance |
---|
1291 | */ |
---|
1292 | // Tell solver we are not in Branch and Cut |
---|
1293 | solver_->setHintParam(OsiDoInBranchAndCut,false,OsiHintDo,NULL) ; |
---|
1294 | tree_->endSearch(); |
---|
1295 | // If we did any sub trees - did we give up on any? |
---|
1296 | if ( numberStoppedSubTrees_) |
---|
1297 | status_=1; |
---|
1298 | if (!status_) { |
---|
1299 | bestPossibleObjective_=bestObjective_; |
---|
1300 | handler_->message(CBC_END_GOOD,messages_) |
---|
1301 | << bestObjective_ << numberIterations_ << numberNodes_ |
---|
1302 | << CoinMessageEol ; |
---|
1303 | } else { |
---|
1304 | handler_->message(CBC_END,messages_) |
---|
1305 | << bestObjective_ <<bestPossibleObjective_ |
---|
1306 | << numberIterations_ << numberNodes_ |
---|
1307 | << CoinMessageEol ; |
---|
1308 | } |
---|
1309 | if (statistics) { |
---|
1310 | // report in some way |
---|
1311 | int * lookup = new int[numberObjects_]; |
---|
1312 | int i; |
---|
1313 | for (i=0;i<numberObjects_;i++) |
---|
1314 | lookup[i]=-1; |
---|
1315 | bool goodIds=true; |
---|
1316 | for (i=0;i<numberObjects_;i++) { |
---|
1317 | int id = object_[i]->id(); |
---|
1318 | int iColumn = object_[i]->columnNumber(); |
---|
1319 | if (iColumn<0) |
---|
1320 | iColumn = id+numberColumns; |
---|
1321 | if(id>=0&&id<numberObjects_) { |
---|
1322 | if (lookup[id]==-1) { |
---|
1323 | lookup[id]=iColumn; |
---|
1324 | } else { |
---|
1325 | goodIds=false; |
---|
1326 | break; |
---|
1327 | } |
---|
1328 | } else { |
---|
1329 | goodIds=false; |
---|
1330 | break; |
---|
1331 | } |
---|
1332 | } |
---|
1333 | if (!goodIds) { |
---|
1334 | delete [] lookup; |
---|
1335 | lookup=NULL; |
---|
1336 | } |
---|
1337 | if (doStatistics==3) { |
---|
1338 | printf(" node parent depth column value obj inf\n"); |
---|
1339 | for ( i=0;i<numberNodes2_;i++) { |
---|
1340 | statistics[i]->print(lookup); |
---|
1341 | } |
---|
1342 | } |
---|
1343 | if (doStatistics>1) { |
---|
1344 | // Find last solution |
---|
1345 | int k; |
---|
1346 | for (k=numberNodes2_-1;k>=0;k--) { |
---|
1347 | if (statistics[k]->endingObjective()!=COIN_DBL_MAX&& |
---|
1348 | !statistics[k]->endingInfeasibility()) |
---|
1349 | break; |
---|
1350 | } |
---|
1351 | if (k>=0) { |
---|
1352 | int depth=statistics[k]->depth(); |
---|
1353 | int * which = new int[depth+1]; |
---|
1354 | for (i=depth;i>=0;i--) { |
---|
1355 | which[i]=k; |
---|
1356 | k=statistics[k]->parentNode(); |
---|
1357 | } |
---|
1358 | printf(" node parent depth column value obj inf\n"); |
---|
1359 | for (i=0;i<=depth;i++) { |
---|
1360 | statistics[which[i]]->print(lookup); |
---|
1361 | } |
---|
1362 | delete [] which; |
---|
1363 | } |
---|
1364 | } |
---|
1365 | // now summary |
---|
1366 | int maxDepth=0; |
---|
1367 | double averageSolutionDepth=0.0; |
---|
1368 | int numberSolutions=0; |
---|
1369 | double averageCutoffDepth=0.0; |
---|
1370 | double averageSolvedDepth=0.0; |
---|
1371 | int numberCutoff=0; |
---|
1372 | int numberDown=0; |
---|
1373 | int numberFirstDown=0; |
---|
1374 | double averageInfDown=0.0; |
---|
1375 | double averageObjDown=0.0; |
---|
1376 | int numberCutoffDown=0; |
---|
1377 | int numberUp=0; |
---|
1378 | int numberFirstUp=0; |
---|
1379 | double averageInfUp=0.0; |
---|
1380 | double averageObjUp=0.0; |
---|
1381 | int numberCutoffUp=0; |
---|
1382 | double averageNumberIterations1=0.0; |
---|
1383 | double averageValue=0.0; |
---|
1384 | for ( i=0;i<numberNodes2_;i++) { |
---|
1385 | int depth = statistics[i]->depth(); |
---|
1386 | int way = statistics[i]->way(); |
---|
1387 | double value = statistics[i]->value(); |
---|
1388 | double startingObjective = statistics[i]->startingObjective(); |
---|
1389 | int startingInfeasibility = statistics[i]->startingInfeasibility(); |
---|
1390 | double endingObjective = statistics[i]->endingObjective(); |
---|
1391 | int endingInfeasibility = statistics[i]->endingInfeasibility(); |
---|
1392 | maxDepth = CoinMax(depth,maxDepth); |
---|
1393 | // Only for completed |
---|
1394 | averageNumberIterations1 += statistics[i]->numberIterations(); |
---|
1395 | averageValue += value; |
---|
1396 | if (endingObjective!=COIN_DBL_MAX&&!endingInfeasibility) { |
---|
1397 | numberSolutions++; |
---|
1398 | averageSolutionDepth += depth; |
---|
1399 | } |
---|
1400 | if (endingObjective==COIN_DBL_MAX) { |
---|
1401 | numberCutoff++; |
---|
1402 | averageCutoffDepth += depth; |
---|
1403 | if (way<0) { |
---|
1404 | numberDown++; |
---|
1405 | numberCutoffDown++; |
---|
1406 | if (way==-1) |
---|
1407 | numberFirstDown++; |
---|
1408 | } else { |
---|
1409 | numberUp++; |
---|
1410 | numberCutoffUp++; |
---|
1411 | if (way==1) |
---|
1412 | numberFirstUp++; |
---|
1413 | } |
---|
1414 | } else { |
---|
1415 | averageSolvedDepth += depth; |
---|
1416 | if (way<0) { |
---|
1417 | numberDown++; |
---|
1418 | averageInfDown += startingInfeasibility-endingInfeasibility; |
---|
1419 | averageObjDown += endingObjective-startingObjective; |
---|
1420 | if (way==-1) |
---|
1421 | numberFirstDown++; |
---|
1422 | } else { |
---|
1423 | numberUp++; |
---|
1424 | averageInfUp += startingInfeasibility-endingInfeasibility; |
---|
1425 | averageObjUp += endingObjective-startingObjective; |
---|
1426 | if (way==1) |
---|
1427 | numberFirstUp++; |
---|
1428 | } |
---|
1429 | } |
---|
1430 | } |
---|
1431 | // Now print |
---|
1432 | if (numberSolutions) |
---|
1433 | averageSolutionDepth /= (double) numberSolutions; |
---|
1434 | int numberSolved = numberNodes2_-numberCutoff; |
---|
1435 | double averageNumberIterations2=numberIterations_-averageNumberIterations1; |
---|
1436 | if(numberCutoff) { |
---|
1437 | averageCutoffDepth /= (double) numberCutoff; |
---|
1438 | averageNumberIterations2 /= (double) numberCutoff; |
---|
1439 | } |
---|
1440 | if (numberNodes2_) |
---|
1441 | averageValue /= (double) numberNodes2_; |
---|
1442 | if (numberSolved) { |
---|
1443 | averageNumberIterations1 /= (double) numberSolved; |
---|
1444 | averageSolvedDepth /= (double) numberSolved; |
---|
1445 | } |
---|
1446 | printf("%d solution(s) were found (by branching) at an average depth of %g\n", |
---|
1447 | numberSolutions,averageSolutionDepth); |
---|
1448 | printf("average value of variable being branched on was %g\n", |
---|
1449 | averageValue); |
---|
1450 | printf("%d nodes were cutoff at an average depth of %g with iteration count of %g\n", |
---|
1451 | numberCutoff,averageCutoffDepth,averageNumberIterations2); |
---|
1452 | printf("%d nodes were solved at an average depth of %g with iteration count of %g\n", |
---|
1453 | numberSolved,averageSolvedDepth,averageNumberIterations1); |
---|
1454 | if (numberDown) { |
---|
1455 | averageInfDown /= (double) numberDown; |
---|
1456 | averageObjDown /= (double) numberDown; |
---|
1457 | } |
---|
1458 | printf("Down %d nodes (%d first, %d second) - %d cutoff, rest decrease numinf %g increase obj %g\n", |
---|
1459 | numberDown,numberFirstDown,numberDown-numberFirstDown,numberCutoffDown, |
---|
1460 | averageInfDown,averageObjDown); |
---|
1461 | if (numberUp) { |
---|
1462 | averageInfUp /= (double) numberUp; |
---|
1463 | averageObjUp /= (double) numberUp; |
---|
1464 | } |
---|
1465 | printf("Up %d nodes (%d first, %d second) - %d cutoff, rest decrease numinf %g increase obj %g\n", |
---|
1466 | numberUp,numberFirstUp,numberUp-numberFirstUp,numberCutoffUp, |
---|
1467 | averageInfUp,averageObjUp); |
---|
1468 | for ( i=0;i<numberNodes2_;i++) |
---|
1469 | delete statistics[i]; |
---|
1470 | delete [] statistics; |
---|
1471 | delete [] lookup; |
---|
1472 | } |
---|
1473 | /* |
---|
1474 | If we think we have a solution, restore and confirm it with a call to |
---|
1475 | setBestSolution(). We need to reset the cutoff value so as not to fathom |
---|
1476 | the solution on bounds. Note that calling setBestSolution( ..., true) |
---|
1477 | leaves the continuousSolver_ bounds vectors fixed at the solution value. |
---|
1478 | |
---|
1479 | Running resolve() here is a failsafe --- setBestSolution has already |
---|
1480 | reoptimised using the continuousSolver_. If for some reason we fail to |
---|
1481 | prove optimality, run the problem again after instructing the solver to |
---|
1482 | tell us more. |
---|
1483 | |
---|
1484 | If all looks good, replace solver_ with continuousSolver_, so that the |
---|
1485 | outside world will be able to obtain information about the solution using |
---|
1486 | public methods. |
---|
1487 | */ |
---|
1488 | if (bestSolution_) |
---|
1489 | { setCutoff(1.0e50) ; // As best solution should be worse than cutoff |
---|
1490 | phase_=5; |
---|
1491 | setBestSolution(CBC_SOLUTION,bestObjective_,bestSolution_,true) ; |
---|
1492 | continuousSolver_->resolve() ; |
---|
1493 | if (!continuousSolver_->isProvenOptimal()) |
---|
1494 | { continuousSolver_->messageHandler()->setLogLevel(2) ; |
---|
1495 | continuousSolver_->initialSolve() ; } |
---|
1496 | delete solver_ ; |
---|
1497 | solver_ = continuousSolver_ ; |
---|
1498 | continuousSolver_ = NULL ; } |
---|
1499 | /* |
---|
1500 | Clean up dangling objects. continuousSolver_ may already be toast. |
---|
1501 | */ |
---|
1502 | delete lastws ; |
---|
1503 | delete [] whichGenerator ; |
---|
1504 | delete [] lowerBefore ; |
---|
1505 | delete [] upperBefore ; |
---|
1506 | delete [] walkback_ ; |
---|
1507 | walkback_ = NULL ; |
---|
1508 | delete [] addedCuts_ ; |
---|
1509 | addedCuts_ = NULL ; |
---|
1510 | if (continuousSolver_) |
---|
1511 | { delete continuousSolver_ ; |
---|
1512 | continuousSolver_ = NULL ; } |
---|
1513 | /* |
---|
1514 | Destroy global cuts by replacing with an empty OsiCuts object. |
---|
1515 | */ |
---|
1516 | globalCuts_= OsiCuts() ; |
---|
1517 | return ; } |
---|
1518 | |
---|
1519 | |
---|
1520 | |
---|
1521 | // Solve the initial LP relaxation |
---|
1522 | void |
---|
1523 | CbcModel::initialSolve() |
---|
1524 | { |
---|
1525 | assert (solver_); |
---|
1526 | solver_->initialSolve(); |
---|
1527 | // But set up so Jon Lee will be happy |
---|
1528 | status_=-1; |
---|
1529 | secondaryStatus_ = -1; |
---|
1530 | originalContinuousObjective_ = solver_->getObjValue()*solver_->getObjSense(); |
---|
1531 | delete [] continuousSolution_; |
---|
1532 | continuousSolution_ = CoinCopyOfArray(solver_->getColSolution(), |
---|
1533 | solver_->getNumCols()); |
---|
1534 | } |
---|
1535 | |
---|
1536 | /*! \brief Get an empty basis object |
---|
1537 | |
---|
1538 | Return an empty CoinWarmStartBasis object with the requested capacity, |
---|
1539 | appropriate for the current solver. The object is cloned from the object |
---|
1540 | cached as emptyWarmStart_. If there is no cached object, the routine |
---|
1541 | queries the solver for a warm start object, empties it, and caches the |
---|
1542 | result. |
---|
1543 | */ |
---|
1544 | |
---|
1545 | CoinWarmStartBasis *CbcModel::getEmptyBasis (int ns, int na) const |
---|
1546 | |
---|
1547 | { CoinWarmStartBasis *emptyBasis ; |
---|
1548 | /* |
---|
1549 | Acquire an empty basis object, if we don't yet have one. |
---|
1550 | */ |
---|
1551 | if (emptyWarmStart_ == 0) |
---|
1552 | { if (solver_ == 0) |
---|
1553 | { throw CoinError("Cannot construct basis without solver!", |
---|
1554 | "getEmptyBasis","CbcModel") ; } |
---|
1555 | emptyBasis = |
---|
1556 | dynamic_cast<CoinWarmStartBasis *>(solver_->getEmptyWarmStart()) ; |
---|
1557 | if (emptyBasis == 0) |
---|
1558 | { throw CoinError( |
---|
1559 | "Solver does not appear to use a basis-oriented warm start.", |
---|
1560 | "getEmptyBasis","CbcModel") ; } |
---|
1561 | emptyBasis->setSize(0,0) ; |
---|
1562 | emptyWarmStart_ = dynamic_cast<CoinWarmStart *>(emptyBasis) ; } |
---|
1563 | /* |
---|
1564 | Clone the empty basis object, resize it as requested, and return. |
---|
1565 | */ |
---|
1566 | emptyBasis = dynamic_cast<CoinWarmStartBasis *>(emptyWarmStart_->clone()) ; |
---|
1567 | assert(emptyBasis) ; |
---|
1568 | if (ns != 0 || na != 0) emptyBasis->setSize(ns,na) ; |
---|
1569 | |
---|
1570 | return (emptyBasis) ; } |
---|
1571 | |
---|
1572 | |
---|
1573 | /** Default Constructor |
---|
1574 | |
---|
1575 | Creates an empty model without an associated solver. |
---|
1576 | */ |
---|
1577 | CbcModel::CbcModel() |
---|
1578 | |
---|
1579 | : |
---|
1580 | solver_(NULL), |
---|
1581 | ourSolver_(true), |
---|
1582 | continuousSolver_(NULL), |
---|
1583 | defaultHandler_(true), |
---|
1584 | emptyWarmStart_(NULL), |
---|
1585 | basis_(NULL), |
---|
1586 | bestObjective_(COIN_DBL_MAX), |
---|
1587 | bestPossibleObjective_(COIN_DBL_MAX), |
---|
1588 | sumChangeObjective1_(0.0), |
---|
1589 | sumChangeObjective2_(0.0), |
---|
1590 | bestSolution_(NULL), |
---|
1591 | currentSolution_(NULL), |
---|
1592 | testSolution_(NULL), |
---|
1593 | minimumDrop_(1.0e-4), |
---|
1594 | numberSolutions_(0), |
---|
1595 | stateOfSearch_(0), |
---|
1596 | hotstartStrategy_(0), |
---|
1597 | numberHeuristicSolutions_(0), |
---|
1598 | numberNodes_(0), |
---|
1599 | numberNodes2_(0), |
---|
1600 | numberIterations_(0), |
---|
1601 | status_(-1), |
---|
1602 | secondaryStatus_(-1), |
---|
1603 | numberIntegers_(0), |
---|
1604 | numberRowsAtContinuous_(0), |
---|
1605 | maximumNumberCuts_(0), |
---|
1606 | phase_(0), |
---|
1607 | currentNumberCuts_(0), |
---|
1608 | maximumDepth_(0), |
---|
1609 | walkback_(NULL), |
---|
1610 | addedCuts_(NULL), |
---|
1611 | nextRowCut_(NULL), |
---|
1612 | currentNode_(NULL), |
---|
1613 | integerVariable_(NULL), |
---|
1614 | continuousSolution_(NULL), |
---|
1615 | usedInSolution_(NULL), |
---|
1616 | specialOptions_(0), |
---|
1617 | subTreeModel_(NULL), |
---|
1618 | numberStoppedSubTrees_(0), |
---|
1619 | presolve_(0), |
---|
1620 | numberStrong_(5), |
---|
1621 | numberBeforeTrust_(0), |
---|
1622 | numberPenalties_(20), |
---|
1623 | penaltyScaleFactor_(3.0), |
---|
1624 | numberInfeasibleNodes_(0), |
---|
1625 | problemType_(0), |
---|
1626 | printFrequency_(0), |
---|
1627 | numberCutGenerators_(0), |
---|
1628 | generator_(NULL), |
---|
1629 | virginGenerator_(NULL), |
---|
1630 | numberHeuristics_(0), |
---|
1631 | heuristic_(NULL), |
---|
1632 | lastHeuristic_(NULL), |
---|
1633 | numberObjects_(0), |
---|
1634 | object_(NULL), |
---|
1635 | originalColumns_(NULL), |
---|
1636 | howOftenGlobalScan_(1), |
---|
1637 | numberGlobalViolations_(0), |
---|
1638 | continuousObjective_(COIN_DBL_MAX), |
---|
1639 | originalContinuousObjective_(COIN_DBL_MAX), |
---|
1640 | continuousInfeasibilities_(INT_MAX), |
---|
1641 | maximumCutPassesAtRoot_(20), |
---|
1642 | maximumCutPasses_(10), |
---|
1643 | resolveAfterTakeOffCuts_(true) |
---|
1644 | { |
---|
1645 | intParam_[CbcMaxNumNode] = 2147483647; |
---|
1646 | intParam_[CbcMaxNumSol] = 9999999; |
---|
1647 | intParam_[CbcFathomDiscipline] = 0; |
---|
1648 | |
---|
1649 | dblParam_[CbcIntegerTolerance] = 1e-6; |
---|
1650 | dblParam_[CbcInfeasibilityWeight] = 0.0; |
---|
1651 | dblParam_[CbcCutoffIncrement] = 1e-5; |
---|
1652 | dblParam_[CbcAllowableGap] = 1.0e-10; |
---|
1653 | dblParam_[CbcAllowableFractionGap] = 0.0; |
---|
1654 | dblParam_[CbcMaximumSeconds] = 1.0e100; |
---|
1655 | dblParam_[CbcStartSeconds] = 0.0; |
---|
1656 | nodeCompare_=new CbcCompareDefault();; |
---|
1657 | problemFeasibility_=new CbcFeasibilityBase(); |
---|
1658 | tree_= new CbcTree(); |
---|
1659 | branchingMethod_=NULL; |
---|
1660 | strategy_=NULL; |
---|
1661 | parentModel_=NULL; |
---|
1662 | appData_=NULL; |
---|
1663 | handler_ = new CoinMessageHandler(); |
---|
1664 | handler_->setLogLevel(2); |
---|
1665 | messages_ = CbcMessage(); |
---|
1666 | } |
---|
1667 | |
---|
1668 | /** Constructor from solver. |
---|
1669 | |
---|
1670 | Creates a model complete with a clone of the solver passed as a parameter. |
---|
1671 | */ |
---|
1672 | |
---|
1673 | CbcModel::CbcModel(const OsiSolverInterface &rhs) |
---|
1674 | : |
---|
1675 | continuousSolver_(NULL), |
---|
1676 | defaultHandler_(true), |
---|
1677 | emptyWarmStart_(NULL), |
---|
1678 | basis_(NULL) , |
---|
1679 | bestObjective_(COIN_DBL_MAX), |
---|
1680 | bestPossibleObjective_(COIN_DBL_MAX), |
---|
1681 | sumChangeObjective1_(0.0), |
---|
1682 | sumChangeObjective2_(0.0), |
---|
1683 | minimumDrop_(1.0e-4), |
---|
1684 | numberSolutions_(0), |
---|
1685 | stateOfSearch_(0), |
---|
1686 | hotstartStrategy_(0), |
---|
1687 | numberHeuristicSolutions_(0), |
---|
1688 | numberNodes_(0), |
---|
1689 | numberNodes2_(0), |
---|
1690 | numberIterations_(0), |
---|
1691 | status_(-1), |
---|
1692 | secondaryStatus_(-1), |
---|
1693 | numberRowsAtContinuous_(0), |
---|
1694 | maximumNumberCuts_(0), |
---|
1695 | phase_(0), |
---|
1696 | currentNumberCuts_(0), |
---|
1697 | maximumDepth_(0), |
---|
1698 | walkback_(NULL), |
---|
1699 | addedCuts_(NULL), |
---|
1700 | nextRowCut_(NULL), |
---|
1701 | currentNode_(NULL), |
---|
1702 | specialOptions_(0), |
---|
1703 | subTreeModel_(NULL), |
---|
1704 | numberStoppedSubTrees_(0), |
---|
1705 | presolve_(0), |
---|
1706 | numberStrong_(5), |
---|
1707 | numberBeforeTrust_(0), |
---|
1708 | numberPenalties_(20), |
---|
1709 | penaltyScaleFactor_(3.0), |
---|
1710 | numberInfeasibleNodes_(0), |
---|
1711 | problemType_(0), |
---|
1712 | printFrequency_(0), |
---|
1713 | numberCutGenerators_(0), |
---|
1714 | generator_(NULL), |
---|
1715 | virginGenerator_(NULL), |
---|
1716 | numberHeuristics_(0), |
---|
1717 | heuristic_(NULL), |
---|
1718 | lastHeuristic_(NULL), |
---|
1719 | numberObjects_(0), |
---|
1720 | object_(NULL), |
---|
1721 | originalColumns_(NULL), |
---|
1722 | howOftenGlobalScan_(1), |
---|
1723 | numberGlobalViolations_(0), |
---|
1724 | continuousObjective_(COIN_DBL_MAX), |
---|
1725 | originalContinuousObjective_(COIN_DBL_MAX), |
---|
1726 | continuousInfeasibilities_(INT_MAX), |
---|
1727 | maximumCutPassesAtRoot_(20), |
---|
1728 | maximumCutPasses_(10), |
---|
1729 | resolveAfterTakeOffCuts_(true) |
---|
1730 | { |
---|
1731 | intParam_[CbcMaxNumNode] = 2147483647; |
---|
1732 | intParam_[CbcMaxNumSol] = 9999999; |
---|
1733 | intParam_[CbcFathomDiscipline] = 0; |
---|
1734 | |
---|
1735 | dblParam_[CbcIntegerTolerance] = 1e-6; |
---|
1736 | dblParam_[CbcInfeasibilityWeight] = 0.0; |
---|
1737 | dblParam_[CbcCutoffIncrement] = 1e-5; |
---|
1738 | dblParam_[CbcAllowableGap] = 1.0e-10; |
---|
1739 | dblParam_[CbcAllowableFractionGap] = 0.0; |
---|
1740 | dblParam_[CbcMaximumSeconds] = 1.0e100; |
---|
1741 | dblParam_[CbcStartSeconds] = 0.0; |
---|
1742 | |
---|
1743 | nodeCompare_=new CbcCompareDefault();; |
---|
1744 | problemFeasibility_=new CbcFeasibilityBase(); |
---|
1745 | tree_= new CbcTree(); |
---|
1746 | branchingMethod_=NULL; |
---|
1747 | strategy_=NULL; |
---|
1748 | parentModel_=NULL; |
---|
1749 | appData_=NULL; |
---|
1750 | handler_ = new CoinMessageHandler(); |
---|
1751 | handler_->setLogLevel(2); |
---|
1752 | messages_ = CbcMessage(); |
---|
1753 | solver_ = rhs.clone(); |
---|
1754 | ourSolver_ = true ; |
---|
1755 | |
---|
1756 | // Initialize solution and integer variable vectors |
---|
1757 | bestSolution_ = NULL; // to say no solution found |
---|
1758 | numberIntegers_=0; |
---|
1759 | int numberColumns = solver_->getNumCols(); |
---|
1760 | int iColumn; |
---|
1761 | if (numberColumns) { |
---|
1762 | // Space for current solution |
---|
1763 | currentSolution_ = new double[numberColumns]; |
---|
1764 | continuousSolution_ = new double[numberColumns]; |
---|
1765 | usedInSolution_ = new int[numberColumns]; |
---|
1766 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
1767 | if( solver_->isInteger(iColumn)) |
---|
1768 | numberIntegers_++; |
---|
1769 | } |
---|
1770 | } else { |
---|
1771 | // empty model |
---|
1772 | currentSolution_=NULL; |
---|
1773 | continuousSolution_=NULL; |
---|
1774 | usedInSolution_=NULL; |
---|
1775 | } |
---|
1776 | testSolution_=currentSolution_; |
---|
1777 | if (numberIntegers_) { |
---|
1778 | integerVariable_ = new int [numberIntegers_]; |
---|
1779 | numberIntegers_=0; |
---|
1780 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
1781 | if( solver_->isInteger(iColumn)) |
---|
1782 | integerVariable_[numberIntegers_++]=iColumn; |
---|
1783 | } |
---|
1784 | } else { |
---|
1785 | integerVariable_ = NULL; |
---|
1786 | } |
---|
1787 | } |
---|
1788 | |
---|
1789 | /* |
---|
1790 | Assign a solver to the model (model assumes ownership) |
---|
1791 | |
---|
1792 | The integer variable vector is initialized if it's not already present. |
---|
1793 | |
---|
1794 | Assuming ownership matches usage in OsiSolverInterface |
---|
1795 | (cf. assignProblem, loadProblem). |
---|
1796 | |
---|
1797 | TODO: What to do about solver parameters? A simple copy likely won't do it, |
---|
1798 | because the SI must push the settings into the underlying solver. In |
---|
1799 | the context of switching solvers in cbc, this means that command line |
---|
1800 | settings will get lost. Stash the command line somewhere and reread it |
---|
1801 | here, maybe? |
---|
1802 | |
---|
1803 | TODO: More generally, how much state should be transferred from the old |
---|
1804 | solver to the new solver? Best perhaps to see how usage develops. |
---|
1805 | What's done here mimics the CbcModel(OsiSolverInterface) constructor. |
---|
1806 | */ |
---|
1807 | void |
---|
1808 | CbcModel::assignSolver(OsiSolverInterface *&solver) |
---|
1809 | |
---|
1810 | { |
---|
1811 | // Keep the current message level for solver (if solver exists) |
---|
1812 | if (solver_) |
---|
1813 | solver->messageHandler()->setLogLevel(solver_->messageHandler()->logLevel()) ; |
---|
1814 | |
---|
1815 | if (ourSolver_) delete solver_ ; |
---|
1816 | solver_ = solver; |
---|
1817 | solver = NULL ; |
---|
1818 | ourSolver_ = true ; |
---|
1819 | /* |
---|
1820 | Basis information is solver-specific. |
---|
1821 | */ |
---|
1822 | if (basis_) |
---|
1823 | { delete basis_ ; |
---|
1824 | basis_ = 0 ; } |
---|
1825 | if (emptyWarmStart_) |
---|
1826 | { delete emptyWarmStart_ ; |
---|
1827 | emptyWarmStart_ = 0 ; } |
---|
1828 | /* |
---|
1829 | Initialize integer variable vector. |
---|
1830 | */ |
---|
1831 | numberIntegers_=0; |
---|
1832 | int numberColumns = solver_->getNumCols(); |
---|
1833 | int iColumn; |
---|
1834 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
1835 | if( solver_->isInteger(iColumn)) |
---|
1836 | numberIntegers_++; |
---|
1837 | } |
---|
1838 | if (numberIntegers_) { |
---|
1839 | delete [] integerVariable_; |
---|
1840 | integerVariable_ = new int [numberIntegers_]; |
---|
1841 | numberIntegers_=0; |
---|
1842 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
1843 | if( solver_->isInteger(iColumn)) |
---|
1844 | integerVariable_[numberIntegers_++]=iColumn; |
---|
1845 | } |
---|
1846 | } else { |
---|
1847 | integerVariable_ = NULL; |
---|
1848 | } |
---|
1849 | |
---|
1850 | return ; |
---|
1851 | } |
---|
1852 | |
---|
1853 | // Copy constructor. |
---|
1854 | |
---|
1855 | CbcModel::CbcModel(const CbcModel & rhs, bool noTree) |
---|
1856 | : |
---|
1857 | continuousSolver_(NULL), |
---|
1858 | defaultHandler_(rhs.defaultHandler_), |
---|
1859 | emptyWarmStart_(NULL), |
---|
1860 | basis_(NULL), |
---|
1861 | bestObjective_(rhs.bestObjective_), |
---|
1862 | bestPossibleObjective_(rhs.bestPossibleObjective_), |
---|
1863 | sumChangeObjective1_(rhs.sumChangeObjective1_), |
---|
1864 | sumChangeObjective2_(rhs.sumChangeObjective2_), |
---|
1865 | minimumDrop_(rhs.minimumDrop_), |
---|
1866 | numberSolutions_(rhs.numberSolutions_), |
---|
1867 | stateOfSearch_(rhs.stateOfSearch_), |
---|
1868 | hotstartStrategy_(rhs.hotstartStrategy_), |
---|
1869 | numberHeuristicSolutions_(rhs.numberHeuristicSolutions_), |
---|
1870 | numberNodes_(rhs.numberNodes_), |
---|
1871 | numberNodes2_(rhs.numberNodes2_), |
---|
1872 | numberIterations_(rhs.numberIterations_), |
---|
1873 | status_(rhs.status_), |
---|
1874 | secondaryStatus_(rhs.secondaryStatus_), |
---|
1875 | specialOptions_(rhs.specialOptions_), |
---|
1876 | subTreeModel_(rhs.subTreeModel_), |
---|
1877 | numberStoppedSubTrees_(rhs.numberStoppedSubTrees_), |
---|
1878 | presolve_(rhs.presolve_), |
---|
1879 | numberStrong_(rhs.numberStrong_), |
---|
1880 | numberBeforeTrust_(rhs.numberBeforeTrust_), |
---|
1881 | numberPenalties_(rhs.numberPenalties_), |
---|
1882 | penaltyScaleFactor_(penaltyScaleFactor_), |
---|
1883 | numberInfeasibleNodes_(rhs.numberInfeasibleNodes_), |
---|
1884 | problemType_(rhs.problemType_), |
---|
1885 | printFrequency_(rhs.printFrequency_), |
---|
1886 | howOftenGlobalScan_(rhs.howOftenGlobalScan_), |
---|
1887 | numberGlobalViolations_(rhs.numberGlobalViolations_), |
---|
1888 | continuousObjective_(rhs.continuousObjective_), |
---|
1889 | originalContinuousObjective_(rhs.originalContinuousObjective_), |
---|
1890 | continuousInfeasibilities_(rhs.continuousInfeasibilities_), |
---|
1891 | maximumCutPassesAtRoot_(rhs.maximumCutPassesAtRoot_), |
---|
1892 | maximumCutPasses_( rhs.maximumCutPasses_), |
---|
1893 | resolveAfterTakeOffCuts_(rhs.resolveAfterTakeOffCuts_) |
---|
1894 | { |
---|
1895 | intParam_[CbcMaxNumNode] = rhs.intParam_[CbcMaxNumNode]; |
---|
1896 | intParam_[CbcMaxNumSol] = rhs.intParam_[CbcMaxNumSol]; |
---|
1897 | intParam_[CbcFathomDiscipline] = rhs.intParam_[CbcFathomDiscipline]; |
---|
1898 | dblParam_[CbcIntegerTolerance] = rhs.dblParam_[CbcIntegerTolerance]; |
---|
1899 | dblParam_[CbcInfeasibilityWeight] = rhs.dblParam_[CbcInfeasibilityWeight]; |
---|
1900 | dblParam_[CbcCutoffIncrement] = rhs.dblParam_[CbcCutoffIncrement]; |
---|
1901 | dblParam_[CbcAllowableGap] = rhs.dblParam_[CbcAllowableGap]; |
---|
1902 | dblParam_[CbcAllowableFractionGap] = rhs.dblParam_[CbcAllowableFractionGap]; |
---|
1903 | dblParam_[CbcMaximumSeconds] = rhs.dblParam_[CbcMaximumSeconds]; |
---|
1904 | dblParam_[CbcStartSeconds] = dblParam_[CbcStartSeconds]; // will be overwritten hopefully |
---|
1905 | if (rhs.emptyWarmStart_) emptyWarmStart_ = rhs.emptyWarmStart_->clone() ; |
---|
1906 | if (rhs.basis_) basis_ = |
---|
1907 | dynamic_cast<CoinWarmStartBasis *>(rhs.basis_->clone()) ; |
---|
1908 | if (defaultHandler_) { |
---|
1909 | handler_ = new CoinMessageHandler(); |
---|
1910 | handler_->setLogLevel(2); |
---|
1911 | } else { |
---|
1912 | handler_ = rhs.handler_; |
---|
1913 | } |
---|
1914 | messageHandler()->setLogLevel(rhs.messageHandler()->logLevel()); |
---|
1915 | numberCutGenerators_ = rhs.numberCutGenerators_; |
---|
1916 | if (numberCutGenerators_) { |
---|
1917 | generator_ = new CbcCutGenerator * [numberCutGenerators_]; |
---|
1918 | virginGenerator_ = new CbcCutGenerator * [numberCutGenerators_]; |
---|
1919 | int i; |
---|
1920 | for (i=0;i<numberCutGenerators_;i++) { |
---|
1921 | generator_[i]=new CbcCutGenerator(*rhs.generator_[i]); |
---|
1922 | virginGenerator_[i]=new CbcCutGenerator(*rhs.virginGenerator_[i]); |
---|
1923 | } |
---|
1924 | } else { |
---|
1925 | generator_=NULL; |
---|
1926 | virginGenerator_=NULL; |
---|
1927 | } |
---|
1928 | if (!noTree) |
---|
1929 | globalCuts_ = rhs.globalCuts_; |
---|
1930 | numberHeuristics_ = rhs.numberHeuristics_; |
---|
1931 | if (numberHeuristics_) { |
---|
1932 | heuristic_ = new CbcHeuristic * [numberHeuristics_]; |
---|
1933 | int i; |
---|
1934 | for (i=0;i<numberHeuristics_;i++) { |
---|
1935 | heuristic_[i]=rhs.heuristic_[i]->clone(); |
---|
1936 | } |
---|
1937 | } else { |
---|
1938 | heuristic_=NULL; |
---|
1939 | } |
---|
1940 | lastHeuristic_ = NULL; |
---|
1941 | numberObjects_=rhs.numberObjects_; |
---|
1942 | if (numberObjects_) { |
---|
1943 | object_ = new CbcObject * [numberObjects_]; |
---|
1944 | int i; |
---|
1945 | for (i=0;i<numberObjects_;i++) |
---|
1946 | object_[i]=(rhs.object_[i])->clone(); |
---|
1947 | } else { |
---|
1948 | object_=NULL; |
---|
1949 | } |
---|
1950 | if (!noTree||!rhs.continuousSolver_) |
---|
1951 | solver_ = rhs.solver_->clone(); |
---|
1952 | else |
---|
1953 | solver_ = rhs.continuousSolver_->clone(); |
---|
1954 | if (rhs.originalColumns_) { |
---|
1955 | int numberColumns = solver_->getNumCols(); |
---|
1956 | originalColumns_= new int [numberColumns]; |
---|
1957 | memcpy(originalColumns_,rhs.originalColumns_,numberColumns*sizeof(int)); |
---|
1958 | } else { |
---|
1959 | originalColumns_=NULL; |
---|
1960 | } |
---|
1961 | nodeCompare_=rhs.nodeCompare_->clone(); |
---|
1962 | problemFeasibility_=rhs.problemFeasibility_->clone(); |
---|
1963 | tree_= rhs.tree_->clone(); |
---|
1964 | branchingMethod_=rhs.branchingMethod_; |
---|
1965 | if (rhs.strategy_) |
---|
1966 | strategy_=rhs.strategy_->clone(); |
---|
1967 | else |
---|
1968 | strategy_=NULL; |
---|
1969 | parentModel_=rhs.parentModel_; |
---|
1970 | appData_=rhs.appData_; |
---|
1971 | messages_ = rhs.messages_; |
---|
1972 | ourSolver_ = true ; |
---|
1973 | messageHandler()->setLogLevel(rhs.messageHandler()->logLevel()); |
---|
1974 | numberIntegers_=rhs.numberIntegers_; |
---|
1975 | if (numberIntegers_) { |
---|
1976 | integerVariable_ = new int [numberIntegers_]; |
---|
1977 | memcpy(integerVariable_,rhs.integerVariable_,numberIntegers_*sizeof(int)); |
---|
1978 | } else { |
---|
1979 | integerVariable_ = NULL; |
---|
1980 | } |
---|
1981 | if (rhs.bestSolution_&&!noTree) { |
---|
1982 | int numberColumns = solver_->getNumCols(); |
---|
1983 | bestSolution_ = new double[numberColumns]; |
---|
1984 | memcpy(bestSolution_,rhs.bestSolution_,numberColumns*sizeof(double)); |
---|
1985 | } else { |
---|
1986 | bestSolution_=NULL; |
---|
1987 | } |
---|
1988 | if (!noTree) { |
---|
1989 | int numberColumns = solver_->getNumCols(); |
---|
1990 | currentSolution_ = CoinCopyOfArray(rhs.currentSolution_,numberColumns); |
---|
1991 | continuousSolution_ = CoinCopyOfArray(rhs.continuousSolution_,numberColumns); |
---|
1992 | usedInSolution_ = CoinCopyOfArray(rhs.usedInSolution_,numberColumns); |
---|
1993 | } else { |
---|
1994 | currentSolution_=NULL; |
---|
1995 | continuousSolution_=NULL; |
---|
1996 | usedInSolution_=NULL; |
---|
1997 | } |
---|
1998 | testSolution_=currentSolution_; |
---|
1999 | numberRowsAtContinuous_ = rhs.numberRowsAtContinuous_; |
---|
2000 | maximumNumberCuts_=rhs.maximumNumberCuts_; |
---|
2001 | phase_ = rhs.phase_; |
---|
2002 | currentNumberCuts_=rhs.currentNumberCuts_; |
---|
2003 | maximumDepth_= rhs.maximumDepth_; |
---|
2004 | if (noTree) { |
---|
2005 | bestObjective_ = COIN_DBL_MAX; |
---|
2006 | numberSolutions_ =0; |
---|
2007 | stateOfSearch_= 0; |
---|
2008 | numberHeuristicSolutions_=0; |
---|
2009 | numberNodes_=0; |
---|
2010 | numberNodes2_=0; |
---|
2011 | numberIterations_=0; |
---|
2012 | status_=0; |
---|
2013 | subTreeModel_=NULL; |
---|
2014 | numberStoppedSubTrees_=0; |
---|
2015 | continuousObjective_=COIN_DBL_MAX; |
---|
2016 | originalContinuousObjective_=COIN_DBL_MAX; |
---|
2017 | continuousInfeasibilities_=INT_MAX; |
---|
2018 | maximumNumberCuts_=0; |
---|
2019 | tree_->cleanTree(this,-COIN_DBL_MAX,bestPossibleObjective_); |
---|
2020 | bestPossibleObjective_ = COIN_DBL_MAX; |
---|
2021 | } |
---|
2022 | // These are only used as temporary arrays so need not be filled |
---|
2023 | if (maximumNumberCuts_) { |
---|
2024 | addedCuts_ = new CbcCountRowCut * [maximumNumberCuts_]; |
---|
2025 | } else { |
---|
2026 | addedCuts_ = NULL; |
---|
2027 | } |
---|
2028 | nextRowCut_ = NULL; |
---|
2029 | currentNode_ = NULL; |
---|
2030 | if (maximumDepth_) |
---|
2031 | walkback_ = new CbcNodeInfo * [maximumDepth_]; |
---|
2032 | else |
---|
2033 | walkback_ = NULL; |
---|
2034 | synchronizeModel(); |
---|
2035 | } |
---|
2036 | |
---|
2037 | // Assignment operator |
---|
2038 | CbcModel & |
---|
2039 | CbcModel::operator=(const CbcModel& rhs) |
---|
2040 | { |
---|
2041 | if (this!=&rhs) { |
---|
2042 | |
---|
2043 | if (defaultHandler_) |
---|
2044 | { delete handler_; |
---|
2045 | handler_ = NULL; } |
---|
2046 | defaultHandler_ = rhs.defaultHandler_; |
---|
2047 | if (defaultHandler_) |
---|
2048 | { handler_ = new CoinMessageHandler(); |
---|
2049 | handler_->setLogLevel(2); } |
---|
2050 | else |
---|
2051 | { handler_ = rhs.handler_; } |
---|
2052 | messages_ = rhs.messages_; |
---|
2053 | messageHandler()->setLogLevel(rhs.messageHandler()->logLevel()); |
---|
2054 | |
---|
2055 | delete solver_; |
---|
2056 | if (rhs.solver_) |
---|
2057 | { solver_ = rhs.solver_->clone() ; } |
---|
2058 | else |
---|
2059 | { solver_ = 0 ; } |
---|
2060 | ourSolver_ = true ; |
---|
2061 | delete continuousSolver_ ; |
---|
2062 | if (rhs.continuousSolver_) |
---|
2063 | { solver_ = rhs.continuousSolver_->clone() ; } |
---|
2064 | else |
---|
2065 | { continuousSolver_ = 0 ; } |
---|
2066 | |
---|
2067 | delete emptyWarmStart_ ; |
---|
2068 | if (rhs.emptyWarmStart_) |
---|
2069 | { emptyWarmStart_ = rhs.emptyWarmStart_->clone() ; } |
---|
2070 | else |
---|
2071 | { emptyWarmStart_ = 0 ; } |
---|
2072 | delete basis_ ; |
---|
2073 | if (rhs.basis_) |
---|
2074 | { basis_ = dynamic_cast<CoinWarmStartBasis *>(rhs.basis_->clone()) ; } |
---|
2075 | else |
---|
2076 | { basis_ = 0 ; } |
---|
2077 | |
---|
2078 | bestObjective_ = rhs.bestObjective_; |
---|
2079 | bestPossibleObjective_=rhs.bestPossibleObjective_; |
---|
2080 | sumChangeObjective1_=rhs.sumChangeObjective1_; |
---|
2081 | sumChangeObjective2_=rhs.sumChangeObjective2_; |
---|
2082 | delete [] bestSolution_; |
---|
2083 | if (rhs.bestSolution_) { |
---|
2084 | int numberColumns = rhs.getNumCols(); |
---|
2085 | bestSolution_ = new double[numberColumns]; |
---|
2086 | memcpy(bestSolution_,rhs.bestSolution_,numberColumns*sizeof(double)); |
---|
2087 | } else { |
---|
2088 | bestSolution_=NULL; |
---|
2089 | } |
---|
2090 | int numberColumns = solver_->getNumCols(); |
---|
2091 | currentSolution_ = CoinCopyOfArray(rhs.currentSolution_,numberColumns); |
---|
2092 | continuousSolution_ = CoinCopyOfArray(rhs.continuousSolution_,numberColumns); |
---|
2093 | usedInSolution_ = CoinCopyOfArray(rhs.usedInSolution_,numberColumns); |
---|
2094 | testSolution_=currentSolution_; |
---|
2095 | minimumDrop_ = rhs.minimumDrop_; |
---|
2096 | numberSolutions_=rhs.numberSolutions_; |
---|
2097 | stateOfSearch_= rhs.stateOfSearch_; |
---|
2098 | hotstartStrategy_=rhs.hotstartStrategy_; |
---|
2099 | numberHeuristicSolutions_=rhs.numberHeuristicSolutions_; |
---|
2100 | numberNodes_ = rhs.numberNodes_; |
---|
2101 | numberNodes2_ = rhs.numberNodes2_; |
---|
2102 | numberIterations_ = rhs.numberIterations_; |
---|
2103 | status_ = rhs.status_; |
---|
2104 | secondaryStatus_ = rhs.secondaryStatus_; |
---|
2105 | specialOptions_ = rhs.specialOptions_; |
---|
2106 | subTreeModel_ = rhs.subTreeModel_; |
---|
2107 | numberStoppedSubTrees_ = rhs.numberStoppedSubTrees_; |
---|
2108 | presolve_ = rhs.presolve_; |
---|
2109 | numberStrong_ = rhs.numberStrong_; |
---|
2110 | numberBeforeTrust_ = rhs.numberBeforeTrust_; |
---|
2111 | numberPenalties_ = rhs.numberPenalties_; |
---|
2112 | penaltyScaleFactor_ = penaltyScaleFactor_; |
---|
2113 | numberInfeasibleNodes_ = rhs.numberInfeasibleNodes_; |
---|
2114 | problemType_ = rhs.problemType_; |
---|
2115 | printFrequency_ = rhs.printFrequency_; |
---|
2116 | howOftenGlobalScan_=rhs.howOftenGlobalScan_; |
---|
2117 | numberGlobalViolations_=rhs.numberGlobalViolations_; |
---|
2118 | continuousObjective_=rhs.continuousObjective_; |
---|
2119 | originalContinuousObjective_ = rhs.originalContinuousObjective_; |
---|
2120 | continuousInfeasibilities_ = rhs.continuousInfeasibilities_; |
---|
2121 | maximumCutPassesAtRoot_ = rhs.maximumCutPassesAtRoot_; |
---|
2122 | maximumCutPasses_ = rhs.maximumCutPasses_; |
---|
2123 | resolveAfterTakeOffCuts_=rhs.resolveAfterTakeOffCuts_; |
---|
2124 | intParam_[CbcMaxNumNode] = rhs.intParam_[CbcMaxNumNode]; |
---|
2125 | intParam_[CbcMaxNumSol] = rhs.intParam_[CbcMaxNumSol]; |
---|
2126 | intParam_[CbcFathomDiscipline] = rhs.intParam_[CbcFathomDiscipline]; |
---|
2127 | dblParam_[CbcIntegerTolerance] = rhs.dblParam_[CbcIntegerTolerance]; |
---|
2128 | dblParam_[CbcInfeasibilityWeight] = rhs.dblParam_[CbcInfeasibilityWeight]; |
---|
2129 | dblParam_[CbcCutoffIncrement] = rhs.dblParam_[CbcCutoffIncrement]; |
---|
2130 | dblParam_[CbcAllowableGap] = rhs.dblParam_[CbcAllowableGap]; |
---|
2131 | dblParam_[CbcAllowableFractionGap] = rhs.dblParam_[CbcAllowableFractionGap]; |
---|
2132 | dblParam_[CbcMaximumSeconds] = rhs.dblParam_[CbcMaximumSeconds]; |
---|
2133 | dblParam_[CbcStartSeconds] = dblParam_[CbcStartSeconds]; // will be overwritten hopefully |
---|
2134 | globalCuts_ = rhs.globalCuts_; |
---|
2135 | int i; |
---|
2136 | for (i=0;i<numberCutGenerators_;i++) { |
---|
2137 | delete generator_[i]; |
---|
2138 | delete virginGenerator_[i]; |
---|
2139 | } |
---|
2140 | delete [] generator_; |
---|
2141 | delete [] virginGenerator_; |
---|
2142 | delete [] heuristic_; |
---|
2143 | lastHeuristic_ = NULL; |
---|
2144 | numberCutGenerators_ = rhs.numberCutGenerators_; |
---|
2145 | if (numberCutGenerators_) { |
---|
2146 | generator_ = new CbcCutGenerator * [numberCutGenerators_]; |
---|
2147 | virginGenerator_ = new CbcCutGenerator * [numberCutGenerators_]; |
---|
2148 | int i; |
---|
2149 | for (i=0;i<numberCutGenerators_;i++) { |
---|
2150 | generator_[i]=new CbcCutGenerator(*rhs.generator_[i]); |
---|
2151 | virginGenerator_[i]=new CbcCutGenerator(*rhs.virginGenerator_[i]); |
---|
2152 | } |
---|
2153 | } else { |
---|
2154 | generator_=NULL; |
---|
2155 | virginGenerator_=NULL; |
---|
2156 | } |
---|
2157 | numberHeuristics_ = rhs.numberHeuristics_; |
---|
2158 | if (numberHeuristics_) { |
---|
2159 | heuristic_ = new CbcHeuristic * [numberHeuristics_]; |
---|
2160 | memcpy(heuristic_,rhs.heuristic_, |
---|
2161 | numberHeuristics_*sizeof(CbcHeuristic *)); |
---|
2162 | } else { |
---|
2163 | heuristic_=NULL; |
---|
2164 | } |
---|
2165 | lastHeuristic_ = NULL; |
---|
2166 | for (i=0;i<numberObjects_;i++) |
---|
2167 | delete object_[i]; |
---|
2168 | delete [] object_; |
---|
2169 | numberObjects_=rhs.numberObjects_; |
---|
2170 | if (numberObjects_) { |
---|
2171 | object_ = new CbcObject * [numberObjects_]; |
---|
2172 | int i; |
---|
2173 | for (i=0;i<numberObjects_;i++) |
---|
2174 | object_[i]=(rhs.object_[i])->clone(); |
---|
2175 | } else { |
---|
2176 | object_=NULL; |
---|
2177 | } |
---|
2178 | delete [] originalColumns_; |
---|
2179 | if (rhs.originalColumns_) { |
---|
2180 | int numberColumns = rhs.getNumCols(); |
---|
2181 | originalColumns_= new int [numberColumns]; |
---|
2182 | memcpy(originalColumns_,rhs.originalColumns_,numberColumns*sizeof(int)); |
---|
2183 | } else { |
---|
2184 | originalColumns_=NULL; |
---|
2185 | } |
---|
2186 | nodeCompare_=rhs.nodeCompare_->clone(); |
---|
2187 | problemFeasibility_=rhs.problemFeasibility_->clone(); |
---|
2188 | delete tree_; |
---|
2189 | tree_= rhs.tree_->clone(); |
---|
2190 | branchingMethod_=rhs.branchingMethod_; |
---|
2191 | delete strategy_; |
---|
2192 | if (rhs.strategy_) |
---|
2193 | strategy_=rhs.strategy_->clone(); |
---|
2194 | else |
---|
2195 | strategy_=NULL; |
---|
2196 | parentModel_=rhs.parentModel_; |
---|
2197 | appData_=rhs.appData_; |
---|
2198 | |
---|
2199 | delete [] integerVariable_; |
---|
2200 | numberIntegers_=rhs.numberIntegers_; |
---|
2201 | if (numberIntegers_) { |
---|
2202 | integerVariable_ = new int [numberIntegers_]; |
---|
2203 | memcpy(integerVariable_,rhs.integerVariable_, |
---|
2204 | numberIntegers_*sizeof(int)); |
---|
2205 | } else { |
---|
2206 | integerVariable_ = NULL; |
---|
2207 | } |
---|
2208 | numberRowsAtContinuous_ = rhs.numberRowsAtContinuous_; |
---|
2209 | maximumNumberCuts_=rhs.maximumNumberCuts_; |
---|
2210 | phase_ = rhs.phase_; |
---|
2211 | currentNumberCuts_=rhs.currentNumberCuts_; |
---|
2212 | maximumDepth_= rhs.maximumDepth_; |
---|
2213 | delete [] addedCuts_; |
---|
2214 | delete [] walkback_; |
---|
2215 | // These are only used as temporary arrays so need not be filled |
---|
2216 | if (maximumNumberCuts_) { |
---|
2217 | addedCuts_ = new CbcCountRowCut * [maximumNumberCuts_]; |
---|
2218 | } else { |
---|
2219 | addedCuts_ = NULL; |
---|
2220 | } |
---|
2221 | nextRowCut_ = NULL; |
---|
2222 | currentNode_ = NULL; |
---|
2223 | if (maximumDepth_) |
---|
2224 | walkback_ = new CbcNodeInfo * [maximumDepth_]; |
---|
2225 | else |
---|
2226 | walkback_ = NULL; |
---|
2227 | synchronizeModel(); |
---|
2228 | } |
---|
2229 | return *this; |
---|
2230 | } |
---|
2231 | |
---|
2232 | // Destructor |
---|
2233 | CbcModel::~CbcModel () |
---|
2234 | { |
---|
2235 | if (defaultHandler_) { |
---|
2236 | delete handler_; |
---|
2237 | handler_ = NULL; |
---|
2238 | } |
---|
2239 | delete tree_; |
---|
2240 | if (ourSolver_) delete solver_; |
---|
2241 | gutsOfDestructor(); |
---|
2242 | } |
---|
2243 | // Clears out as much as possible (except solver) |
---|
2244 | void |
---|
2245 | CbcModel::gutsOfDestructor() |
---|
2246 | { |
---|
2247 | delete emptyWarmStart_ ; |
---|
2248 | emptyWarmStart_ =NULL; |
---|
2249 | delete basis_ ; |
---|
2250 | basis_ =NULL; |
---|
2251 | delete continuousSolver_; |
---|
2252 | continuousSolver_=NULL; |
---|
2253 | delete [] bestSolution_; |
---|
2254 | bestSolution_=NULL; |
---|
2255 | delete [] currentSolution_; |
---|
2256 | currentSolution_=NULL; |
---|
2257 | delete [] continuousSolution_; |
---|
2258 | continuousSolution_=NULL; |
---|
2259 | delete [] usedInSolution_; |
---|
2260 | usedInSolution_ = NULL; |
---|
2261 | testSolution_=NULL; |
---|
2262 | delete [] integerVariable_; |
---|
2263 | integerVariable_=NULL; |
---|
2264 | int i; |
---|
2265 | for (i=0;i<numberCutGenerators_;i++) { |
---|
2266 | delete generator_[i]; |
---|
2267 | delete virginGenerator_[i]; |
---|
2268 | } |
---|
2269 | delete [] generator_; |
---|
2270 | delete [] virginGenerator_; |
---|
2271 | generator_=NULL; |
---|
2272 | virginGenerator_=NULL; |
---|
2273 | for (i=0;i<numberHeuristics_;i++) |
---|
2274 | delete heuristic_[i]; |
---|
2275 | delete [] heuristic_; |
---|
2276 | heuristic_=NULL; |
---|
2277 | lastHeuristic_ = NULL; |
---|
2278 | delete nodeCompare_; |
---|
2279 | nodeCompare_=NULL; |
---|
2280 | delete problemFeasibility_; |
---|
2281 | problemFeasibility_=NULL; |
---|
2282 | delete [] addedCuts_; |
---|
2283 | addedCuts_=NULL; |
---|
2284 | nextRowCut_ = NULL; |
---|
2285 | currentNode_ = NULL; |
---|
2286 | delete [] walkback_; |
---|
2287 | walkback_=NULL; |
---|
2288 | for (i=0;i<numberObjects_;i++) |
---|
2289 | delete object_[i]; |
---|
2290 | delete [] object_; |
---|
2291 | object_=NULL; |
---|
2292 | delete [] originalColumns_; |
---|
2293 | originalColumns_=NULL; |
---|
2294 | delete strategy_; |
---|
2295 | } |
---|
2296 | // Are there a numerical difficulties? |
---|
2297 | bool |
---|
2298 | CbcModel::isAbandoned() const |
---|
2299 | { |
---|
2300 | return status_ == 2; |
---|
2301 | } |
---|
2302 | // Is optimality proven? |
---|
2303 | bool |
---|
2304 | CbcModel::isProvenOptimal() const |
---|
2305 | { |
---|
2306 | if (!status_ && bestObjective_<1.0e30) |
---|
2307 | return true; |
---|
2308 | else |
---|
2309 | return false; |
---|
2310 | } |
---|
2311 | // Is infeasiblity proven (or none better than cutoff)? |
---|
2312 | bool |
---|
2313 | CbcModel::isProvenInfeasible() const |
---|
2314 | { |
---|
2315 | if (!status_ && bestObjective_>=1.0e30) |
---|
2316 | return true; |
---|
2317 | else |
---|
2318 | return false; |
---|
2319 | } |
---|
2320 | // Node limit reached? |
---|
2321 | bool |
---|
2322 | CbcModel::isNodeLimitReached() const |
---|
2323 | { |
---|
2324 | return numberNodes_ >= intParam_[CbcMaxNumNode]; |
---|
2325 | } |
---|
2326 | // Time limit reached? |
---|
2327 | bool |
---|
2328 | CbcModel::isSecondsLimitReached() const |
---|
2329 | { |
---|
2330 | if (status_==1&&secondaryStatus_==4) |
---|
2331 | return true; |
---|
2332 | else |
---|
2333 | return false; |
---|
2334 | } |
---|
2335 | // Solution limit reached? |
---|
2336 | bool |
---|
2337 | CbcModel::isSolutionLimitReached() const |
---|
2338 | { |
---|
2339 | return numberSolutions_ >= intParam_[CbcMaxNumSol]; |
---|
2340 | } |
---|
2341 | // Set language |
---|
2342 | void |
---|
2343 | CbcModel::newLanguage(CoinMessages::Language language) |
---|
2344 | { |
---|
2345 | messages_ = CbcMessage(language); |
---|
2346 | } |
---|
2347 | void |
---|
2348 | CbcModel::setNumberStrong(int number) |
---|
2349 | { |
---|
2350 | if (number<0) |
---|
2351 | numberStrong_=0; |
---|
2352 | else |
---|
2353 | numberStrong_=number; |
---|
2354 | } |
---|
2355 | void |
---|
2356 | CbcModel::setNumberBeforeTrust(int number) |
---|
2357 | { |
---|
2358 | if (number<=0) { |
---|
2359 | numberBeforeTrust_=0; |
---|
2360 | } else { |
---|
2361 | numberBeforeTrust_=number; |
---|
2362 | //numberStrong_ = CoinMax(numberStrong_,1); |
---|
2363 | } |
---|
2364 | } |
---|
2365 | void |
---|
2366 | CbcModel::setNumberPenalties(int number) |
---|
2367 | { |
---|
2368 | if (number<=0) { |
---|
2369 | numberPenalties_=0; |
---|
2370 | } else { |
---|
2371 | numberPenalties_=number; |
---|
2372 | } |
---|
2373 | } |
---|
2374 | void |
---|
2375 | CbcModel::setPenaltyScaleFactor(double value) |
---|
2376 | { |
---|
2377 | if (value<=0) { |
---|
2378 | penaltyScaleFactor_=3.0; |
---|
2379 | } else { |
---|
2380 | penaltyScaleFactor_=value; |
---|
2381 | } |
---|
2382 | } |
---|
2383 | void |
---|
2384 | CbcModel::setHowOftenGlobalScan(int number) |
---|
2385 | { |
---|
2386 | if (number<-1) |
---|
2387 | howOftenGlobalScan_=0; |
---|
2388 | else |
---|
2389 | howOftenGlobalScan_=number; |
---|
2390 | } |
---|
2391 | |
---|
2392 | // Add one generator |
---|
2393 | void |
---|
2394 | CbcModel::addCutGenerator(CglCutGenerator * generator, |
---|
2395 | int howOften, const char * name, |
---|
2396 | bool normal, bool atSolution, |
---|
2397 | bool whenInfeasible,int howOftenInSub, |
---|
2398 | int whatDepth, int whatDepthInSub) |
---|
2399 | { |
---|
2400 | CbcCutGenerator ** temp = generator_; |
---|
2401 | generator_ = new CbcCutGenerator * [numberCutGenerators_+1]; |
---|
2402 | memcpy(generator_,temp,numberCutGenerators_*sizeof(CbcCutGenerator *)); |
---|
2403 | delete[] temp ; |
---|
2404 | generator_[numberCutGenerators_]= |
---|
2405 | new CbcCutGenerator(this,generator, howOften, name, |
---|
2406 | normal,atSolution,whenInfeasible,howOftenInSub, |
---|
2407 | whatDepth, whatDepthInSub); |
---|
2408 | // and before any cahnges |
---|
2409 | temp = virginGenerator_; |
---|
2410 | virginGenerator_ = new CbcCutGenerator * [numberCutGenerators_+1]; |
---|
2411 | memcpy(virginGenerator_,temp,numberCutGenerators_*sizeof(CbcCutGenerator *)); |
---|
2412 | delete[] temp ; |
---|
2413 | virginGenerator_[numberCutGenerators_++]= |
---|
2414 | new CbcCutGenerator(this,generator, howOften, name, |
---|
2415 | normal,atSolution,whenInfeasible,howOftenInSub, |
---|
2416 | whatDepth, whatDepthInSub); |
---|
2417 | |
---|
2418 | } |
---|
2419 | // Add one heuristic |
---|
2420 | void |
---|
2421 | CbcModel::addHeuristic(CbcHeuristic * generator) |
---|
2422 | { |
---|
2423 | CbcHeuristic ** temp = heuristic_; |
---|
2424 | heuristic_ = new CbcHeuristic * [numberHeuristics_+1]; |
---|
2425 | memcpy(heuristic_,temp,numberHeuristics_*sizeof(CbcHeuristic *)); |
---|
2426 | delete [] temp; |
---|
2427 | heuristic_[numberHeuristics_++]=generator->clone(); |
---|
2428 | } |
---|
2429 | |
---|
2430 | /* |
---|
2431 | The last subproblem handled by the solver is not necessarily related to the |
---|
2432 | one being recreated, so the first action is to remove all cuts from the |
---|
2433 | constraint system. Next, traverse the tree from node to the root to |
---|
2434 | determine the basis size required for this subproblem and create an empty |
---|
2435 | basis with the right capacity. Finally, traverse the tree from root to |
---|
2436 | node, adjusting bounds in the constraint system, adjusting the basis, and |
---|
2437 | collecting the cuts that must be added to the constraint system. |
---|
2438 | applyToModel does the heavy lifting. |
---|
2439 | |
---|
2440 | addCuts1 is used in contexts where all that's desired is the list of cuts: |
---|
2441 | the node is already fathomed, and we're collecting cuts so that we can |
---|
2442 | adjust reference counts as we prune nodes. Arguably the two functions |
---|
2443 | should be separated. The culprit is applyToModel, which performs cut |
---|
2444 | collection and model adjustment. |
---|
2445 | |
---|
2446 | Certainly in the contexts where all we need is a list of cuts, there's no |
---|
2447 | point in passing in a valid basis --- an empty basis will do just fine. |
---|
2448 | */ |
---|
2449 | void CbcModel::addCuts1 (CbcNode * node, CoinWarmStartBasis *&lastws) |
---|
2450 | { int i; |
---|
2451 | int nNode=0; |
---|
2452 | int numberColumns = getNumCols(); |
---|
2453 | CbcNodeInfo * nodeInfo = node->nodeInfo(); |
---|
2454 | |
---|
2455 | /* |
---|
2456 | Remove all cuts from the constraint system. |
---|
2457 | (original comment includes ``see note below for later efficiency'', but |
---|
2458 | the reference isn't clear to me). |
---|
2459 | */ |
---|
2460 | int currentNumberCuts = solver_->getNumRows()-numberRowsAtContinuous_; |
---|
2461 | int *which = new int[currentNumberCuts]; |
---|
2462 | for (i = 0 ; i < currentNumberCuts ; i++) |
---|
2463 | which[i] = i+numberRowsAtContinuous_; |
---|
2464 | solver_->deleteRows(currentNumberCuts,which); |
---|
2465 | delete [] which; |
---|
2466 | /* |
---|
2467 | Accumulate the path from node to the root in walkback_, and accumulate a |
---|
2468 | cut count in currentNumberCuts. |
---|
2469 | |
---|
2470 | original comment: when working then just unwind until where new node joins |
---|
2471 | old node (for cuts?) |
---|
2472 | */ |
---|
2473 | currentNumberCuts = 0; |
---|
2474 | while (nodeInfo) { |
---|
2475 | //printf("nNode = %d, nodeInfo = %x\n",nNode,nodeInfo); |
---|
2476 | walkback_[nNode++]=nodeInfo; |
---|
2477 | currentNumberCuts += nodeInfo->numberCuts() ; |
---|
2478 | nodeInfo = nodeInfo->parent() ; |
---|
2479 | if (nNode==maximumDepth_) { |
---|
2480 | maximumDepth_ *= 2; |
---|
2481 | CbcNodeInfo ** temp = new CbcNodeInfo * [maximumDepth_]; |
---|
2482 | for (i=0;i<nNode;i++) |
---|
2483 | temp[i] = walkback_[i]; |
---|
2484 | delete [] walkback_; |
---|
2485 | walkback_ = temp; |
---|
2486 | } |
---|
2487 | } |
---|
2488 | /* |
---|
2489 | Create an empty basis with sufficient capacity for the constraint system |
---|
2490 | we'll construct: original system plus cuts. Make sure we have capacity to |
---|
2491 | record those cuts in addedCuts_. |
---|
2492 | |
---|
2493 | The method of adjusting the basis at a FullNodeInfo object (the root, for |
---|
2494 | example) is to use a copy constructor to duplicate the basis held in the |
---|
2495 | nodeInfo, then resize it and return the new basis object. Guaranteed, |
---|
2496 | lastws will point to a different basis when it returns. We pass in a basis |
---|
2497 | because we need the parameter to return the allocated basis, and it's an |
---|
2498 | easy way to pass in the size. But we take a hit for memory allocation. |
---|
2499 | */ |
---|
2500 | currentNumberCuts_=currentNumberCuts; |
---|
2501 | if (currentNumberCuts >= maximumNumberCuts_) { |
---|
2502 | maximumNumberCuts_ = currentNumberCuts; |
---|
2503 | delete [] addedCuts_; |
---|
2504 | addedCuts_ = new CbcCountRowCut * [maximumNumberCuts_]; |
---|
2505 | } |
---|
2506 | lastws->setSize(numberColumns,numberRowsAtContinuous_+currentNumberCuts); |
---|
2507 | /* |
---|
2508 | This last bit of code traverses the path collected in walkback_ from the |
---|
2509 | root back to node. At the end of the loop, |
---|
2510 | * lastws will be an appropriate basis for node; |
---|
2511 | * variable bounds in the constraint system will be set to be correct for |
---|
2512 | node; and |
---|
2513 | * addedCuts_ will be set to a list of cuts that need to be added to the |
---|
2514 | constraint system at node. |
---|
2515 | applyToModel does all the heavy lifting. |
---|
2516 | */ |
---|
2517 | currentNumberCuts=0; |
---|
2518 | while (nNode) { |
---|
2519 | --nNode; |
---|
2520 | walkback_[nNode]->applyToModel(this,lastws,addedCuts_,currentNumberCuts); |
---|
2521 | } |
---|
2522 | } |
---|
2523 | |
---|
2524 | /* |
---|
2525 | adjustCuts might be a better name: If the node is feasible, we sift through |
---|
2526 | the cuts we've collected, add the ones that are tight and omit the ones that |
---|
2527 | are loose. If the node is infeasible, we just adjust the reference counts to |
---|
2528 | reflect that we're about to prune this node and its descendants. |
---|
2529 | |
---|
2530 | The reason we need to pass in lastws is that OsiClp automagically corrects |
---|
2531 | the basis when it deletes constraints. So when all cuts are stripped within |
---|
2532 | addCuts1, we lose their basis entries, hence the ability to determine if |
---|
2533 | they are loose or tight. The question is whether we really need to pass in |
---|
2534 | a basis or if we can capture it here. I'm thinking we can capture it here |
---|
2535 | and pass it back out if required. |
---|
2536 | */ |
---|
2537 | int CbcModel::addCuts (CbcNode *node, CoinWarmStartBasis *&lastws) |
---|
2538 | { |
---|
2539 | /* |
---|
2540 | addCuts1 performs step 1 of restoring the subproblem at this node; see the |
---|
2541 | comments there. |
---|
2542 | */ |
---|
2543 | addCuts1(node,lastws); |
---|
2544 | int i; |
---|
2545 | int numberColumns = getNumCols(); |
---|
2546 | CbcNodeInfo * nodeInfo = node->nodeInfo(); |
---|
2547 | double cutoff = getCutoff() ; |
---|
2548 | int currentNumberCuts=currentNumberCuts_; |
---|
2549 | /* |
---|
2550 | If the node can't be fathomed by bound, reinstall tight cuts in the |
---|
2551 | constraint system. |
---|
2552 | */ |
---|
2553 | if (node->objectiveValue() < cutoff) |
---|
2554 | { int numberToAdd = 0; |
---|
2555 | const OsiRowCut * * addCuts; |
---|
2556 | if (currentNumberCuts == 0) |
---|
2557 | addCuts = NULL; |
---|
2558 | else |
---|
2559 | addCuts = new const OsiRowCut * [currentNumberCuts]; |
---|
2560 | # ifdef CHECK_CUT_COUNTS |
---|
2561 | printf("addCuts: expanded basis; rows %d+%d\n", |
---|
2562 | numberRowsAtContinuous_,currentNumberCuts); |
---|
2563 | lastws->print(); |
---|
2564 | # endif |
---|
2565 | /* |
---|
2566 | Adjust the basis and constraint system so that we retain only active cuts. |
---|
2567 | There are three steps: |
---|
2568 | 1) Scan the basis. If the logical associated with the cut is basic, it's |
---|
2569 | loose and we drop it. The status of the logical for tight cuts is |
---|
2570 | written back into the status array, compressing as we go. |
---|
2571 | 2) Resize the basis to fit the number of active cuts, stash a clone, and |
---|
2572 | install with a call to setWarmStart(). |
---|
2573 | 3) Install the tight cuts into the constraint system (applyRowCuts). |
---|
2574 | |
---|
2575 | TODO: After working through the code in createInfo, I'm more comfortable if |
---|
2576 | inactive cuts are retained in lastws. So, instead of cloning |
---|
2577 | lastws into basis_ after the compression loop, do it ahead of time |
---|
2578 | and then recover lastws from basis_ after the setWarmStart(). |
---|
2579 | (Minimal code change :-). See CbcNode::createInfo for more. |
---|
2580 | */ |
---|
2581 | if (basis_) delete basis_ ; |
---|
2582 | basis_= dynamic_cast<CoinWarmStartBasis *>(lastws->clone()) ; |
---|
2583 | for (i=0;i<currentNumberCuts;i++) { |
---|
2584 | CoinWarmStartBasis::Status status = |
---|
2585 | lastws->getArtifStatus(i+numberRowsAtContinuous_); |
---|
2586 | if (status != CoinWarmStartBasis::basic&&addedCuts_[i]) { |
---|
2587 | # ifdef CHECK_CUT_COUNTS |
---|
2588 | printf("Using cut %d %x as row %d\n",i,addedCuts_[i], |
---|
2589 | numberRowsAtContinuous_+numberToAdd); |
---|
2590 | # endif |
---|
2591 | lastws->setArtifStatus(numberToAdd+numberRowsAtContinuous_,status); |
---|
2592 | addCuts[numberToAdd++] = new OsiRowCut(*addedCuts_[i]); |
---|
2593 | } else { |
---|
2594 | # ifdef CHECK_CUT_COUNTS |
---|
2595 | printf("Dropping cut %d %x\n",i,addedCuts_[i]); |
---|
2596 | # endif |
---|
2597 | addedCuts_[i]=NULL; |
---|
2598 | } |
---|
2599 | } |
---|
2600 | int numberRowsNow=numberRowsAtContinuous_+numberToAdd; |
---|
2601 | lastws->resize(numberRowsNow,numberColumns); |
---|
2602 | #ifdef FULL_DEBUG |
---|
2603 | printf("addCuts: stripped basis; rows %d + %d\n", |
---|
2604 | numberRowsAtContinuous_,numberToAdd); |
---|
2605 | lastws->print(); |
---|
2606 | #endif |
---|
2607 | /* |
---|
2608 | Apply the cuts and set the basis in the solver. |
---|
2609 | */ |
---|
2610 | solver_->applyRowCuts(numberToAdd,addCuts); |
---|
2611 | solver_->setWarmStart(lastws); |
---|
2612 | /* |
---|
2613 | TODO: Undo the debugging change. Delete lastws and assign basis_. |
---|
2614 | */ |
---|
2615 | delete lastws ; |
---|
2616 | lastws = basis_ ; |
---|
2617 | basis_ = 0 ; |
---|
2618 | |
---|
2619 | #if 0 |
---|
2620 | if ((numberNodes_%printFrequency_)==0) { |
---|
2621 | printf("Objective %g, depth %d, unsatisfied %d\n", |
---|
2622 | node->objectiveValue(), |
---|
2623 | node->depth(),node->numberUnsatisfied()); |
---|
2624 | } |
---|
2625 | #endif |
---|
2626 | /* |
---|
2627 | Clean up and we're out of here. |
---|
2628 | */ |
---|
2629 | for (i=0;i<numberToAdd;i++) |
---|
2630 | delete addCuts[i]; |
---|
2631 | delete [] addCuts; |
---|
2632 | numberNodes_++; |
---|
2633 | return 0; |
---|
2634 | } |
---|
2635 | /* |
---|
2636 | This node has been fathomed by bound as we try to revive it out of the live |
---|
2637 | set. Adjust the cut reference counts to reflect that we no longer need to |
---|
2638 | explore the remaining branch arms, hence they will no longer reference any |
---|
2639 | cuts. Cuts whose reference count falls to zero are deleted. |
---|
2640 | */ |
---|
2641 | else |
---|
2642 | { int i; |
---|
2643 | int numberLeft = nodeInfo->numberBranchesLeft(); |
---|
2644 | for (i = 0 ; i < currentNumberCuts ; i++) |
---|
2645 | { if (addedCuts_[i]) |
---|
2646 | { if (!addedCuts_[i]->decrement(numberLeft)) |
---|
2647 | { delete addedCuts_[i]; |
---|
2648 | addedCuts_[i] = NULL; } } } |
---|
2649 | return 1 ; } |
---|
2650 | } |
---|
2651 | |
---|
2652 | |
---|
2653 | /* |
---|
2654 | Perform reduced cost fixing on integer variables. |
---|
2655 | |
---|
2656 | The variables in question are already nonbasic at bound. We're just nailing |
---|
2657 | down the current situation. |
---|
2658 | */ |
---|
2659 | |
---|
2660 | void CbcModel::reducedCostFix () |
---|
2661 | |
---|
2662 | { double cutoff = getCutoff() ; |
---|
2663 | double direction = solver_->getObjSense() ; |
---|
2664 | double gap = cutoff - solver_->getObjValue()*direction ; |
---|
2665 | double integerTolerance = getDblParam(CbcIntegerTolerance) ; |
---|
2666 | |
---|
2667 | const double *lower = solver_->getColLower() ; |
---|
2668 | const double *upper = solver_->getColUpper() ; |
---|
2669 | const double *solution = solver_->getColSolution() ; |
---|
2670 | const double *reducedCost = solver_->getReducedCost() ; |
---|
2671 | |
---|
2672 | int numberFixed = 0 ; |
---|
2673 | for (int i = 0 ; i < numberIntegers_ ; i++) |
---|
2674 | { int iColumn = integerVariable_[i] ; |
---|
2675 | double djValue = direction*reducedCost[iColumn] ; |
---|
2676 | if (upper[iColumn]-lower[iColumn] > integerTolerance) |
---|
2677 | { if (solution[iColumn] < lower[iColumn]+integerTolerance && djValue > gap) |
---|
2678 | { solver_->setColUpper(iColumn,lower[iColumn]) ; |
---|
2679 | numberFixed++ ; } |
---|
2680 | else |
---|
2681 | if (solution[iColumn] > upper[iColumn]-integerTolerance && -djValue > gap) |
---|
2682 | { solver_->setColLower(iColumn,upper[iColumn]) ; |
---|
2683 | numberFixed++ ; } } } |
---|
2684 | |
---|
2685 | return ; } |
---|
2686 | |
---|
2687 | // Collect coding to replace whichGenerator |
---|
2688 | static int * newWhichGenerator(int numberNow, int numberAfter, |
---|
2689 | int & maximumWhich, int * whichGenerator) |
---|
2690 | { |
---|
2691 | if (numberAfter > maximumWhich) { |
---|
2692 | maximumWhich = CoinMax(maximumWhich*2+100,numberAfter) ; |
---|
2693 | int * temp = new int[2*maximumWhich] ; |
---|
2694 | memcpy(temp,whichGenerator,numberNow*sizeof(int)) ; |
---|
2695 | delete [] whichGenerator ; |
---|
2696 | whichGenerator = temp ; |
---|
2697 | } |
---|
2698 | return whichGenerator; |
---|
2699 | } |
---|
2700 | |
---|
2701 | /** Solve the model using cuts |
---|
2702 | |
---|
2703 | This version takes off redundant cuts from node. |
---|
2704 | Returns true if feasible. |
---|
2705 | |
---|
2706 | \todo |
---|
2707 | Why do I need to resolve the problem? What has been done between the last |
---|
2708 | relaxation and calling solveWithCuts? |
---|
2709 | |
---|
2710 | If numberTries == 0 then user did not want any cuts. |
---|
2711 | */ |
---|
2712 | |
---|
2713 | bool |
---|
2714 | CbcModel::solveWithCuts (OsiCuts &cuts, int numberTries, CbcNode *node, |
---|
2715 | int &numberOldActiveCuts, int &numberNewCuts, |
---|
2716 | int &maximumWhich, int *&whichGenerator) |
---|
2717 | /* |
---|
2718 | Parameters: |
---|
2719 | numberTries: (i) the maximum number of iterations for this round of cut |
---|
2720 | generation; if negative then we don't mind if drop is tiny. |
---|
2721 | |
---|
2722 | cuts: (o) all cuts generated in this round of cut generation |
---|
2723 | whichGenerator: (i/o) whichGenerator[i] is loaded with the index of the |
---|
2724 | generator that produced cuts[i]; reallocated as |
---|
2725 | required |
---|
2726 | numberOldActiveCuts: (o) the number of active cuts at this node from |
---|
2727 | previous rounds of cut generation |
---|
2728 | numberNewCuts: (o) the number of cuts produced in this round of cut |
---|
2729 | generation |
---|
2730 | maximumWhich: (i/o) capacity of whichGenerator; may be updated if |
---|
2731 | whichGenerator grows. |
---|
2732 | |
---|
2733 | node: (i) So we can update dynamic pseudo costs |
---|
2734 | */ |
---|
2735 | |
---|
2736 | |
---|
2737 | { bool feasible = true ; |
---|
2738 | int lastNumberCuts = 0 ; |
---|
2739 | double lastObjective = -1.0e100 ; |
---|
2740 | int violated = 0 ; |
---|
2741 | int numberRowsAtStart = solver_->getNumRows() ; |
---|
2742 | int numberColumns = solver_->getNumCols() ; |
---|
2743 | |
---|
2744 | numberOldActiveCuts = numberRowsAtStart-numberRowsAtContinuous_ ; |
---|
2745 | numberNewCuts = 0 ; |
---|
2746 | |
---|
2747 | bool onOptimalPath = false ; |
---|
2748 | const OsiRowCutDebugger *debugger = NULL; |
---|
2749 | if ((specialOptions_&1)!=0) { |
---|
2750 | /* |
---|
2751 | See OsiRowCutDebugger for details. In a nutshell, make sure that current |
---|
2752 | variable values do not conflict with a known optimal solution. (Obviously |
---|
2753 | this can be fooled when there are multiple solutions.) |
---|
2754 | */ |
---|
2755 | debugger = solver_->getRowCutDebugger() ; |
---|
2756 | if (debugger) |
---|
2757 | onOptimalPath = (debugger->onOptimalPath(*solver_)) ; |
---|
2758 | } |
---|
2759 | /* |
---|
2760 | Resolve the problem. If we've lost feasibility, might as well bail out right |
---|
2761 | after the debug stuff. |
---|
2762 | */ |
---|
2763 | double objectiveValue = solver_->getObjValue()*solver_->getObjSense(); |
---|
2764 | if (node) |
---|
2765 | objectiveValue= node->objectiveValue(); |
---|
2766 | feasible = resolve() ; |
---|
2767 | if (problemFeasibility_->feasible(this,0)<0) { |
---|
2768 | feasible=false; // pretend infeasible |
---|
2769 | } |
---|
2770 | |
---|
2771 | // Update branching information if wanted |
---|
2772 | if(node &&branchingMethod_) |
---|
2773 | branchingMethod_->updateInformation(solver_,node); |
---|
2774 | |
---|
2775 | #ifdef CBC_DEBUG |
---|
2776 | if (feasible) |
---|
2777 | { printf("Obj value %g (%s) %d rows\n",solver_->getObjValue(), |
---|
2778 | (solver_->isProvenOptimal())?"proven":"unproven", |
---|
2779 | solver_->getNumRows()) ; } |
---|
2780 | |
---|
2781 | else |
---|
2782 | { printf("Infeasible %d rows\n",solver_->getNumRows()) ; } |
---|
2783 | #endif |
---|
2784 | if ((specialOptions_&1)!=0) { |
---|
2785 | /* |
---|
2786 | If the RowCutDebugger said we were compatible with the optimal solution, |
---|
2787 | and now we're suddenly infeasible, we might be confused. Then again, we |
---|
2788 | may have fathomed by bound, heading for a rediscovery of an optimal solution. |
---|
2789 | */ |
---|
2790 | if (onOptimalPath && !solver_->isDualObjectiveLimitReached()) { |
---|
2791 | if (!feasible) { |
---|
2792 | solver_->writeMps("infeas"); |
---|
2793 | CoinWarmStartBasis *slack = |
---|
2794 | dynamic_cast<CoinWarmStartBasis *>(solver_->getEmptyWarmStart()) ; |
---|
2795 | solver_->setWarmStart(slack); |
---|
2796 | delete slack ; |
---|
2797 | solver_->setHintParam(OsiDoReducePrint,false,OsiHintDo,0) ; |
---|
2798 | solver_->initialSolve(); |
---|
2799 | } |
---|
2800 | assert(feasible) ; |
---|
2801 | } |
---|
2802 | } |
---|
2803 | |
---|
2804 | if (!feasible) { |
---|
2805 | numberInfeasibleNodes_++; |
---|
2806 | return (false) ; |
---|
2807 | } |
---|
2808 | sumChangeObjective1_ += solver_->getObjValue()*solver_->getObjSense() |
---|
2809 | - objectiveValue ; |
---|
2810 | if ( CoinCpuTime()-dblParam_[CbcStartSeconds] > dblParam_[CbcMaximumSeconds] ) |
---|
2811 | numberTries=0; // exit |
---|
2812 | //if ((numberNodes_%100)==0) |
---|
2813 | //printf("XXa sum obj changed by %g\n",sumChangeObjective1_); |
---|
2814 | objectiveValue = solver_->getObjValue()*solver_->getObjSense(); |
---|
2815 | // Return at once if numberTries zero |
---|
2816 | if (!numberTries) { |
---|
2817 | cuts=OsiCuts(); |
---|
2818 | numberNewCuts=0; |
---|
2819 | return true; |
---|
2820 | } |
---|
2821 | /* |
---|
2822 | Do reduced cost fixing, and then grab the primal solution and bounds vectors. |
---|
2823 | */ |
---|
2824 | reducedCostFix() ; |
---|
2825 | const double *lower = solver_->getColLower() ; |
---|
2826 | const double *upper = solver_->getColUpper() ; |
---|
2827 | const double *solution = solver_->getColSolution() ; |
---|
2828 | /* |
---|
2829 | Set up for at most numberTries rounds of cut generation. If numberTries is |
---|
2830 | negative, we'll ignore the minimumDrop_ cutoff and keep generating cuts for |
---|
2831 | the specified number of rounds. |
---|
2832 | */ |
---|
2833 | double minimumDrop = minimumDrop_ ; |
---|
2834 | if (numberTries<0) |
---|
2835 | { numberTries = -numberTries ; |
---|
2836 | minimumDrop = -1.0 ; } |
---|
2837 | /* |
---|
2838 | Is it time to scan the cuts in order to remove redundant cuts? If so, set |
---|
2839 | up to do it. |
---|
2840 | */ |
---|
2841 | # define SCANCUTS 100 |
---|
2842 | int *countColumnCuts = NULL ; |
---|
2843 | // Always accumulate row cut counts |
---|
2844 | int * countRowCuts =new int[numberCutGenerators_+numberHeuristics_] ; |
---|
2845 | memset(countRowCuts,0, |
---|
2846 | (numberCutGenerators_+numberHeuristics_)*sizeof(int)) ; |
---|
2847 | bool fullScan = false ; |
---|
2848 | if ((numberNodes_%SCANCUTS) == 0) |
---|
2849 | { fullScan = true ; |
---|
2850 | countColumnCuts = new int[numberCutGenerators_+numberHeuristics_] ; |
---|
2851 | memset(countColumnCuts,0, |
---|
2852 | (numberCutGenerators_+numberHeuristics_)*sizeof(int)) ; } |
---|
2853 | |
---|
2854 | double direction = solver_->getObjSense() ; |
---|
2855 | double startObjective = solver_->getObjValue()*direction ; |
---|
2856 | |
---|
2857 | currentPassNumber_ = 0 ; |
---|
2858 | double primalTolerance = 1.0e-7 ; |
---|
2859 | /* |
---|
2860 | Begin cut generation loop. Cuts generated during each iteration are |
---|
2861 | collected in theseCuts. The loop can be divided into four phases: |
---|
2862 | 1) Prep: Fix variables using reduced cost. In the first iteration only, |
---|
2863 | consider scanning globalCuts_ and activating any applicable cuts. |
---|
2864 | 2) Cut Generation: Call each generator and heuristic registered in the |
---|
2865 | generator_ and heuristic_ arrays. Newly generated global cuts are |
---|
2866 | copied to globalCuts_ at this time. |
---|
2867 | 3) Cut Installation and Reoptimisation: Install column and row cuts in |
---|
2868 | the solver. Copy row cuts to cuts (parameter). Reoptimise. |
---|
2869 | 4) Cut Purging: takeOffCuts() removes inactive cuts from the solver, and |
---|
2870 | does the necessary bookkeeping in the model. |
---|
2871 | */ |
---|
2872 | do |
---|
2873 | { currentPassNumber_++ ; |
---|
2874 | numberTries-- ; |
---|
2875 | OsiCuts theseCuts ; |
---|
2876 | /* |
---|
2877 | Scan previously generated global column and row cuts to see if any are |
---|
2878 | useful. |
---|
2879 | I can't see why this code |
---|
2880 | needs its own copy of the primal solution. Removed the dec'l. |
---|
2881 | */ |
---|
2882 | int numberViolated=0; |
---|
2883 | if (currentPassNumber_ == 1 && howOftenGlobalScan_ > 0 && |
---|
2884 | (numberNodes_%howOftenGlobalScan_) == 0) |
---|
2885 | { int numberCuts = globalCuts_.sizeColCuts() ; |
---|
2886 | int i; |
---|
2887 | // possibly extend whichGenerator |
---|
2888 | whichGenerator = newWhichGenerator(numberViolated, numberViolated+numberCuts, |
---|
2889 | maximumWhich, whichGenerator); |
---|
2890 | for ( i = 0 ; i < numberCuts ; i++) |
---|
2891 | { const OsiColCut *thisCut = globalCuts_.colCutPtr(i) ; |
---|
2892 | if (thisCut->violated(solution)>primalTolerance) { |
---|
2893 | printf("Global cut added - violation %g\n", |
---|
2894 | thisCut->violated(solution)) ; |
---|
2895 | whichGenerator[numberViolated++]=-1; |
---|
2896 | theseCuts.insert(*thisCut) ; |
---|
2897 | } |
---|
2898 | } |
---|
2899 | numberCuts = globalCuts_.sizeRowCuts() ; |
---|
2900 | // possibly extend whichGenerator |
---|
2901 | whichGenerator = newWhichGenerator(numberViolated, numberViolated+numberCuts, |
---|
2902 | maximumWhich, whichGenerator); |
---|
2903 | for ( i = 0;i<numberCuts;i++) { |
---|
2904 | const OsiRowCut * thisCut = globalCuts_.rowCutPtr(i) ; |
---|
2905 | if (thisCut->violated(solution)>primalTolerance) { |
---|
2906 | //printf("Global cut added - violation %g\n", |
---|
2907 | // thisCut->violated(solution)) ; |
---|
2908 | whichGenerator[numberViolated++]=-1; |
---|
2909 | theseCuts.insert(*thisCut) ; |
---|
2910 | } |
---|
2911 | } |
---|
2912 | numberGlobalViolations_+=numberViolated; |
---|
2913 | } |
---|
2914 | /* |
---|
2915 | Generate new cuts (global and/or local) and/or apply heuristics. If |
---|
2916 | CglProbing is used, then it should be first as it can fix continuous |
---|
2917 | variables. |
---|
2918 | |
---|
2919 | At present, CglProbing is the only case where generateCuts will return |
---|
2920 | true. generateCuts actually modifies variable bounds in the solver when |
---|
2921 | CglProbing indicates that it can fix a variable. Reoptimisation is required |
---|
2922 | to take full advantage. |
---|
2923 | */ |
---|
2924 | if (nextRowCut_) { |
---|
2925 | // branch was a cut - add it |
---|
2926 | theseCuts.insert(*nextRowCut_); |
---|
2927 | //nextRowCut_->print(); |
---|
2928 | const OsiRowCut * cut=nextRowCut_; |
---|
2929 | const double * solution = solver_->getColSolution(); |
---|
2930 | double lb = cut->lb(); |
---|
2931 | double ub = cut->ub(); |
---|
2932 | int n=cut->row().getNumElements(); |
---|
2933 | const int * column = cut->row().getIndices(); |
---|
2934 | const double * element = cut->row().getElements(); |
---|
2935 | double sum=0.0; |
---|
2936 | for (int i=0;i<n;i++) { |
---|
2937 | int iColumn = column[i]; |
---|
2938 | double value = element[i]; |
---|
2939 | //if (solution[iColumn]>1.0e-7) |
---|
2940 | //printf("value of %d is %g\n",iColumn,solution[iColumn]); |
---|
2941 | sum += value * solution[iColumn]; |
---|
2942 | } |
---|
2943 | delete nextRowCut_; |
---|
2944 | nextRowCut_=NULL; |
---|
2945 | if (handler_->logLevel()>1) |
---|
2946 | printf("applying branch cut, sum is %g, bounds %g %g\n",sum,lb,ub); |
---|
2947 | // possibly extend whichGenerator |
---|
2948 | whichGenerator = newWhichGenerator(numberViolated, numberViolated+1, |
---|
2949 | maximumWhich, whichGenerator); |
---|
2950 | // set whichgenerator (also serves as marker to say don't delete0 |
---|
2951 | whichGenerator[numberViolated++]=-2; |
---|
2952 | } |
---|
2953 | double * newSolution = new double [numberColumns] ; |
---|
2954 | double heuristicValue = getCutoff() ; |
---|
2955 | int found = -1; // no solution found |
---|
2956 | |
---|
2957 | for (int i = 0;i<numberCutGenerators_+numberHeuristics_;i++) { |
---|
2958 | int numberRowCutsBefore = theseCuts.sizeRowCuts() ; |
---|
2959 | int numberColumnCutsBefore = theseCuts.sizeColCuts() ; |
---|
2960 | if (i<numberCutGenerators_) { |
---|
2961 | if (generator_[i]->normal()) { |
---|
2962 | bool mustResolve = |
---|
2963 | generator_[i]->generateCuts(theseCuts,fullScan,node) ; |
---|
2964 | #ifdef CBC_DEBUG |
---|
2965 | { |
---|
2966 | int numberRowCutsAfter = theseCuts.sizeRowCuts() ; |
---|
2967 | int k ; |
---|
2968 | for (k = numberRowCutsBefore;k<numberRowCutsAfter;k++) { |
---|
2969 | OsiRowCut thisCut = theseCuts.rowCut(k) ; |
---|
2970 | /* check size of elements. |
---|
2971 | We can allow smaller but this helps debug generators as it |
---|
2972 | is unsafe to have small elements */ |
---|
2973 | int n=thisCut.row().getNumElements(); |
---|
2974 | const int * column = thisCut.row().getIndices(); |
---|
2975 | const double * element = thisCut.row().getElements(); |
---|
2976 | //assert (n); |
---|
2977 | for (int i=0;i<n;i++) { |
---|
2978 | int iColumn = column[i]; |
---|
2979 | double value = element[i]; |
---|
2980 | assert(fabs(value)>1.0e-12&&fabs(value)<1.0e20); |
---|
2981 | } |
---|
2982 | } |
---|
2983 | } |
---|
2984 | #endif |
---|
2985 | if (mustResolve) { |
---|
2986 | feasible = resolve() ; |
---|
2987 | if ((specialOptions_&1)!=0) { |
---|
2988 | debugger = solver_->getRowCutDebugger() ; |
---|
2989 | if (debugger) |
---|
2990 | onOptimalPath = (debugger->onOptimalPath(*solver_)) ; |
---|
2991 | else |
---|
2992 | onOptimalPath=false; |
---|
2993 | if (onOptimalPath && !solver_->isDualObjectiveLimitReached()) |
---|
2994 | assert(feasible) ; |
---|
2995 | } |
---|
2996 | if (!feasible) |
---|
2997 | break ; |
---|
2998 | } |
---|
2999 | } |
---|
3000 | } else { |
---|
3001 | // see if heuristic will do anything |
---|
3002 | double saveValue = heuristicValue ; |
---|
3003 | int ifSol = |
---|
3004 | heuristic_[i-numberCutGenerators_]->solution(heuristicValue, |
---|
3005 | newSolution, |
---|
3006 | theseCuts) ; |
---|
3007 | if (ifSol>0) { |
---|
3008 | // better solution found |
---|
3009 | found = i ; |
---|
3010 | incrementUsed(newSolution); |
---|
3011 | } else if (ifSol<0) { |
---|
3012 | heuristicValue = saveValue ; |
---|
3013 | } |
---|
3014 | } |
---|
3015 | int numberRowCutsAfter = theseCuts.sizeRowCuts() ; |
---|
3016 | int numberColumnCutsAfter = theseCuts.sizeColCuts() ; |
---|
3017 | |
---|
3018 | if ((specialOptions_&1)!=0) { |
---|
3019 | if (onOptimalPath) { |
---|
3020 | int k ; |
---|
3021 | for (k = numberRowCutsBefore;k<numberRowCutsAfter;k++) { |
---|
3022 | OsiRowCut thisCut = theseCuts.rowCut(k) ; |
---|
3023 | if(debugger->invalidCut(thisCut)) { |
---|
3024 | solver_->writeMps("badCut"); |
---|
3025 | } |
---|
3026 | assert(!debugger->invalidCut(thisCut)) ; |
---|
3027 | } |
---|
3028 | } |
---|
3029 | } |
---|
3030 | |
---|
3031 | /* |
---|
3032 | The cut generator/heuristic has done its thing, and maybe it generated some |
---|
3033 | cuts and/or a new solution. Do a bit of bookkeeping: load |
---|
3034 | whichGenerator[i] with the index of the generator responsible for a cut, |
---|
3035 | and place cuts flagged as global in the global cut pool for the model. |
---|
3036 | |
---|
3037 | lastNumberCuts is the sum of cuts added in previous iterations; it's the |
---|
3038 | offset to the proper starting position in whichGenerator. |
---|
3039 | */ |
---|
3040 | int numberBefore = |
---|
3041 | numberRowCutsBefore+numberColumnCutsBefore+lastNumberCuts ; |
---|
3042 | int numberAfter = |
---|
3043 | numberRowCutsAfter+numberColumnCutsAfter+lastNumberCuts ; |
---|
3044 | // possibly extend whichGenerator |
---|
3045 | whichGenerator = newWhichGenerator(numberBefore, numberAfter, |
---|
3046 | maximumWhich, whichGenerator); |
---|
3047 | int j ; |
---|
3048 | if (fullScan) { |
---|
3049 | // counts |
---|
3050 | countColumnCuts[i] += numberColumnCutsAfter-numberColumnCutsBefore ; |
---|
3051 | } |
---|
3052 | countRowCuts[i] += numberRowCutsAfter-numberRowCutsBefore ; |
---|
3053 | |
---|
3054 | for (j = numberRowCutsBefore;j<numberRowCutsAfter;j++) { |
---|
3055 | whichGenerator[numberBefore++] = i ; |
---|
3056 | const OsiRowCut * thisCut = theseCuts.rowCutPtr(j) ; |
---|
3057 | if (thisCut->lb()>thisCut->ub()) |
---|
3058 | violated=-2; // sub-problem is infeasible |
---|
3059 | if (thisCut->globallyValid()) { |
---|
3060 | // add to global list |
---|
3061 | globalCuts_.insert(*thisCut) ; |
---|
3062 | } |
---|
3063 | } |
---|
3064 | for (j = numberColumnCutsBefore;j<numberColumnCutsAfter;j++) { |
---|
3065 | whichGenerator[numberBefore++] = i ; |
---|
3066 | const OsiColCut * thisCut = theseCuts.colCutPtr(j) ; |
---|
3067 | if (thisCut->globallyValid()) { |
---|
3068 | // add to global list |
---|
3069 | globalCuts_.insert(*thisCut) ; |
---|
3070 | } |
---|
3071 | } |
---|
3072 | } |
---|
3073 | // If at root node and first pass do heuristics without cuts |
---|
3074 | if (!numberNodes_&¤tPassNumber_==1) { |
---|
3075 | // Save number solutions |
---|
3076 | int saveNumberSolutions = numberSolutions_; |
---|
3077 | int saveNumberHeuristicSolutions = numberHeuristicSolutions_; |
---|
3078 | for (int i = 0;i<numberHeuristics_;i++) { |
---|
3079 | // see if heuristic will do anything |
---|
3080 | double saveValue = heuristicValue ; |
---|
3081 | int ifSol = heuristic_[i]->solution(heuristicValue, |
---|
3082 | newSolution); |
---|
3083 | if (ifSol>0) { |
---|
3084 | // better solution found |
---|
3085 | found = i ; |
---|
3086 | incrementUsed(newSolution); |
---|
3087 | // increment number of solutions so other heuristics can test |
---|
3088 | numberSolutions_++; |
---|
3089 | numberHeuristicSolutions_++; |
---|
3090 | } else { |
---|
3091 | heuristicValue = saveValue ; |
---|
3092 | } |
---|
3093 | } |
---|
3094 | // Restore number solutions |
---|
3095 | numberSolutions_ = saveNumberSolutions; |
---|
3096 | numberHeuristicSolutions_ = saveNumberHeuristicSolutions; |
---|
3097 | } |
---|
3098 | /* |
---|
3099 | End of the loop to exercise each generator/heuristic. |
---|
3100 | |
---|
3101 | Did any of the heuristics turn up a new solution? Record it before we free |
---|
3102 | the vector. |
---|
3103 | */ |
---|
3104 | if (found >= 0) { |
---|
3105 | phase_=4; |
---|
3106 | incrementUsed(newSolution); |
---|
3107 | setBestSolution(CBC_ROUNDING,heuristicValue,newSolution) ; |
---|
3108 | lastHeuristic_ = heuristic_[found]; |
---|
3109 | } |
---|
3110 | delete [] newSolution ; |
---|
3111 | |
---|
3112 | #if 0 |
---|
3113 | // switch on to get all cuts printed |
---|
3114 | theseCuts.printCuts() ; |
---|
3115 | #endif |
---|
3116 | int numberColumnCuts = theseCuts.sizeColCuts() ; |
---|
3117 | int numberRowCuts = theseCuts.sizeRowCuts() ; |
---|
3118 | if (violated>=0) |
---|
3119 | violated = numberRowCuts + numberColumnCuts ; |
---|
3120 | /* |
---|
3121 | Apply column cuts (aka bound tightening). This may be partially redundant |
---|
3122 | for column cuts returned by CglProbing, as generateCuts installs bounds |
---|
3123 | from CglProbing when it determines it can fix a variable. |
---|
3124 | |
---|
3125 | TODO: Looks like the use of violated has evolved. The value set above is |
---|
3126 | completely ignored. All that's left is violated == -1 indicates some |
---|
3127 | cut is violated, violated == -2 indicates infeasibility. Only |
---|
3128 | infeasibility warrants exceptional action. |
---|
3129 | |
---|
3130 | TODO: Strikes me that this code will fail to detect infeasibility, because |
---|
3131 | the breaks escape the inner loops but the outer loop keeps going. |
---|
3132 | Infeasibility in an early cut will be overwritten if a later cut is |
---|
3133 | merely violated. |
---|
3134 | */ |
---|
3135 | if (numberColumnCuts) { |
---|
3136 | |
---|
3137 | #ifdef CBC_DEBUG |
---|
3138 | double * oldLower = new double [numberColumns] ; |
---|
3139 | double * oldUpper = new double [numberColumns] ; |
---|
3140 | memcpy(oldLower,lower,numberColumns*sizeof(double)) ; |
---|
3141 | memcpy(oldUpper,upper,numberColumns*sizeof(double)) ; |
---|
3142 | #endif |
---|
3143 | |
---|
3144 | double integerTolerance = getDblParam(CbcIntegerTolerance) ; |
---|
3145 | for (int i = 0;i<numberColumnCuts;i++) { |
---|
3146 | const OsiColCut * thisCut = theseCuts.colCutPtr(i) ; |
---|
3147 | const CoinPackedVector & lbs = thisCut->lbs() ; |
---|
3148 | const CoinPackedVector & ubs = thisCut->ubs() ; |
---|
3149 | int j ; |
---|
3150 | int n ; |
---|
3151 | const int * which ; |
---|
3152 | const double * values ; |
---|
3153 | n = lbs.getNumElements() ; |
---|
3154 | which = lbs.getIndices() ; |
---|
3155 | values = lbs.getElements() ; |
---|
3156 | for (j = 0;j<n;j++) { |
---|
3157 | int iColumn = which[j] ; |
---|
3158 | double value = solution[iColumn] ; |
---|
3159 | #if CBC_DEBUG>1 |
---|
3160 | printf("%d %g %g %g %g\n",iColumn,oldLower[iColumn], |
---|
3161 | solution[iColumn],oldUpper[iColumn],values[j]) ; |
---|
3162 | #endif |
---|
3163 | solver_->setColLower(iColumn,values[j]) ; |
---|
3164 | if (value<values[j]-integerTolerance) |
---|
3165 | violated = -1 ; |
---|
3166 | if (values[j]>upper[iColumn]+integerTolerance) { |
---|
3167 | // infeasible |
---|
3168 | violated = -2 ; |
---|
3169 | break ; |
---|
3170 | } |
---|
3171 | } |
---|
3172 | n = ubs.getNumElements() ; |
---|
3173 | which = ubs.getIndices() ; |
---|
3174 | values = ubs.getElements() ; |
---|
3175 | for (j = 0;j<n;j++) { |
---|
3176 | int iColumn = which[j] ; |
---|
3177 | double value = solution[iColumn] ; |
---|
3178 | #if CBC_DEBUG>1 |
---|
3179 | printf("%d %g %g %g %g\n",iColumn,oldLower[iColumn], |
---|
3180 | solution[iColumn],oldUpper[iColumn],values[j]) ; |
---|
3181 | #endif |
---|
3182 | solver_->setColUpper(iColumn,values[j]) ; |
---|
3183 | if (value>values[j]+integerTolerance) |
---|
3184 | violated = -1 ; |
---|
3185 | if (values[j]<lower[iColumn]-integerTolerance) { |
---|
3186 | // infeasible |
---|
3187 | violated = -2 ; |
---|
3188 | break ; |
---|
3189 | } |
---|
3190 | } |
---|
3191 | } |
---|
3192 | #ifdef CBC_DEBUG |
---|
3193 | delete [] oldLower ; |
---|
3194 | delete [] oldUpper ; |
---|
3195 | #endif |
---|
3196 | } |
---|
3197 | /* |
---|
3198 | End installation of column cuts. The break here escapes the numberTries |
---|
3199 | loop. |
---|
3200 | */ |
---|
3201 | if (violated == -2||!feasible) { |
---|
3202 | // infeasible |
---|
3203 | feasible = false ; |
---|
3204 | violated = -2; |
---|
3205 | if (!numberNodes_) |
---|
3206 | messageHandler()->message(CBC_INFEAS, |
---|
3207 | messages()) |
---|
3208 | << CoinMessageEol ; |
---|
3209 | break ; |
---|
3210 | } |
---|
3211 | /* |
---|
3212 | Now apply the row (constraint) cuts. This is a bit more work because we need |
---|
3213 | to obtain and augment the current basis. |
---|
3214 | |
---|
3215 | TODO: Why do this work, if there are no row cuts? The current basis will do |
---|
3216 | just fine. |
---|
3217 | */ |
---|
3218 | int numberRowsNow = solver_->getNumRows() ; |
---|
3219 | assert(numberRowsNow == numberRowsAtStart+lastNumberCuts) ; |
---|
3220 | int numberToAdd = theseCuts.sizeRowCuts() ; |
---|
3221 | numberNewCuts = lastNumberCuts+numberToAdd ; |
---|
3222 | /* |
---|
3223 | Get a basis by asking the solver for warm start information. Resize it |
---|
3224 | (retaining the basis) so it can accommodate the cuts. |
---|
3225 | */ |
---|
3226 | delete basis_ ; |
---|
3227 | basis_ = dynamic_cast<CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
3228 | assert(basis_ != NULL); // make sure not volume |
---|
3229 | basis_->resize(numberRowsAtStart+numberNewCuts,numberColumns) ; |
---|
3230 | /* |
---|
3231 | Now actually add the row cuts and reoptimise. |
---|
3232 | |
---|
3233 | Install the cuts in the solver using applyRowCuts and |
---|
3234 | augment the basis with the corresponding slack. We also add each row cut to |
---|
3235 | the set of row cuts (cuts.insert()) supplied as a parameter. The new basis |
---|
3236 | must be set with setWarmStart(). |
---|
3237 | |
---|
3238 | TODO: It's not clear to me why we can't separate this into two sections. |
---|
3239 | The first would add the row cuts, and be executed only if row cuts |
---|
3240 | need to be installed. The second would call resolve() and would be |
---|
3241 | executed if either row or column cuts have been installed. |
---|
3242 | |
---|
3243 | TODO: Seems to me the original code could allocate addCuts with size 0, if |
---|
3244 | numberRowCuts was 0 and numberColumnCuts was nonzero. That might |
---|
3245 | explain the memory fault noted in the comment by AJK. Unfortunately, |
---|
3246 | just commenting out the delete[] results in massive memory leaks. Try |
---|
3247 | a revision to separate the row cut case. Why do we need addCuts at |
---|
3248 | all? A typing issue, apparently: OsiCut vs. OsiRowCut. |
---|
3249 | |
---|
3250 | TODO: It looks to me as if numberToAdd and numberRowCuts are identical at |
---|
3251 | this point. Confirm & get rid of one of them. |
---|
3252 | |
---|
3253 | TODO: Any reason why the three loops can't be consolidated? |
---|
3254 | */ |
---|
3255 | if (numberRowCuts > 0 || numberColumnCuts > 0) |
---|
3256 | { if (numberToAdd > 0) |
---|
3257 | { int i ; |
---|
3258 | // Faster to add all at once |
---|
3259 | const OsiRowCut ** addCuts = new const OsiRowCut * [numberToAdd] ; |
---|
3260 | for (i = 0 ; i < numberToAdd ; i++) |
---|
3261 | { addCuts[i] = &theseCuts.rowCut(i) ; } |
---|
3262 | solver_->applyRowCuts(numberToAdd,addCuts) ; |
---|
3263 | // AJK this caused a memory fault on Win32 |
---|
3264 | // may be okay now with ** form |
---|
3265 | delete [] addCuts ; |
---|
3266 | for (i = 0 ; i < numberToAdd ; i++) |
---|
3267 | { cuts.insert(theseCuts.rowCut(i)) ; } |
---|
3268 | for (i = 0 ; i < numberToAdd ; i++) |
---|
3269 | { basis_->setArtifStatus(numberRowsNow+i, |
---|
3270 | CoinWarmStartBasis::basic) ; } |
---|
3271 | if (solver_->setWarmStart(basis_) == false) |
---|
3272 | { throw CoinError("Fail setWarmStart() after cut installation.", |
---|
3273 | "solveWithCuts","CbcModel") ; } } |
---|
3274 | feasible = resolve() ; |
---|
3275 | if ( CoinCpuTime()-dblParam_[CbcStartSeconds] > dblParam_[CbcMaximumSeconds] ) |
---|
3276 | numberTries=0; // exit |
---|
3277 | # ifdef CBC_DEBUG |
---|
3278 | printf("Obj value after cuts %g %d rows\n",solver_->getObjValue(), |
---|
3279 | solver_->getNumRows()) ; |
---|
3280 | if (onOptimalPath && !solver_->isDualObjectiveLimitReached()) |
---|
3281 | assert(feasible) ; |
---|
3282 | # endif |
---|
3283 | } |
---|
3284 | /* |
---|
3285 | No cuts. Cut short the cut generation (numberTries) loop. |
---|
3286 | */ |
---|
3287 | else |
---|
3288 | { numberTries = 0 ; } |
---|
3289 | /* |
---|
3290 | If the problem is still feasible, first, call takeOffCuts() to remove cuts |
---|
3291 | that are now slack. takeOffCuts() will call the solver to reoptimise if |
---|
3292 | that's needed to restore a valid solution. |
---|
3293 | |
---|
3294 | Next, see if we should quit due to diminishing returns: |
---|
3295 | * we've tried three rounds of cut generation and we're getting |
---|
3296 | insufficient improvement in the objective; or |
---|
3297 | * we generated no cuts; or |
---|
3298 | * the solver declared optimality with 0 iterations after we added the |
---|
3299 | cuts generated in this round. |
---|
3300 | If we decide to keep going, prep for the next iteration. |
---|
3301 | |
---|
3302 | It just seems more safe to tell takeOffCuts() to call resolve(), even if |
---|
3303 | we're not continuing cut generation. Otherwise code executed between here |
---|
3304 | and final disposition of the node will need to be careful not to access the |
---|
3305 | lp solution. It can happen that we lose feasibility in takeOffCuts --- |
---|
3306 | numerical jitters when the cutoff bound is epsilon less than the current |
---|
3307 | best, and we're evaluating an alternative optimum. |
---|
3308 | |
---|
3309 | TODO: After successive rounds of code motion, there seems no need to |
---|
3310 | distinguish between the two checks for aborting the cut generation |
---|
3311 | loop. Confirm and clean up. |
---|
3312 | */ |
---|
3313 | if (feasible) |
---|
3314 | { int cutIterations = solver_->getIterationCount() ; |
---|
3315 | if (numberOldActiveCuts+numberNewCuts) { |
---|
3316 | takeOffCuts(cuts,whichGenerator,numberOldActiveCuts, |
---|
3317 | numberNewCuts,resolveAfterTakeOffCuts_) ; |
---|
3318 | if (solver_->isDualObjectiveLimitReached()&&resolveAfterTakeOffCuts_) |
---|
3319 | { feasible = false ; |
---|
3320 | # ifdef CBC_DEBUG |
---|
3321 | double z = solver_->getObjValue() ; |
---|
3322 | double cut = getCutoff() ; |
---|
3323 | printf("Lost feasibility by %g in takeOffCuts; z = %g, cutoff = %g\n", |
---|
3324 | z-cut,z,cut) ; |
---|
3325 | # endif |
---|
3326 | } |
---|
3327 | } |
---|
3328 | if (feasible) |
---|
3329 | { numberRowsAtStart = numberOldActiveCuts+numberRowsAtContinuous_ ; |
---|
3330 | lastNumberCuts = numberNewCuts ; |
---|
3331 | if (direction*solver_->getObjValue() < lastObjective+minimumDrop && |
---|
3332 | currentPassNumber_ >= 3) |
---|
3333 | { numberTries = 0 ; } |
---|
3334 | if (numberRowCuts+numberColumnCuts == 0 || cutIterations == 0) |
---|
3335 | { break ; } |
---|
3336 | if (numberTries > 0) |
---|
3337 | { reducedCostFix() ; |
---|
3338 | lastObjective = direction*solver_->getObjValue() ; |
---|
3339 | lower = solver_->getColLower() ; |
---|
3340 | upper = solver_->getColUpper() ; |
---|
3341 | solution = solver_->getColSolution() ; } } } |
---|
3342 | /* |
---|
3343 | We've lost feasibility --- this node won't be referencing the cuts we've |
---|
3344 | been collecting, so decrement the reference counts. |
---|
3345 | |
---|
3346 | TODO: Presumably this is in preparation for backtracking. Seems like it |
---|
3347 | should be the `else' off the previous `if'. |
---|
3348 | */ |
---|
3349 | if (!feasible) |
---|
3350 | { int i ; |
---|
3351 | for (i = 0;i<currentNumberCuts_;i++) { |
---|
3352 | // take off node |
---|
3353 | if (addedCuts_[i]) { |
---|
3354 | if (!addedCuts_[i]->decrement()) |
---|
3355 | delete addedCuts_[i] ; |
---|
3356 | addedCuts_[i] = NULL ; |
---|
3357 | } |
---|
3358 | } |
---|
3359 | numberTries = 0 ; |
---|
3360 | } |
---|
3361 | } while (numberTries>0) ; |
---|
3362 | // Reduced cost fix at end |
---|
3363 | //reducedCostFix(); |
---|
3364 | // If at root node do heuristics |
---|
3365 | if (!numberNodes_) { |
---|
3366 | // mark so heuristics can tell |
---|
3367 | int savePass=currentPassNumber_; |
---|
3368 | currentPassNumber_=999999; |
---|
3369 | double * newSolution = new double [numberColumns] ; |
---|
3370 | double heuristicValue = getCutoff() ; |
---|
3371 | int found = -1; // no solution found |
---|
3372 | for (int i = 0;i<numberHeuristics_;i++) { |
---|
3373 | // see if heuristic will do anything |
---|
3374 | double saveValue = heuristicValue ; |
---|
3375 | int ifSol = heuristic_[i]->solution(heuristicValue, |
---|
3376 | newSolution); |
---|
3377 | if (ifSol>0) { |
---|
3378 | // better solution found |
---|
3379 | found = i ; |
---|
3380 | incrementUsed(newSolution); |
---|
3381 | } else { |
---|
3382 | heuristicValue = saveValue ; |
---|
3383 | } |
---|
3384 | } |
---|
3385 | currentPassNumber_=savePass; |
---|
3386 | if (found >= 0) { |
---|
3387 | phase_=4; |
---|
3388 | incrementUsed(newSolution); |
---|
3389 | setBestSolution(CBC_ROUNDING,heuristicValue,newSolution) ; |
---|
3390 | lastHeuristic_ = heuristic_[found]; |
---|
3391 | } |
---|
3392 | delete [] newSolution ; |
---|
3393 | } |
---|
3394 | // Up change due to cuts |
---|
3395 | if (feasible) |
---|
3396 | sumChangeObjective2_ += solver_->getObjValue()*solver_->getObjSense() |
---|
3397 | - objectiveValue ; |
---|
3398 | //if ((numberNodes_%100)==0) |
---|
3399 | //printf("XXb sum obj changed by %g\n",sumChangeObjective2_); |
---|
3400 | /* |
---|
3401 | End of cut generation loop. |
---|
3402 | |
---|
3403 | Now, consider if we want to disable or adjust the frequency of use for any |
---|
3404 | of the cut generators. If the client specified a positive number for |
---|
3405 | howOften, it will never change. If the original value was negative, it'll |
---|
3406 | be converted to 1000000+|howOften|, and this value will be adjusted each |
---|
3407 | time fullScan is true. Actual cut generation is performed every |
---|
3408 | howOften%1000000 nodes; the 1000000 offset is just a convenient way to |
---|
3409 | specify that the frequency is adjustable. |
---|
3410 | |
---|
3411 | During cut generation, we recorded the number of cuts produced by each |
---|
3412 | generator for this node. For all cuts, whichGenerator records the generator |
---|
3413 | that produced a cut. |
---|
3414 | |
---|
3415 | TODO: All this should probably be hidden in a method of the CbcCutGenerator |
---|
3416 | class. |
---|
3417 | */ |
---|
3418 | if (fullScan&&numberCutGenerators_) { |
---|
3419 | /* If cuts just at root node then it will probably be faster to |
---|
3420 | update matrix and leave all in */ |
---|
3421 | bool willBeCutsInTree=false; |
---|
3422 | // Root node or every so often - see what to turn off |
---|
3423 | int i ; |
---|
3424 | double thisObjective = solver_->getObjValue()*direction ; |
---|
3425 | double totalCuts = 0.0 ; |
---|
3426 | for (i = 0;i<numberCutGenerators_;i++) |
---|
3427 | totalCuts += countRowCuts[i] + 5.0*countColumnCuts[i] ; |
---|
3428 | if (!numberNodes_) |
---|
3429 | handler_->message(CBC_ROOT,messages_) |
---|
3430 | <<numberNewCuts |
---|
3431 | <<startObjective<<thisObjective |
---|
3432 | <<currentPassNumber_ |
---|
3433 | <<CoinMessageEol ; |
---|
3434 | int * count = new int[numberCutGenerators_] ; |
---|
3435 | memset(count,0,numberCutGenerators_*sizeof(int)) ; |
---|
3436 | for (i = 0;i<numberNewCuts;i++) { |
---|
3437 | int iGenerator = whichGenerator[i]; |
---|
3438 | if (iGenerator>=0) |
---|
3439 | count[iGenerator]++ ; |
---|
3440 | } |
---|
3441 | double small = (0.5* totalCuts) / |
---|
3442 | ((double) numberCutGenerators_) ; |
---|
3443 | for (i = 0;i<numberCutGenerators_;i++) { |
---|
3444 | int howOften = generator_[i]->howOften() ; |
---|
3445 | if (howOften<-99) |
---|
3446 | continue ; |
---|
3447 | if (howOften<0||howOften >= 1000000) { |
---|
3448 | // If small number switch mostly off |
---|
3449 | double thisCuts = countRowCuts[i] + 5.0*countColumnCuts[i] ; |
---|
3450 | if (!thisCuts||howOften == -99) { |
---|
3451 | if (howOften == -99) |
---|
3452 | howOften = -100 ; |
---|
3453 | else |
---|
3454 | howOften = 1000000+SCANCUTS; // wait until next time |
---|
3455 | } else if (thisCuts<small) { |
---|
3456 | int k = (int) sqrt(small/thisCuts) ; |
---|
3457 | howOften = k+1000000 ; |
---|
3458 | } else { |
---|
3459 | howOften = 1+1000000 ; |
---|
3460 | } |
---|
3461 | // If cuts useless switch off |
---|
3462 | if (numberNodes_>=10&&sumChangeObjective1_>1.0e2*(sumChangeObjective2_+1.0e-12)) { |
---|
3463 | howOften = 1000000+SCANCUTS; // wait until next time |
---|
3464 | //printf("switch off cut %d due to lack of use\n",i); |
---|
3465 | } |
---|
3466 | } |
---|
3467 | if (howOften>=0&&generator_[i]->generator()->mayGenerateRowCutsInTree()) |
---|
3468 | willBeCutsInTree=true; |
---|
3469 | |
---|
3470 | generator_[i]->setHowOften(howOften) ; |
---|
3471 | if (howOften>=1000000&&howOften<2000000&&0) { |
---|
3472 | // Go to depth |
---|
3473 | int bias=1; |
---|
3474 | if (howOften==1+1000000) |
---|
3475 | generator_[i]->setWhatDepth(bias+1); |
---|
3476 | else if (howOften<=10+1000000) |
---|
3477 | generator_[i]->setWhatDepth(bias+2); |
---|
3478 | else |
---|
3479 | generator_[i]->setWhatDepth(bias+1000); |
---|
3480 | } |
---|
3481 | int newFrequency = generator_[i]->howOften()%1000000 ; |
---|
3482 | // increment cut counts |
---|
3483 | generator_[i]->incrementNumberCutsInTotal(countRowCuts[i]); |
---|
3484 | generator_[i]->incrementNumberCutsActive(count[i]); |
---|
3485 | if (handler_->logLevel()>1||!numberNodes_) { |
---|
3486 | handler_->message(CBC_GENERATOR,messages_) |
---|
3487 | <<i |
---|
3488 | <<generator_[i]->cutGeneratorName() |
---|
3489 | <<countRowCuts[i]<<count[i] |
---|
3490 | <<countColumnCuts[i]; |
---|
3491 | handler_->printing(!numberNodes_&&generator_[i]->timing()) |
---|
3492 | <<generator_[i]->timeInCutGenerator(); |
---|
3493 | handler_->message() |
---|
3494 | <<newFrequency |
---|
3495 | <<CoinMessageEol ; |
---|
3496 | } |
---|
3497 | } |
---|
3498 | delete [] count ; |
---|
3499 | if( !numberNodes_) { |
---|
3500 | if( !willBeCutsInTree) { |
---|
3501 | // Take off cuts |
---|
3502 | cuts = OsiCuts(); |
---|
3503 | numberNewCuts=0; |
---|
3504 | // update size of problem |
---|
3505 | numberRowsAtContinuous_ = solver_->getNumRows() ; |
---|
3506 | #ifdef COIN_USE_CLP |
---|
3507 | OsiClpSolverInterface * clpSolver |
---|
3508 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
3509 | if (clpSolver) { |
---|
3510 | // Maybe solver might like to know only column bounds will change |
---|
3511 | //int options = clpSolver->specialOptions(); |
---|
3512 | //clpSolver->setSpecialOptions(options|128); |
---|
3513 | clpSolver->synchronizeModel(); |
---|
3514 | } |
---|
3515 | #endif |
---|
3516 | } else { |
---|
3517 | #ifdef COIN_USE_CLP |
---|
3518 | OsiClpSolverInterface * clpSolver |
---|
3519 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
3520 | if (clpSolver) { |
---|
3521 | // make sure factorization can't carry over |
---|
3522 | int options = clpSolver->specialOptions(); |
---|
3523 | clpSolver->setSpecialOptions(options&(~8)); |
---|
3524 | } |
---|
3525 | #endif |
---|
3526 | } |
---|
3527 | } |
---|
3528 | } else if (numberCutGenerators_) { |
---|
3529 | int i; |
---|
3530 | // add to counts anyway |
---|
3531 | for (i = 0;i<numberCutGenerators_;i++) |
---|
3532 | generator_[i]->incrementNumberCutsInTotal(countRowCuts[i]); |
---|
3533 | // What if not feasible as cuts may have helped |
---|
3534 | if (feasible) { |
---|
3535 | for (i = 0;i<numberNewCuts;i++) { |
---|
3536 | int iGenerator = whichGenerator[i]; |
---|
3537 | if (iGenerator>=0) |
---|
3538 | generator_[iGenerator]->incrementNumberCutsActive(); |
---|
3539 | } |
---|
3540 | } |
---|
3541 | } |
---|
3542 | |
---|
3543 | delete [] countRowCuts ; |
---|
3544 | delete [] countColumnCuts ; |
---|
3545 | |
---|
3546 | |
---|
3547 | #ifdef CHECK_CUT_COUNTS |
---|
3548 | if (feasible) |
---|
3549 | { delete basis_ ; |
---|
3550 | basis_ = dynamic_cast<CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
3551 | printf("solveWithCuts: Number of rows at end (only active cuts) %d\n", |
---|
3552 | numberRowsAtContinuous_+numberNewCuts+numberOldActiveCuts) ; |
---|
3553 | basis_->print() ; } |
---|
3554 | #endif |
---|
3555 | #ifdef CBC_DEBUG |
---|
3556 | if (onOptimalPath && !solver_->isDualObjectiveLimitReached()) |
---|
3557 | assert(feasible) ; |
---|
3558 | #endif |
---|
3559 | |
---|
3560 | return feasible ; } |
---|
3561 | |
---|
3562 | |
---|
3563 | /* |
---|
3564 | Remove slack cuts. We obtain a basis and scan it. Cuts with basic slacks |
---|
3565 | are purged. If any cuts are purged, resolve() is called to restore the |
---|
3566 | solution held in the solver. If resolve() pivots, there's the possibility |
---|
3567 | that a slack may be pivoted in (trust me :-), so the process iterates. |
---|
3568 | Setting allowResolve to false will suppress reoptimisation (but see note |
---|
3569 | below). |
---|
3570 | |
---|
3571 | At the level of the solver's constraint system, loose cuts are really |
---|
3572 | deleted. There's an implicit assumption that deleteRows will also update |
---|
3573 | the active basis in the solver. |
---|
3574 | |
---|
3575 | At the level of nodes and models, it's more complicated. |
---|
3576 | |
---|
3577 | New cuts exist only in the collection of cuts passed as a parameter. They |
---|
3578 | are deleted from the collection and that's the end of them. |
---|
3579 | |
---|
3580 | Older cuts have made it into addedCuts_. Two separate actions are needed. |
---|
3581 | The reference count for the CbcCountRowCut object is decremented. If this |
---|
3582 | count falls to 0, the node which owns the cut is located, the reference to |
---|
3583 | the cut is removed, and then the cut object is destroyed (courtesy of the |
---|
3584 | CbcCountRowCut destructor). We also need to set the addedCuts_ entry to |
---|
3585 | NULL. This is important so that when it comes time to generate basis edits |
---|
3586 | we can tell this cut was dropped from the basis during processing of the |
---|
3587 | node. |
---|
3588 | |
---|
3589 | NOTE: In general, it's necessary to call resolve() after purging slack |
---|
3590 | cuts. Deleting constraints constitutes a change in the problem, and |
---|
3591 | an OSI is not required to maintain a valid solution when the problem |
---|
3592 | is changed. But ... it's really useful to maintain the active basis, |
---|
3593 | and the OSI is supposed to do that. (Yes, it's splitting hairs.) In |
---|
3594 | some places, it's possible to know that the solution will never be |
---|
3595 | consulted after this call, only the basis. (E.g., this routine is |
---|
3596 | called as a last act before generating info to place the node in the |
---|
3597 | live set.) For such use, set allowResolve to false. |
---|
3598 | |
---|
3599 | TODO: No real harm would be done if we just ignored the rare occasion when |
---|
3600 | the call to resolve() pivoted a slack back into the basis. It's a |
---|
3601 | minor inefficiency, at worst. But it does break assertions which |
---|
3602 | check that there are no loose cuts in the basis. It might be better |
---|
3603 | to remove the assertions. |
---|
3604 | */ |
---|
3605 | |
---|
3606 | void |
---|
3607 | CbcModel::takeOffCuts (OsiCuts &newCuts, int *whichGenerator, |
---|
3608 | int &numberOldActiveCuts, int &numberNewCuts, |
---|
3609 | bool allowResolve) |
---|
3610 | |
---|
3611 | { // int resolveIterations = 0 ; |
---|
3612 | int firstOldCut = numberRowsAtContinuous_ ; |
---|
3613 | int totalNumberCuts = numberNewCuts+numberOldActiveCuts ; |
---|
3614 | int *solverCutIndices = new int[totalNumberCuts] ; |
---|
3615 | int *newCutIndices = new int[numberNewCuts] ; |
---|
3616 | const CoinWarmStartBasis* ws ; |
---|
3617 | CoinWarmStartBasis::Status status ; |
---|
3618 | bool needPurge = true ; |
---|
3619 | /* |
---|
3620 | The outer loop allows repetition of purge in the event that reoptimisation |
---|
3621 | changes the basis. To start an iteration, clear the deletion counts and grab |
---|
3622 | the current basis. |
---|
3623 | */ |
---|
3624 | while (needPurge) |
---|
3625 | { int numberNewToDelete = 0 ; |
---|
3626 | int numberOldToDelete = 0 ; |
---|
3627 | int i ; |
---|
3628 | ws = dynamic_cast<const CoinWarmStartBasis*>(solver_->getWarmStart()) ; |
---|
3629 | /* |
---|
3630 | Scan the basis entries of the old cuts generated prior to this round of cut |
---|
3631 | generation. Loose cuts are `removed' by decrementing their reference count |
---|
3632 | and setting the addedCuts_ entry to NULL. (If the reference count falls to |
---|
3633 | 0, they're really deleted. See CbcModel and CbcCountRowCut doc'n for |
---|
3634 | principles of cut handling.) |
---|
3635 | */ |
---|
3636 | int oldCutIndex = 0 ; |
---|
3637 | for (i = 0 ; i < numberOldActiveCuts ; i++) |
---|
3638 | { status = ws->getArtifStatus(i+firstOldCut) ; |
---|
3639 | while (!addedCuts_[oldCutIndex]) oldCutIndex++ ; |
---|
3640 | assert(oldCutIndex < currentNumberCuts_) ; |
---|
3641 | // always leave if from nextRowCut_ |
---|
3642 | if (status == CoinWarmStartBasis::basic&& |
---|
3643 | addedCuts_[oldCutIndex]->effectiveness()!=COIN_DBL_MAX) |
---|
3644 | { solverCutIndices[numberOldToDelete++] = i+firstOldCut ; |
---|
3645 | if (addedCuts_[oldCutIndex]->decrement() == 0) |
---|
3646 | delete addedCuts_[oldCutIndex] ; |
---|
3647 | addedCuts_[oldCutIndex] = NULL ; |
---|
3648 | oldCutIndex++ ; } |
---|
3649 | else |
---|
3650 | { oldCutIndex++ ; } } |
---|
3651 | /* |
---|
3652 | Scan the basis entries of the new cuts generated with this round of cut |
---|
3653 | generation. At this point, newCuts is the only record of the new cuts, so |
---|
3654 | when we delete loose cuts from newCuts, they're really gone. newCuts is a |
---|
3655 | vector, so it's most efficient to compress it (eraseRowCut) from back to |
---|
3656 | front. |
---|
3657 | */ |
---|
3658 | int firstNewCut = firstOldCut+numberOldActiveCuts ; |
---|
3659 | int k = 0 ; |
---|
3660 | for (i = 0 ; i < numberNewCuts ; i++) |
---|
3661 | { status = ws->getArtifStatus(i+firstNewCut) ; |
---|
3662 | if (status == CoinWarmStartBasis::basic&&whichGenerator[i]!=-2) |
---|
3663 | { solverCutIndices[numberNewToDelete+numberOldToDelete] = i+firstNewCut ; |
---|
3664 | newCutIndices[numberNewToDelete++] = i ; } |
---|
3665 | else |
---|
3666 | { // save which generator did it |
---|
3667 | whichGenerator[k++] = whichGenerator[i] ; } } |
---|
3668 | delete ws ; |
---|
3669 | for (i = numberNewToDelete-1 ; i >= 0 ; i--) |
---|
3670 | { int iCut = newCutIndices[i] ; |
---|
3671 | newCuts.eraseRowCut(iCut) ; } |
---|
3672 | /* |
---|
3673 | Did we delete anything? If so, delete the cuts from the constraint system |
---|
3674 | held in the solver and reoptimise unless we're forbidden to do so. If the |
---|
3675 | call to resolve() results in pivots, there's the possibility we again have |
---|
3676 | basic slacks. Repeat the purging loop. |
---|
3677 | */ |
---|
3678 | if (numberNewToDelete+numberOldToDelete > 0) |
---|
3679 | { solver_->deleteRows(numberNewToDelete+numberOldToDelete, |
---|
3680 | solverCutIndices) ; |
---|
3681 | numberNewCuts -= numberNewToDelete ; |
---|
3682 | numberOldActiveCuts -= numberOldToDelete ; |
---|
3683 | # ifdef CBC_DEBUG |
---|
3684 | printf("takeOffCuts: purged %d+%d cuts\n", numberOldToDelete, |
---|
3685 | numberNewToDelete ); |
---|
3686 | # endif |
---|
3687 | if (allowResolve) |
---|
3688 | { |
---|
3689 | phase_=3; |
---|
3690 | // can do quick optimality check |
---|
3691 | int easy=2; |
---|
3692 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,&easy) ; |
---|
3693 | solver_->resolve() ; |
---|
3694 | solver_->setHintParam(OsiDoInBranchAndCut,true,OsiHintDo,NULL) ; |
---|
3695 | if (solver_->getIterationCount() == 0) |
---|
3696 | { needPurge = false ; } |
---|
3697 | # ifdef CBC_DEBUG |
---|
3698 | else |
---|
3699 | { printf( "Repeating purging loop. %d iters.\n", |
---|
3700 | solver_->getIterationCount()); |
---|
3701 | # endif |
---|
3702 | } |
---|
3703 | else |
---|
3704 | { needPurge = false ; } } |
---|
3705 | else |
---|
3706 | { needPurge = false ; } } |
---|
3707 | /* |
---|
3708 | Clean up and return. |
---|
3709 | */ |
---|
3710 | delete [] solverCutIndices ; |
---|
3711 | delete [] newCutIndices ; |
---|
3712 | } |
---|
3713 | |
---|
3714 | |
---|
3715 | |
---|
3716 | bool |
---|
3717 | CbcModel::resolve() |
---|
3718 | { |
---|
3719 | // We may have deliberately added in violated cuts - check to avoid message |
---|
3720 | int iRow; |
---|
3721 | int numberRows = solver_->getNumRows(); |
---|
3722 | const double * rowLower = solver_->getRowLower(); |
---|
3723 | const double * rowUpper = solver_->getRowUpper(); |
---|
3724 | bool feasible=true; |
---|
3725 | for (iRow= numberRowsAtContinuous_;iRow<numberRows;iRow++) { |
---|
3726 | if (rowLower[iRow]>rowUpper[iRow]+1.0e-8) |
---|
3727 | feasible=false; |
---|
3728 | } |
---|
3729 | // Can't happen if strong branching as would have been found before |
---|
3730 | if (!numberStrong_&&numberObjects_>numberIntegers_) { |
---|
3731 | int iColumn; |
---|
3732 | int numberColumns = solver_->getNumCols(); |
---|
3733 | const double * columnLower = solver_->getColLower(); |
---|
3734 | const double * columnUpper = solver_->getColUpper(); |
---|
3735 | for (iColumn= 0;iColumn<numberColumns;iColumn++) { |
---|
3736 | if (columnLower[iColumn]>columnUpper[iColumn]+1.0e-5) |
---|
3737 | feasible=false; |
---|
3738 | } |
---|
3739 | } |
---|
3740 | /* |
---|
3741 | Reoptimize. Consider the possibility that we should fathom on bounds. But be |
---|
3742 | careful --- where the objective takes on integral values, we may want to keep |
---|
3743 | a solution where the objective is right on the cutoff. |
---|
3744 | */ |
---|
3745 | if (feasible) |
---|
3746 | { |
---|
3747 | solver_->resolve() ; |
---|
3748 | numberIterations_ += solver_->getIterationCount() ; |
---|
3749 | feasible = (solver_->isProvenOptimal() && |
---|
3750 | !solver_->isDualObjectiveLimitReached()) ; } |
---|
3751 | if (!feasible&& continuousObjective_ <-1.0e30) { |
---|
3752 | // at root node - double double check |
---|
3753 | bool saveTakeHint; |
---|
3754 | OsiHintStrength saveStrength; |
---|
3755 | solver_->getHintParam(OsiDoDualInResolve,saveTakeHint,saveStrength); |
---|
3756 | if (saveTakeHint||saveStrength==OsiHintIgnore) { |
---|
3757 | solver_->setHintParam(OsiDoDualInResolve,false,OsiHintDo) ; |
---|
3758 | solver_->resolve(); |
---|
3759 | solver_->setHintParam(OsiDoDualInResolve,saveTakeHint,saveStrength); |
---|
3760 | numberIterations_ += solver_->getIterationCount() ; |
---|
3761 | feasible = solver_->isProvenOptimal(); |
---|
3762 | } |
---|
3763 | } |
---|
3764 | return feasible ; } |
---|
3765 | |
---|
3766 | |
---|
3767 | /* Set up objects. Only do ones whose length is in range. |
---|
3768 | If makeEquality true then a new model may be returned if |
---|
3769 | modifications had to be made, otherwise "this" is returned. |
---|
3770 | |
---|
3771 | Could use Probing at continuous to extend objects |
---|
3772 | */ |
---|
3773 | CbcModel * |
---|
3774 | CbcModel::findCliques(bool makeEquality, |
---|
3775 | int atLeastThisMany, int lessThanThis, int defaultValue) |
---|
3776 | { |
---|
3777 | // No objects are allowed to exist |
---|
3778 | assert(numberObjects_==numberIntegers_||!numberObjects_); |
---|
3779 | CoinPackedMatrix matrixByRow(*solver_->getMatrixByRow()); |
---|
3780 | int numberRows = solver_->getNumRows(); |
---|
3781 | int numberColumns = solver_->getNumCols(); |
---|
3782 | |
---|
3783 | // We may want to add columns |
---|
3784 | int numberSlacks=0; |
---|
3785 | int * rows = new int[numberRows]; |
---|
3786 | double * element =new double[numberRows]; |
---|
3787 | |
---|
3788 | int iRow; |
---|
3789 | |
---|
3790 | findIntegers(true); |
---|
3791 | numberObjects_=numberIntegers_; |
---|
3792 | |
---|
3793 | int numberCliques=0; |
---|
3794 | CbcObject ** object = new CbcObject * [numberRows]; |
---|
3795 | int * which = new int[numberIntegers_]; |
---|
3796 | char * type = new char[numberIntegers_]; |
---|
3797 | int * lookup = new int[numberColumns]; |
---|
3798 | int i; |
---|
3799 | for (i=0;i<numberColumns;i++) |
---|
3800 | lookup[i]=-1; |
---|
3801 | for (i=0;i<numberIntegers_;i++) |
---|
3802 | lookup[integerVariable_[i]]=i; |
---|
3803 | |
---|
3804 | // Statistics |
---|
3805 | int totalP1=0,totalM1=0; |
---|
3806 | int numberBig=0,totalBig=0; |
---|
3807 | int numberFixed=0; |
---|
3808 | |
---|
3809 | // Row copy |
---|
3810 | const double * elementByRow = matrixByRow.getElements(); |
---|
3811 | const int * column = matrixByRow.getIndices(); |
---|
3812 | const CoinBigIndex * rowStart = matrixByRow.getVectorStarts(); |
---|
3813 | const int * rowLength = matrixByRow.getVectorLengths(); |
---|
3814 | |
---|
3815 | // Column lengths for slacks |
---|
3816 | const int * columnLength = solver_->getMatrixByCol()->getVectorLengths(); |
---|
3817 | |
---|
3818 | const double * lower = getColLower(); |
---|
3819 | const double * upper = getColUpper(); |
---|
3820 | const double * rowLower = getRowLower(); |
---|
3821 | const double * rowUpper = getRowUpper(); |
---|
3822 | |
---|
3823 | for (iRow=0;iRow<numberRows;iRow++) { |
---|
3824 | int numberP1=0, numberM1=0; |
---|
3825 | int j; |
---|
3826 | double upperValue=rowUpper[iRow]; |
---|
3827 | double lowerValue=rowLower[iRow]; |
---|
3828 | bool good=true; |
---|
3829 | int slack = -1; |
---|
3830 | for (j=rowStart[iRow];j<rowStart[iRow]+rowLength[iRow];j++) { |
---|
3831 | int iColumn = column[j]; |
---|
3832 | int iInteger=lookup[iColumn]; |
---|
3833 | if (upper[iColumn]-lower[iColumn]<1.0e-8) { |
---|
3834 | // fixed |
---|
3835 | upperValue -= lower[iColumn]*elementByRow[j]; |
---|
3836 | lowerValue -= lower[iColumn]*elementByRow[j]; |
---|
3837 | continue; |
---|
3838 | } else if (upper[iColumn]!=1.0||lower[iColumn]!=0.0) { |
---|
3839 | good = false; |
---|
3840 | break; |
---|
3841 | } else if (iInteger<0) { |
---|
3842 | good = false; |
---|
3843 | break; |
---|
3844 | } else { |
---|
3845 | if (columnLength[iColumn]==1) |
---|
3846 | slack = iInteger; |
---|
3847 | } |
---|
3848 | if (fabs(elementByRow[j])!=1.0) { |
---|
3849 | good=false; |
---|
3850 | break; |
---|
3851 | } else if (elementByRow[j]>0.0) { |
---|
3852 | which[numberP1++]=iInteger; |
---|
3853 | } else { |
---|
3854 | numberM1++; |
---|
3855 | which[numberIntegers_-numberM1]=iInteger; |
---|
3856 | } |
---|
3857 | } |
---|
3858 | int iUpper = (int) floor(upperValue+1.0e-5); |
---|
3859 | int iLower = (int) ceil(lowerValue-1.0e-5); |
---|
3860 | int state=0; |
---|
3861 | if (upperValue<1.0e6) { |
---|
3862 | if (iUpper==1-numberM1) |
---|
3863 | state=1; |
---|
3864 | else if (iUpper==-numberM1) |
---|
3865 | state=2; |
---|
3866 | else if (iUpper<-numberM1) |
---|
3867 | state=3; |
---|
3868 | } |
---|
3869 | if (!state&&lowerValue>-1.0e6) { |
---|
3870 | if (-iLower==1-numberP1) |
---|
3871 | state=-1; |
---|
3872 | else if (-iLower==-numberP1) |
---|
3873 | state=-2; |
---|
3874 | else if (-iLower<-numberP1) |
---|
3875 | state=-3; |
---|
3876 | } |
---|
3877 | if (good&&state) { |
---|
3878 | if (abs(state)==3) { |
---|
3879 | // infeasible |
---|
3880 | numberObjects_=-1; |
---|
3881 | break; |
---|
3882 | } else if (abs(state)==2) { |
---|
3883 | // we can fix all |
---|
3884 | numberFixed += numberP1+numberM1; |
---|
3885 | if (state>0) { |
---|
3886 | // fix all +1 at 0, -1 at 1 |
---|
3887 | for (i=0;i<numberP1;i++) |
---|
3888 | solver_->setColUpper(integerVariable_[which[i]],0.0); |
---|
3889 | for (i=0;i<numberM1;i++) |
---|
3890 | solver_->setColLower(integerVariable_[which[numberIntegers_-i-1]], |
---|
3891 | 1.0); |
---|
3892 | } else { |
---|
3893 | // fix all +1 at 1, -1 at 0 |
---|
3894 | for (i=0;i<numberP1;i++) |
---|
3895 | solver_->setColLower(integerVariable_[which[i]],1.0); |
---|
3896 | for (i=0;i<numberM1;i++) |
---|
3897 | solver_->setColUpper(integerVariable_[which[numberIntegers_-i-1]], |
---|
3898 | 0.0); |
---|
3899 | } |
---|
3900 | } else { |
---|
3901 | int length = numberP1+numberM1; |
---|
3902 | if (length >= atLeastThisMany&&length<lessThanThis) { |
---|
3903 | // create object |
---|
3904 | bool addOne=false; |
---|
3905 | int objectType; |
---|
3906 | if (iLower==iUpper) { |
---|
3907 | objectType=1; |
---|
3908 | } else { |
---|
3909 | if (makeEquality) { |
---|
3910 | objectType=1; |
---|
3911 | element[numberSlacks]=state; |
---|
3912 | rows[numberSlacks++]=iRow; |
---|
3913 | addOne=true; |
---|
3914 | } else { |
---|
3915 | objectType=0; |
---|
3916 | } |
---|
3917 | } |
---|
3918 | if (state>0) { |
---|
3919 | totalP1 += numberP1; |
---|
3920 | totalM1 += numberM1; |
---|
3921 | for (i=0;i<numberP1;i++) |
---|
3922 | type[i]=1; |
---|
3923 | for (i=0;i<numberM1;i++) { |
---|
3924 | which[numberP1]=which[numberIntegers_-i-1]; |
---|
3925 | type[numberP1++]=0; |
---|
3926 | } |
---|
3927 | } else { |
---|
3928 | totalP1 += numberM1; |
---|
3929 | totalM1 += numberP1; |
---|
3930 | for (i=0;i<numberP1;i++) |
---|
3931 | type[i]=0; |
---|
3932 | for (i=0;i<numberM1;i++) { |
---|
3933 | which[numberP1]=which[numberIntegers_-i-1]; |
---|
3934 | type[numberP1++]=1; |
---|
3935 | } |
---|
3936 | } |
---|
3937 | if (addOne) { |
---|
3938 | // add in slack |
---|
3939 | which[numberP1]=numberIntegers_+numberSlacks-1; |
---|
3940 | slack = numberP1; |
---|
3941 | type[numberP1++]=1; |
---|
3942 | } else if (slack >= 0) { |
---|
3943 | for (i=0;i<numberP1;i++) { |
---|
3944 | if (which[i]==slack) { |
---|
3945 | slack=i; |
---|
3946 | } |
---|
3947 | } |
---|
3948 | } |
---|
3949 | object[numberCliques] = new CbcClique(this,objectType,numberP1, |
---|
3950 | which,type, |
---|
3951 | 1000000+numberCliques,slack); |
---|
3952 | numberCliques++; |
---|
3953 | } else if (numberP1+numberM1 >= lessThanThis) { |
---|
3954 | // too big |
---|
3955 | numberBig++; |
---|
3956 | totalBig += numberP1+numberM1; |
---|
3957 | } |
---|
3958 | } |
---|
3959 | } |
---|
3960 | } |
---|
3961 | delete [] which; |
---|
3962 | delete [] type; |
---|
3963 | delete [] lookup; |
---|
3964 | if (numberCliques<0) { |
---|
3965 | printf("*** Problem infeasible\n"); |
---|
3966 | } else { |
---|
3967 | if (numberCliques) |
---|
3968 | printf("%d cliques of average size %g found, %d P1, %d M1\n", |
---|
3969 | numberCliques, |
---|
3970 | ((double)(totalP1+totalM1))/((double) numberCliques), |
---|
3971 | totalP1,totalM1); |
---|
3972 | else |
---|
3973 | printf("No cliques found\n"); |
---|
3974 | if (numberBig) |
---|
3975 | printf("%d large cliques ( >= %d) found, total %d\n", |
---|
3976 | numberBig,lessThanThis,totalBig); |
---|
3977 | if (numberFixed) |
---|
3978 | printf("%d variables fixed\n",numberFixed); |
---|
3979 | } |
---|
3980 | if (numberCliques>0&&numberSlacks&&makeEquality) { |
---|
3981 | printf("adding %d integer slacks\n",numberSlacks); |
---|
3982 | // add variables to make equality rows |
---|
3983 | int * temp = new int[numberIntegers_+numberSlacks]; |
---|
3984 | memcpy(temp,integerVariable_,numberIntegers_*sizeof(int)); |
---|
3985 | // Get new model |
---|
3986 | CbcModel * newModel = new CbcModel(*this); |
---|
3987 | OsiSolverInterface * newSolver = newModel->solver(); |
---|
3988 | for (i=0;i<numberSlacks;i++) { |
---|
3989 | temp[i+numberIntegers_]=i+numberColumns; |
---|
3990 | int iRow = rows[i]; |
---|
3991 | double value = element[i]; |
---|
3992 | double lowerValue = 0.0; |
---|
3993 | double upperValue = 1.0; |
---|
3994 | double objValue = 0.0; |
---|
3995 | CoinPackedVector column(1,&iRow,&value); |
---|
3996 | newSolver->addCol(column,lowerValue,upperValue,objValue); |
---|
3997 | // set integer |
---|
3998 | newSolver->setInteger(numberColumns+i); |
---|
3999 | if (value >0) |
---|
4000 | newSolver->setRowLower(iRow,rowUpper[iRow]); |
---|
4001 | else |
---|
4002 | newSolver->setRowUpper(iRow,rowLower[iRow]); |
---|
4003 | } |
---|
4004 | // replace list of integers |
---|
4005 | for (i=0;i<newModel->numberObjects_;i++) |
---|
4006 | delete newModel->object_[i]; |
---|
4007 | newModel->numberObjects_ = 0; |
---|
4008 | delete [] newModel->object_; |
---|
4009 | newModel->object_=NULL; |
---|
4010 | newModel->findIntegers(true); //Set up all integer objects |
---|
4011 | for (i=0;i<numberIntegers_;i++) { |
---|
4012 | newModel->modifiableObject(i)->setPriority(object_[i]->priority()); |
---|
4013 | } |
---|
4014 | if (originalColumns_) { |
---|
4015 | // old model had originalColumns |
---|
4016 | delete [] newModel->originalColumns_; |
---|
4017 | newModel->originalColumns_ = new int[numberColumns+numberSlacks]; |
---|
4018 | memcpy(newModel->originalColumns_,originalColumns_,numberColumns*sizeof(int)); |
---|
4019 | // mark as not in previous model |
---|
4020 | for (i=numberColumns;i<numberColumns+numberSlacks;i++) |
---|
4021 | newModel->originalColumns_[i]=-1; |
---|
4022 | } |
---|
4023 | delete [] rows; |
---|
4024 | delete [] element; |
---|
4025 | newModel->addObjects(numberCliques,object); |
---|
4026 | for (;i<numberCliques;i++) |
---|
4027 | delete object[i]; |
---|
4028 | delete [] object; |
---|
4029 | newModel->synchronizeModel(); |
---|
4030 | return newModel; |
---|
4031 | } else { |
---|
4032 | if (numberCliques>0) { |
---|
4033 | addObjects(numberCliques,object); |
---|
4034 | for (;i<numberCliques;i++) |
---|
4035 | delete object[i]; |
---|
4036 | synchronizeModel(); |
---|
4037 | } |
---|
4038 | delete [] object; |
---|
4039 | delete [] rows; |
---|
4040 | delete [] element; |
---|
4041 | return this; |
---|
4042 | } |
---|
4043 | } |
---|
4044 | |
---|
4045 | /* |
---|
4046 | Set branching priorities. |
---|
4047 | |
---|
4048 | Setting integer priorities looks pretty robust; the call to findIntegers |
---|
4049 | makes sure that SimpleInteger objects are in place. Setting priorities for |
---|
4050 | other objects is entirely dependent on their existence, and the routine may |
---|
4051 | quietly fail in several directions. |
---|
4052 | */ |
---|
4053 | |
---|
4054 | void |
---|
4055 | CbcModel::passInPriorities (const int * priorities, |
---|
4056 | bool ifObject) |
---|
4057 | { |
---|
4058 | findIntegers(false); |
---|
4059 | int i; |
---|
4060 | if (priorities) { |
---|
4061 | int i0=0; |
---|
4062 | int i1=numberObjects_-1; |
---|
4063 | if (ifObject) { |
---|
4064 | for (i=numberIntegers_;i<numberObjects_;i++) { |
---|
4065 | object_[i]->setPriority(priorities[i-numberIntegers_]); |
---|
4066 | } |
---|
4067 | i0=numberIntegers_; |
---|
4068 | } else { |
---|
4069 | for (i=0;i<numberIntegers_;i++) { |
---|
4070 | object_[i]->setPriority(priorities[i]); |
---|
4071 | } |
---|
4072 | i1=numberIntegers_-1; |
---|
4073 | } |
---|
4074 | messageHandler()->message(CBC_PRIORITY, |
---|
4075 | messages()) |
---|
4076 | << i0<<i1<<numberObjects_ << CoinMessageEol ; |
---|
4077 | } |
---|
4078 | } |
---|
4079 | |
---|
4080 | // Delete all object information |
---|
4081 | void |
---|
4082 | CbcModel::deleteObjects() |
---|
4083 | { |
---|
4084 | int i; |
---|
4085 | for (i=0;i<numberObjects_;i++) |
---|
4086 | delete object_[i]; |
---|
4087 | delete [] object_; |
---|
4088 | object_ = NULL; |
---|
4089 | numberObjects_=0; |
---|
4090 | findIntegers(true); |
---|
4091 | } |
---|
4092 | |
---|
4093 | /*! |
---|
4094 | Ensure all attached objects (CbcObjects, heuristics, and cut |
---|
4095 | generators) point to this model. |
---|
4096 | */ |
---|
4097 | void CbcModel::synchronizeModel() |
---|
4098 | { |
---|
4099 | int i; |
---|
4100 | for (i=0;i<numberHeuristics_;i++) |
---|
4101 | heuristic_[i]->setModel(this); |
---|
4102 | for (i=0;i<numberObjects_;i++) |
---|
4103 | object_[i]->setModel(this); |
---|
4104 | for (i=0;i<numberCutGenerators_;i++) |
---|
4105 | generator_[i]->refreshModel(this); |
---|
4106 | } |
---|
4107 | |
---|
4108 | // Fill in integers and create objects |
---|
4109 | |
---|
4110 | /** |
---|
4111 | The routine first does a scan to count the number of integer variables. |
---|
4112 | It then creates an array, integerVariable_, to store the indices of the |
---|
4113 | integer variables, and an array of `objects', one for each variable. |
---|
4114 | |
---|
4115 | The scan is repeated, this time recording the index of each integer |
---|
4116 | variable in integerVariable_, and creating an CbcSimpleInteger object that |
---|
4117 | contains information about the integer variable. Initially, this is just |
---|
4118 | the index and upper & lower bounds. |
---|
4119 | |
---|
4120 | \todo |
---|
4121 | Note the assumption in cbc that the first numberIntegers_ objects are |
---|
4122 | CbcSimpleInteger. In particular, the code which handles the startAgain |
---|
4123 | case assumes that if the object_ array exists it can simply replace the first |
---|
4124 | numberInteger_ objects. This is arguably unsafe. |
---|
4125 | |
---|
4126 | I am going to re-order if necessary |
---|
4127 | */ |
---|
4128 | |
---|
4129 | void |
---|
4130 | CbcModel::findIntegers(bool startAgain) |
---|
4131 | { |
---|
4132 | assert(solver_); |
---|
4133 | /* |
---|
4134 | No need to do this if we have previous information, unless forced to start |
---|
4135 | over. |
---|
4136 | */ |
---|
4137 | if (numberIntegers_&&!startAgain&&object_) |
---|
4138 | return; |
---|
4139 | /* |
---|
4140 | Clear out the old integer variable list, then count the number of integer |
---|
4141 | variables. |
---|
4142 | */ |
---|
4143 | delete [] integerVariable_; |
---|
4144 | numberIntegers_=0; |
---|
4145 | int numberColumns = getNumCols(); |
---|
4146 | int iColumn; |
---|
4147 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
4148 | if (isInteger(iColumn)) |
---|
4149 | numberIntegers_++; |
---|
4150 | } |
---|
4151 | // Find out how many old non-integer objects there are |
---|
4152 | int nObjects=0; |
---|
4153 | CbcObject ** oldObject = object_; |
---|
4154 | int iObject; |
---|
4155 | for (iObject = 0;iObject<numberObjects_;iObject++) { |
---|
4156 | CbcSimpleInteger * obj = |
---|
4157 | dynamic_cast <CbcSimpleInteger *>(oldObject[iObject]) ; |
---|
4158 | if (obj) |
---|
4159 | delete oldObject[iObject]; |
---|
4160 | else |
---|
4161 | oldObject[nObjects++]=oldObject[iObject]; |
---|
4162 | } |
---|
4163 | |
---|
4164 | /* |
---|
4165 | Found any? Allocate an array to hold the indices of the integer variables. |
---|
4166 | Make a large enough array for all objects |
---|
4167 | */ |
---|
4168 | object_ = new CbcObject * [numberIntegers_+nObjects]; |
---|
4169 | numberObjects_=numberIntegers_+nObjects;; |
---|
4170 | integerVariable_ = new int [numberIntegers_]; |
---|
4171 | /* |
---|
4172 | Walk the variables again, filling in the indices and creating objects for |
---|
4173 | the integer variables. Initially, the objects hold the index and upper & |
---|
4174 | lower bounds. |
---|
4175 | */ |
---|
4176 | numberIntegers_=0; |
---|
4177 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
4178 | if(isInteger(iColumn)) { |
---|
4179 | object_[numberIntegers_] = |
---|
4180 | new CbcSimpleInteger(this,numberIntegers_,iColumn); |
---|
4181 | integerVariable_[numberIntegers_++]=iColumn; |
---|
4182 | } |
---|
4183 | } |
---|
4184 | // Now append other objects |
---|
4185 | memcpy(object_+numberIntegers_,oldObject,nObjects*sizeof(CbcObject *)); |
---|
4186 | // Delete old array (just array) |
---|
4187 | delete [] oldObject; |
---|
4188 | |
---|
4189 | if (!numberObjects_) |
---|
4190 | handler_->message(CBC_NOINT,messages_) << CoinMessageEol ; |
---|
4191 | } |
---|
4192 | /* If numberBeforeTrust >0 then we are going to use CbcBranchDynamic. |
---|
4193 | Scan and convert CbcSimpleInteger objects |
---|
4194 | */ |
---|
4195 | void |
---|
4196 | CbcModel::convertToDynamic() |
---|
4197 | { |
---|
4198 | int iObject; |
---|
4199 | for (iObject = 0;iObject<numberObjects_;iObject++) { |
---|
4200 | CbcSimpleInteger * obj1 = |
---|
4201 | dynamic_cast <CbcSimpleInteger *>(object_[iObject]) ; |
---|
4202 | CbcSimpleIntegerDynamicPseudoCost * obj2 = |
---|
4203 | dynamic_cast <CbcSimpleIntegerDynamicPseudoCost *>(object_[iObject]) ; |
---|
4204 | if (obj1&&!obj2) { |
---|
4205 | // replace |
---|
4206 | int iColumn = obj1->columnNumber(); |
---|
4207 | delete object_[iObject]; |
---|
4208 | CbcSimpleIntegerDynamicPseudoCost * newObject = |
---|
4209 | new CbcSimpleIntegerDynamicPseudoCost(this,iObject,iColumn,0.3); |
---|
4210 | newObject->setNumberBeforeTrust(numberBeforeTrust_); |
---|
4211 | object_[iObject] = newObject; |
---|
4212 | } |
---|
4213 | } |
---|
4214 | if (branchingMethod_&&(branchingMethod_->whichMethod()&1)==0) { |
---|
4215 | // Need a method which can do better |
---|
4216 | branchingMethod_=NULL; |
---|
4217 | } |
---|
4218 | } |
---|
4219 | |
---|
4220 | /* Add in any object information (objects are cloned - owner can delete |
---|
4221 | originals */ |
---|
4222 | void |
---|
4223 | CbcModel::addObjects(int numberObjects, CbcObject ** objects) |
---|
4224 | { |
---|
4225 | // If integers but not enough objects fudge |
---|
4226 | if (numberIntegers_>numberObjects_) |
---|
4227 | findIntegers(true); |
---|
4228 | /* But if incoming objects inherit from simple integer we just want |
---|
4229 | to replace */ |
---|
4230 | int numberColumns = solver_->getNumCols(); |
---|
4231 | /** mark is -1 if not integer, >=0 if using existing simple integer and |
---|
4232 | >=numberColumns if using new integer */ |
---|
4233 | int * mark = new int[numberColumns]; |
---|
4234 | int i; |
---|
4235 | for (i=0;i<numberColumns;i++) |
---|
4236 | mark[i]=-1; |
---|
4237 | int newNumberObjects = numberObjects; |
---|
4238 | int newIntegers=0; |
---|
4239 | for (i=0;i<numberObjects;i++) { |
---|
4240 | CbcSimpleInteger * obj = |
---|
4241 | dynamic_cast <CbcSimpleInteger *>(objects[i]) ; |
---|
4242 | if (obj) { |
---|
4243 | int iColumn = obj->columnNumber(); |
---|
4244 | mark[iColumn]=i+numberColumns; |
---|
4245 | newIntegers++; |
---|
4246 | } |
---|
4247 | } |
---|
4248 | // and existing |
---|
4249 | for (i=0;i<numberObjects_;i++) { |
---|
4250 | CbcSimpleInteger * obj = |
---|
4251 | dynamic_cast <CbcSimpleInteger *>(object_[i]) ; |
---|
4252 | if (obj) { |
---|
4253 | int iColumn = obj->columnNumber(); |
---|
4254 | if (mark[iColumn]<0) { |
---|
4255 | newIntegers++; |
---|
4256 | newNumberObjects++; |
---|
4257 | mark[iColumn]=i; |
---|
4258 | } |
---|
4259 | } |
---|
4260 | } |
---|
4261 | delete [] integerVariable_; |
---|
4262 | integerVariable_=NULL; |
---|
4263 | if (newIntegers!=numberIntegers_) |
---|
4264 | printf("changing number of integers from %d to %d\n", |
---|
4265 | numberIntegers_,newIntegers); |
---|
4266 | numberIntegers_ = newIntegers; |
---|
4267 | integerVariable_ = new int [numberIntegers_]; |
---|
4268 | CbcObject ** temp = new CbcObject * [newNumberObjects]; |
---|
4269 | // Put integers first |
---|
4270 | newIntegers=0; |
---|
4271 | numberIntegers_=0; |
---|
4272 | for (i=0;i<numberColumns;i++) { |
---|
4273 | int which = mark[i]; |
---|
4274 | if (which>=0) { |
---|
4275 | if (!isInteger(i)) { |
---|
4276 | newIntegers++; |
---|
4277 | solver_->setInteger(i); |
---|
4278 | } |
---|
4279 | if (which<numberColumns) { |
---|
4280 | temp[numberIntegers_]=object_[which]; |
---|
4281 | object_[which]=NULL; |
---|
4282 | } else { |
---|
4283 | temp[numberIntegers_]=objects[which-numberColumns]->clone(); |
---|
4284 | temp[numberIntegers_]->setModel(this); |
---|
4285 | } |
---|
4286 | integerVariable_[numberIntegers_++]=i; |
---|
4287 | } |
---|
4288 | } |
---|
4289 | if (newIntegers) |
---|
4290 | printf("%d variables were declared integer\n",newIntegers); |
---|
4291 | int n=numberIntegers_; |
---|
4292 | // Now rest of old |
---|
4293 | for (i=0;i<numberObjects_;i++) { |
---|
4294 | if (object_[i]) { |
---|
4295 | CbcSimpleInteger * obj = |
---|
4296 | dynamic_cast <CbcSimpleInteger *>(object_[i]) ; |
---|
4297 | if (obj) { |
---|
4298 | delete object_[i]; |
---|
4299 | } else { |
---|
4300 | temp[n++]=object_[i]; |
---|
4301 | } |
---|
4302 | } |
---|
4303 | } |
---|
4304 | // and rest of new |
---|
4305 | for (i=0;i<numberObjects;i++) { |
---|
4306 | CbcSimpleInteger * obj = |
---|
4307 | dynamic_cast <CbcSimpleInteger *>(objects[i]) ; |
---|
4308 | if (!obj) { |
---|
4309 | temp[n]=objects[i]->clone(); |
---|
4310 | temp[n++]->setModel(this); |
---|
4311 | } |
---|
4312 | } |
---|
4313 | delete [] mark; |
---|
4314 | delete [] object_; |
---|
4315 | object_ = temp; |
---|
4316 | assert (n==newNumberObjects); |
---|
4317 | numberObjects_ = newNumberObjects; |
---|
4318 | } |
---|
4319 | |
---|
4320 | /** |
---|
4321 | This routine sets the objective cutoff value used for fathoming and |
---|
4322 | determining monotonic variables. |
---|
4323 | |
---|
4324 | If the fathoming discipline is strict, a small tolerance is added to the |
---|
4325 | new cutoff. This avoids problems due to roundoff when the target value |
---|
4326 | is exact. The common example would be an IP with only integer variables in |
---|
4327 | the objective. If the target is set to the exact value z of the optimum, |
---|
4328 | it's possible to end up fathoming an ancestor of the solution because the |
---|
4329 | solver returns z+epsilon. |
---|
4330 | |
---|
4331 | Determining if strict fathoming is needed is best done by analysis. |
---|
4332 | In cbc, that's analyseObjective. The default is false. |
---|
4333 | |
---|
4334 | In cbc we always minimize so add epsilon |
---|
4335 | */ |
---|
4336 | |
---|
4337 | void CbcModel::setCutoff (double value) |
---|
4338 | |
---|
4339 | { |
---|
4340 | #if 0 |
---|
4341 | double tol = 0 ; |
---|
4342 | int fathomStrict = getIntParam(CbcFathomDiscipline) ; |
---|
4343 | if (fathomStrict == 1) |
---|
4344 | { solver_->getDblParam(OsiDualTolerance,tol) ; |
---|
4345 | tol = tol*(1+fabs(value)) ; |
---|
4346 | |
---|
4347 | value += tol ; } |
---|
4348 | #endif |
---|
4349 | // Solvers know about direction |
---|
4350 | double direction = solver_->getObjSense(); |
---|
4351 | solver_->setDblParam(OsiDualObjectiveLimit,value*direction); } |
---|
4352 | |
---|
4353 | |
---|
4354 | |
---|
4355 | /* |
---|
4356 | Call this to really test if a valid solution can be feasible. The cutoff is |
---|
4357 | passed in as a parameter so that we don't need to worry here after swapping |
---|
4358 | solvers. The solution is assumed to be numberColumns in size. If |
---|
4359 | fixVariables is true then the bounds of the continuous solver are updated. |
---|
4360 | The routine returns the objective value determined by reoptimizing from |
---|
4361 | scratch. If the solution is rejected, this will be worse than the cutoff. |
---|
4362 | |
---|
4363 | TODO: There's an issue with getting the correct cutoff value: We update the |
---|
4364 | cutoff in the regular solver, but not in continuousSolver_. But our only |
---|
4365 | use for continuousSolver_ is verifying candidate solutions. Would it |
---|
4366 | make sense to update the cutoff? Then we wouldn't need to step around |
---|
4367 | isDualObjectiveLimitReached(). |
---|
4368 | */ |
---|
4369 | double |
---|
4370 | CbcModel::checkSolution (double cutoff, const double *solution, |
---|
4371 | bool fixVariables) |
---|
4372 | |
---|
4373 | { int numberColumns = solver_->getNumCols(); |
---|
4374 | |
---|
4375 | /* |
---|
4376 | Grab the continuous solver (the pristine copy of the problem, made before |
---|
4377 | starting to work on the root node). Save the bounds on the variables. |
---|
4378 | Install the solution passed as a parameter, and copy it to the model's |
---|
4379 | currentSolution_. |
---|
4380 | |
---|
4381 | TODO: This is a belt-and-suspenders approach. Once the code has settled |
---|
4382 | a bit, we can cast a critical eye here. |
---|
4383 | */ |
---|
4384 | OsiSolverInterface * saveSolver = solver_; |
---|
4385 | if (continuousSolver_) |
---|
4386 | solver_ = continuousSolver_; |
---|
4387 | // move solution to continuous copy |
---|
4388 | solver_->setColSolution(solution); |
---|
4389 | // Put current solution in safe place |
---|
4390 | // Point to current solution |
---|
4391 | const double * save = testSolution_; |
---|
4392 | // Safe as will be const inside infeasibility() |
---|
4393 | testSolution_ = solver_->getColSolution(); |
---|
4394 | //memcpy(currentSolution_,solver_->getColSolution(), |
---|
4395 | // numberColumns*sizeof(double)); |
---|
4396 | //solver_->messageHandler()->setLogLevel(4); |
---|
4397 | |
---|
4398 | // save original bounds |
---|
4399 | double * saveUpper = new double[numberColumns]; |
---|
4400 | double * saveLower = new double[numberColumns]; |
---|
4401 | memcpy(saveUpper,getColUpper(),numberColumns*sizeof(double)); |
---|
4402 | memcpy(saveLower,getColLower(),numberColumns*sizeof(double)); |
---|
4403 | |
---|
4404 | /* |
---|
4405 | Run through the objects and use feasibleRegion() to set variable bounds |
---|
4406 | so as to fix the variables specified in the objects at their value in this |
---|
4407 | solution. Since the object list contains (at least) one object for every |
---|
4408 | integer variable, this has the effect of fixing all integer variables. |
---|
4409 | */ |
---|
4410 | int i; |
---|
4411 | for (i=0;i<numberObjects_;i++) |
---|
4412 | object_[i]->feasibleRegion(); |
---|
4413 | // We can switch off check |
---|
4414 | if ((specialOptions_&4)==0) { |
---|
4415 | if ((specialOptions_&2)==0) { |
---|
4416 | /* |
---|
4417 | Remove any existing warm start information to be sure there is no |
---|
4418 | residual influence on initialSolve(). |
---|
4419 | */ |
---|
4420 | CoinWarmStartBasis *slack = |
---|
4421 | dynamic_cast<CoinWarmStartBasis *>(solver_->getEmptyWarmStart()) ; |
---|
4422 | solver_->setWarmStart(slack); |
---|
4423 | delete slack ; |
---|
4424 | } |
---|
4425 | // Give a hint not to do scaling |
---|
4426 | //bool saveTakeHint; |
---|
4427 | //OsiHintStrength saveStrength; |
---|
4428 | //bool gotHint = (solver_->getHintParam(OsiDoScale,saveTakeHint,saveStrength)); |
---|
4429 | //assert (gotHint); |
---|
4430 | //solver_->setHintParam(OsiDoScale,false,OsiHintTry); |
---|
4431 | solver_->initialSolve(); |
---|
4432 | //solver_->setHintParam(OsiDoScale,saveTakeHint,saveStrength); |
---|
4433 | if (!solver_->isProvenOptimal()) |
---|
4434 | { printf("checkSolution infeas! Retrying wihout scaling.\n"); |
---|
4435 | bool saveTakeHint; |
---|
4436 | OsiHintStrength saveStrength; |
---|
4437 | bool savePrintHint; |
---|
4438 | solver_->writeMps("infeas"); |
---|
4439 | bool gotHint = (solver_->getHintParam(OsiDoReducePrint,savePrintHint,saveStrength)); |
---|
4440 | gotHint = (solver_->getHintParam(OsiDoScale,saveTakeHint,saveStrength)); |
---|
4441 | solver_->setHintParam(OsiDoScale,false,OsiHintTry); |
---|
4442 | solver_->setHintParam(OsiDoReducePrint,false,OsiHintTry) ; |
---|
4443 | solver_->initialSolve(); |
---|
4444 | solver_->setHintParam(OsiDoScale,saveTakeHint,saveStrength); |
---|
4445 | solver_->setHintParam(OsiDoReducePrint,savePrintHint,OsiHintTry) ; |
---|
4446 | } |
---|
4447 | //assert(solver_->isProvenOptimal()); |
---|
4448 | } |
---|
4449 | double objectiveValue = solver_->getObjValue()*solver_->getObjSense(); |
---|
4450 | |
---|
4451 | /* |
---|
4452 | Check that the solution still beats the objective cutoff. |
---|
4453 | |
---|
4454 | If it passes, make a copy of the primal variable values and do some |
---|
4455 | cleanup and checks: |
---|
4456 | + Values of all variables are are within original bounds and values of |
---|
4457 | all integer variables are within tolerance of integral. |
---|
4458 | + There are no constraint violations. |
---|
4459 | There really should be no need for the check against original bounds. |
---|
4460 | Perhaps an opportunity for a sanity check? |
---|
4461 | */ |
---|
4462 | if ((solver_->isProvenOptimal()||(specialOptions_&4)!=0) && objectiveValue <= cutoff) |
---|
4463 | { |
---|
4464 | double * solution = new double[numberColumns]; |
---|
4465 | memcpy(solution ,solver_->getColSolution(),numberColumns*sizeof(double)) ; |
---|
4466 | |
---|
4467 | const double * rowLower = solver_->getRowLower() ; |
---|
4468 | const double * rowUpper = solver_->getRowUpper() ; |
---|
4469 | int numberRows = solver_->getNumRows() ; |
---|
4470 | double *rowActivity = new double[numberRows] ; |
---|
4471 | memset(rowActivity,0,numberRows*sizeof(double)) ; |
---|
4472 | |
---|
4473 | double integerTolerance = getIntegerTolerance() ; |
---|
4474 | |
---|
4475 | int iColumn; |
---|
4476 | for (iColumn = 0 ; iColumn < numberColumns ; iColumn++) |
---|
4477 | { double value = solution[iColumn] ; |
---|
4478 | value = CoinMax(value, saveLower[iColumn]) ; |
---|
4479 | value = CoinMin(value, saveUpper[iColumn]) ; |
---|
4480 | if (solver_->isInteger(iColumn)) |
---|
4481 | assert(fabs(value-solution[iColumn]) <= integerTolerance) ; |
---|
4482 | solution[iColumn] = value ; } |
---|
4483 | |
---|
4484 | solver_->getMatrixByCol()->times(solution,rowActivity) ; |
---|
4485 | delete [] solution; |
---|
4486 | double primalTolerance ; |
---|
4487 | solver_->getDblParam(OsiPrimalTolerance,primalTolerance) ; |
---|
4488 | double largestInfeasibility =0.0; |
---|
4489 | for (i=0 ; i < numberRows ; i++) { |
---|
4490 | largestInfeasibility = CoinMax(largestInfeasibility, |
---|
4491 | rowLower[i]-rowActivity[i]); |
---|
4492 | largestInfeasibility = CoinMax(largestInfeasibility, |
---|
4493 | rowActivity[i]-rowUpper[i]); |
---|
4494 | } |
---|
4495 | if (largestInfeasibility>100.0*primalTolerance) |
---|
4496 | handler_->message(CBC_NOTFEAS3, messages_) |
---|
4497 | << largestInfeasibility << CoinMessageEol ; |
---|
4498 | |
---|
4499 | delete [] rowActivity ; } |
---|
4500 | else |
---|
4501 | { objectiveValue=1.0e50 ; } |
---|
4502 | /* |
---|
4503 | Regardless of what we think of the solution, we may need to restore the |
---|
4504 | original bounds of the continuous solver. Unfortunately, const'ness |
---|
4505 | prevents us from simply reversing the memcpy used to make these snapshots. |
---|
4506 | */ |
---|
4507 | if (!fixVariables) |
---|
4508 | { for (int iColumn = 0 ; iColumn < numberColumns ; iColumn++) |
---|
4509 | { solver_->setColLower(iColumn,saveLower[iColumn]) ; |
---|
4510 | solver_->setColUpper(iColumn,saveUpper[iColumn]) ; } } |
---|
4511 | delete [] saveLower; |
---|
4512 | delete [] saveUpper; |
---|
4513 | |
---|
4514 | /* |
---|
4515 | Restore the usual solver. |
---|
4516 | */ |
---|
4517 | solver_=saveSolver; |
---|
4518 | testSolution_ = save; |
---|
4519 | return objectiveValue; |
---|
4520 | } |
---|
4521 | |
---|
4522 | /* |
---|
4523 | Call this routine from anywhere when a solution is found. The solution |
---|
4524 | vector is assumed to contain one value for each structural variable. |
---|
4525 | |
---|
4526 | The first action is to run checkSolution() to confirm the objective and |
---|
4527 | feasibility. If this check causes the solution to be rejected, we're done. |
---|
4528 | If fixVariables = true, the variable bounds held by the continuous solver |
---|
4529 | will be left fixed to the values in the solution; otherwise they are |
---|
4530 | restored to the original values. |
---|
4531 | |
---|
4532 | If the solution is accepted, install it as the best solution. |
---|
4533 | |
---|
4534 | The routine also contains a hook to run any cut generators that are flagged |
---|
4535 | to run when a new solution is discovered. There's a potential hazard because |
---|
4536 | the cut generators see the continuous solver >after< possible restoration of |
---|
4537 | original bounds (which may well invalidate the solution). |
---|
4538 | */ |
---|
4539 | |
---|
4540 | void |
---|
4541 | CbcModel::setBestSolution (CBC_Message how, |
---|
4542 | double & objectiveValue, const double *solution, |
---|
4543 | bool fixVariables) |
---|
4544 | |
---|
4545 | { double cutoff = getCutoff() ; |
---|
4546 | |
---|
4547 | /* |
---|
4548 | Double check the solution to catch pretenders. |
---|
4549 | */ |
---|
4550 | objectiveValue = checkSolution(cutoff,solution,fixVariables); |
---|
4551 | if (objectiveValue > cutoff) |
---|
4552 | { if (objectiveValue>1.0e30) |
---|
4553 | handler_->message(CBC_NOTFEAS1, messages_) << CoinMessageEol ; |
---|
4554 | else |
---|
4555 | handler_->message(CBC_NOTFEAS2, messages_) |
---|
4556 | << objectiveValue << cutoff << CoinMessageEol ; } |
---|
4557 | /* |
---|
4558 | We have a winner. Install it as the new incumbent. |
---|
4559 | Bump the objective cutoff value and solution counts. Give the user the |
---|
4560 | good news. |
---|
4561 | */ |
---|
4562 | else |
---|
4563 | { bestObjective_ = objectiveValue; |
---|
4564 | int numberColumns = solver_->getNumCols(); |
---|
4565 | if (!bestSolution_) |
---|
4566 | bestSolution_ = new double[numberColumns]; |
---|
4567 | CoinCopyN(solution,numberColumns,bestSolution_); |
---|
4568 | |
---|
4569 | cutoff = bestObjective_-dblParam_[CbcCutoffIncrement]; |
---|
4570 | // This is not correct - that way cutoff can go up if maximization |
---|
4571 | //double direction = solver_->getObjSense(); |
---|
4572 | //setCutoff(cutoff*direction); |
---|
4573 | setCutoff(cutoff); |
---|
4574 | |
---|
4575 | if (how==CBC_ROUNDING) |
---|
4576 | numberHeuristicSolutions_++; |
---|
4577 | numberSolutions_++; |
---|
4578 | if (numberHeuristicSolutions_==numberSolutions_) |
---|
4579 | stateOfSearch_ = 1; |
---|
4580 | else |
---|
4581 | stateOfSearch_ = 2; |
---|
4582 | |
---|
4583 | handler_->message(how,messages_) |
---|
4584 | <<bestObjective_<<numberIterations_ |
---|
4585 | <<numberNodes_ |
---|
4586 | <<CoinMessageEol; |
---|
4587 | /* |
---|
4588 | Now step through the cut generators and see if any of them are flagged to |
---|
4589 | run when a new solution is discovered. Only global cuts are useful. (The |
---|
4590 | solution being evaluated may not correspond to the current location in the |
---|
4591 | search tree --- discovered by heuristic, for example.) |
---|
4592 | */ |
---|
4593 | OsiCuts theseCuts; |
---|
4594 | int i; |
---|
4595 | int lastNumberCuts=0; |
---|
4596 | for (i=0;i<numberCutGenerators_;i++) { |
---|
4597 | if (generator_[i]->atSolution()) { |
---|
4598 | generator_[i]->generateCuts(theseCuts,true,NULL); |
---|
4599 | int numberCuts = theseCuts.sizeRowCuts(); |
---|
4600 | for (int j=lastNumberCuts;j<numberCuts;j++) { |
---|
4601 | const OsiRowCut * thisCut = theseCuts.rowCutPtr(j); |
---|
4602 | if (thisCut->globallyValid()) { |
---|
4603 | if ((specialOptions_&1)!=0) { |
---|
4604 | /* As these are global cuts - |
---|
4605 | a) Always get debugger object |
---|
4606 | b) Not fatal error to cutoff optimal (if we have just got optimal) |
---|
4607 | */ |
---|
4608 | const OsiRowCutDebugger *debugger = solver_->getRowCutDebuggerAlways() ; |
---|
4609 | if (debugger) { |
---|
4610 | if(debugger->invalidCut(*thisCut)) |
---|
4611 | printf("ZZZZ Global cut - cuts off optimal solution!\n"); |
---|
4612 | } |
---|
4613 | } |
---|
4614 | // add to global list |
---|
4615 | globalCuts_.insert(*thisCut); |
---|
4616 | generator_[i]->incrementNumberCutsInTotal(); |
---|
4617 | } |
---|
4618 | } |
---|
4619 | } |
---|
4620 | } |
---|
4621 | int numberCuts = theseCuts.sizeColCuts(); |
---|
4622 | for (i=0;i<numberCuts;i++) { |
---|
4623 | const OsiColCut * thisCut = theseCuts.colCutPtr(i); |
---|
4624 | if (thisCut->globallyValid()) { |
---|
4625 | // add to global list |
---|
4626 | globalCuts_.insert(*thisCut); |
---|
4627 | } |
---|
4628 | } |
---|
4629 | } |
---|
4630 | |
---|
4631 | return ; } |
---|
4632 | |
---|
4633 | |
---|
4634 | /* Test the current solution for feasibility. |
---|
4635 | |
---|
4636 | Calculate the number of standard integer infeasibilities, then scan the |
---|
4637 | remaining objects to see if any of them report infeasibilities. |
---|
4638 | |
---|
4639 | Currently (2003.08) the only object besides SimpleInteger is Clique, hence |
---|
4640 | the comments about `odd ones' infeasibilities. |
---|
4641 | */ |
---|
4642 | bool |
---|
4643 | CbcModel::feasibleSolution(int & numberIntegerInfeasibilities, |
---|
4644 | int & numberObjectInfeasibilities) const |
---|
4645 | { |
---|
4646 | int numberUnsatisfied=0; |
---|
4647 | double sumUnsatisfied=0.0; |
---|
4648 | int preferredWay; |
---|
4649 | int j; |
---|
4650 | // Point to current solution |
---|
4651 | const double * save = testSolution_; |
---|
4652 | // Safe as will be const inside infeasibility() |
---|
4653 | testSolution_ = solver_->getColSolution(); |
---|
4654 | // Put current solution in safe place |
---|
4655 | //memcpy(currentSolution_,solver_->getColSolution(), |
---|
4656 | // solver_->getNumCols()*sizeof(double)); |
---|
4657 | for (j=0;j<numberIntegers_;j++) { |
---|
4658 | const CbcObject * object = object_[j]; |
---|
4659 | double infeasibility = object->infeasibility(preferredWay); |
---|
4660 | if (infeasibility) { |
---|
4661 | assert (infeasibility>0); |
---|
4662 | numberUnsatisfied++; |
---|
4663 | sumUnsatisfied += infeasibility; |
---|
4664 | } |
---|
4665 | } |
---|
4666 | numberIntegerInfeasibilities = numberUnsatisfied; |
---|
4667 | for (;j<numberObjects_;j++) { |
---|
4668 | const CbcObject * object = object_[j]; |
---|
4669 | double infeasibility = object->infeasibility(preferredWay); |
---|
4670 | if (infeasibility) { |
---|
4671 | assert (infeasibility>0); |
---|
4672 | numberUnsatisfied++; |
---|
4673 | sumUnsatisfied += infeasibility; |
---|
4674 | } |
---|
4675 | } |
---|
4676 | // and restore |
---|
4677 | testSolution_ = save; |
---|
4678 | numberObjectInfeasibilities = numberUnsatisfied-numberIntegerInfeasibilities; |
---|
4679 | return (!numberUnsatisfied); |
---|
4680 | } |
---|
4681 | |
---|
4682 | /* For all vubs see if we can tighten bounds by solving Lp's |
---|
4683 | type - 0 just vubs |
---|
4684 | 1 all (could be very slow) |
---|
4685 | -1 just vubs where variable away from bound |
---|
4686 | Returns false if not feasible |
---|
4687 | */ |
---|
4688 | bool |
---|
4689 | CbcModel::tightenVubs(int type, bool allowMultipleBinary, double useCutoff) |
---|
4690 | { |
---|
4691 | |
---|
4692 | CoinPackedMatrix matrixByRow(*solver_->getMatrixByRow()); |
---|
4693 | int numberRows = solver_->getNumRows(); |
---|
4694 | int numberColumns = solver_->getNumCols(); |
---|
4695 | |
---|
4696 | int iRow,iColumn; |
---|
4697 | |
---|
4698 | // Row copy |
---|
4699 | //const double * elementByRow = matrixByRow.getElements(); |
---|
4700 | const int * column = matrixByRow.getIndices(); |
---|
4701 | const CoinBigIndex * rowStart = matrixByRow.getVectorStarts(); |
---|
4702 | const int * rowLength = matrixByRow.getVectorLengths(); |
---|
4703 | |
---|
4704 | const double * colUpper = solver_->getColUpper(); |
---|
4705 | const double * colLower = solver_->getColLower(); |
---|
4706 | //const double * rowUpper = solver_->getRowUpper(); |
---|
4707 | //const double * rowLower = solver_->getRowLower(); |
---|
4708 | |
---|
4709 | const double * objective = solver_->getObjCoefficients(); |
---|
4710 | //double direction = solver_->getObjSense(); |
---|
4711 | const double * colsol = solver_->getColSolution(); |
---|
4712 | |
---|
4713 | int numberVub=0; |
---|
4714 | int * continuous = new int[numberColumns]; |
---|
4715 | if (type >= 0) { |
---|
4716 | double * sort = new double[numberColumns]; |
---|
4717 | for (iRow=0;iRow<numberRows;iRow++) { |
---|
4718 | int j; |
---|
4719 | int numberBinary=0; |
---|
4720 | int numberUnsatisfiedBinary=0; |
---|
4721 | int numberContinuous=0; |
---|
4722 | int iCont=-1; |
---|
4723 | double weight=1.0e30; |
---|
4724 | for (j=rowStart[iRow];j<rowStart[iRow]+rowLength[iRow];j++) { |
---|
4725 | int iColumn = column[j]; |
---|
4726 | if (colUpper[iColumn]-colLower[iColumn]>1.0e-8) { |
---|
4727 | if (solver_->isFreeBinary(iColumn)) { |
---|
4728 | numberBinary++; |
---|
4729 | /* For sort I make naive assumption: |
---|
4730 | x - a * delta <=0 or |
---|
4731 | -x + a * delta >= 0 |
---|
4732 | */ |
---|
4733 | if (colsol[iColumn]>colLower[iColumn]+1.0e-6&& |
---|
4734 | colsol[iColumn]<colUpper[iColumn]-1.0e-6) { |
---|
4735 | numberUnsatisfiedBinary++; |
---|
4736 | weight = CoinMin(weight,fabs(objective[iColumn])); |
---|
4737 | } |
---|
4738 | } else { |
---|
4739 | numberContinuous++; |
---|
4740 | iCont=iColumn; |
---|
4741 | } |
---|
4742 | } |
---|
4743 | } |
---|
4744 | if (numberContinuous==1&&numberBinary) { |
---|
4745 | if (numberBinary==1||allowMultipleBinary) { |
---|
4746 | // treat as vub |
---|
4747 | if (!numberUnsatisfiedBinary) |
---|
4748 | weight=-1.0; // at end |
---|
4749 | sort[numberVub]=-weight; |
---|
4750 | continuous[numberVub++] = iCont; |
---|
4751 | } |
---|
4752 | } |
---|
4753 | } |
---|
4754 | if (type>0) { |
---|
4755 | // take so many |
---|
4756 | CoinSort_2(sort,sort+numberVub,continuous); |
---|
4757 | numberVub = CoinMin(numberVub,type); |
---|
4758 | } |
---|
4759 | delete [] sort; |
---|
4760 | } else { |
---|
4761 | for (iColumn=0;iColumn<numberColumns;iColumn++) |
---|
4762 | continuous[iColumn]=iColumn; |
---|
4763 | numberVub=numberColumns; |
---|
4764 | } |
---|
4765 | bool feasible = tightenVubs(numberVub,continuous,useCutoff); |
---|
4766 | delete [] continuous; |
---|
4767 | |
---|
4768 | return feasible; |
---|
4769 | } |
---|
4770 | // This version is just handed a list of variables |
---|
4771 | bool |
---|
4772 | CbcModel::tightenVubs(int numberSolves, const int * which, |
---|
4773 | double useCutoff) |
---|
4774 | { |
---|
4775 | |
---|
4776 | int numberColumns = solver_->getNumCols(); |
---|
4777 | |
---|
4778 | int iColumn; |
---|
4779 | |
---|
4780 | OsiSolverInterface * solver = solver_; |
---|
4781 | double saveCutoff = getCutoff() ; |
---|
4782 | |
---|
4783 | double * objective = new double[numberColumns]; |
---|
4784 | memcpy(objective,solver_->getObjCoefficients(),numberColumns*sizeof(double)); |
---|
4785 | double direction = solver_->getObjSense(); |
---|
4786 | |
---|
4787 | // add in objective if there is a cutoff |
---|
4788 | if (useCutoff<1.0e30) { |
---|
4789 | // get new version of model |
---|
4790 | solver = solver_->clone(); |
---|
4791 | CoinPackedVector newRow; |
---|
4792 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
4793 | solver->setObjCoeff(iColumn,0.0); // zero out in new model |
---|
4794 | if (objective[iColumn]) |
---|
4795 | newRow.insert(iColumn,direction * objective[iColumn]); |
---|
4796 | |
---|
4797 | } |
---|
4798 | solver->addRow(newRow,-COIN_DBL_MAX,useCutoff); |
---|
4799 | // signal no objective |
---|
4800 | delete [] objective; |
---|
4801 | objective=NULL; |
---|
4802 | } |
---|
4803 | setCutoff(COIN_DBL_MAX); |
---|
4804 | |
---|
4805 | |
---|
4806 | bool * vub = new bool [numberColumns]; |
---|
4807 | int iVub; |
---|
4808 | |
---|
4809 | // mark vub columns |
---|
4810 | for (iColumn=0;iColumn<numberColumns;iColumn++) |
---|
4811 | vub[iColumn]=false; |
---|
4812 | for (iVub=0;iVub<numberSolves;iVub++) |
---|
4813 | vub[which[iVub]]=true; |
---|
4814 | OsiCuts cuts; |
---|
4815 | // First tighten bounds anyway if CglProbing there |
---|
4816 | CglProbing* generator = NULL; |
---|
4817 | int iGen; |
---|
4818 | for (iGen=0;iGen<numberCutGenerators_;iGen++) { |
---|
4819 | generator = dynamic_cast<CglProbing*>(generator_[iGen]->generator()); |
---|
4820 | if (generator) |
---|
4821 | break; |
---|
4822 | } |
---|
4823 | int numberFixed=0; |
---|
4824 | int numberTightened=0; |
---|
4825 | int numberFixedByProbing=0; |
---|
4826 | int numberTightenedByProbing=0; |
---|
4827 | int printFrequency = (numberSolves+19)/20; // up to 20 messages |
---|
4828 | int save[4]; |
---|
4829 | if (generator) { |
---|
4830 | // set to cheaper and then restore at end |
---|
4831 | save[0]=generator->getMaxPass(); |
---|
4832 | save[1]=generator->getMaxProbe(); |
---|
4833 | save[2]=generator->getMaxLook(); |
---|
4834 | save[3]=generator->rowCuts(); |
---|
4835 | generator->setMaxPass(1); |
---|
4836 | generator->setMaxProbe(10); |
---|
4837 | generator->setMaxLook(50); |
---|
4838 | generator->setRowCuts(0); |
---|
4839 | |
---|
4840 | // Probing - return tight column bounds |
---|
4841 | generator->generateCutsAndModify(*solver,cuts); |
---|
4842 | const double * tightLower = generator->tightLower(); |
---|
4843 | const double * lower = solver->getColLower(); |
---|
4844 | const double * tightUpper = generator->tightUpper(); |
---|
4845 | const double * upper = solver->getColUpper(); |
---|
4846 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
4847 | double newUpper = tightUpper[iColumn]; |
---|
4848 | double newLower = tightLower[iColumn]; |
---|
4849 | if (newUpper<upper[iColumn]-1.0e-8*(fabs(upper[iColumn])+1)|| |
---|
4850 | newLower>lower[iColumn]+1.0e-8*(fabs(lower[iColumn])+1)) { |
---|
4851 | if (newUpper<newLower) { |
---|
4852 | fprintf(stderr,"Problem is infeasible\n"); |
---|
4853 | return false; |
---|
4854 | } |
---|
4855 | if (newUpper==newLower) { |
---|
4856 | numberFixed++; |
---|
4857 | numberFixedByProbing++; |
---|
4858 | solver->setColLower(iColumn,newLower); |
---|
4859 | solver->setColUpper(iColumn,newUpper); |
---|
4860 | printf("Column %d, new bounds %g %g\n",iColumn, |
---|
4861 | newLower,newUpper); |
---|
4862 | } else if (vub[iColumn]) { |
---|
4863 | numberTightened++; |
---|
4864 | numberTightenedByProbing++; |
---|
4865 | if (!solver->isInteger(iColumn)) { |
---|
4866 | // relax |
---|
4867 | newLower=CoinMax(lower[iColumn], |
---|
4868 | newLower |
---|
4869 | -1.0e-5*(fabs(lower[iColumn])+1)); |
---|
4870 | newUpper=CoinMin(upper[iColumn], |
---|
4871 | newUpper |
---|
4872 | +1.0e-5*(fabs(upper[iColumn])+1)); |
---|
4873 | } |
---|
4874 | solver->setColLower(iColumn,newLower); |
---|
4875 | solver->setColUpper(iColumn,newUpper); |
---|
4876 | } |
---|
4877 | } |
---|
4878 | } |
---|
4879 | } |
---|
4880 | CoinWarmStart * ws = solver->getWarmStart(); |
---|
4881 | double * solution = new double [numberColumns]; |
---|
4882 | memcpy(solution,solver->getColSolution(),numberColumns*sizeof(double)); |
---|
4883 | for (iColumn=0;iColumn<numberColumns;iColumn++) |
---|
4884 | solver->setObjCoeff(iColumn,0.0); |
---|
4885 | //solver->messageHandler()->setLogLevel(2); |
---|
4886 | for (iVub=0;iVub<numberSolves;iVub++) { |
---|
4887 | iColumn = which[iVub]; |
---|
4888 | int iTry; |
---|
4889 | for (iTry=0;iTry<2;iTry++) { |
---|
4890 | double saveUpper = solver->getColUpper()[iColumn]; |
---|
4891 | double saveLower = solver->getColLower()[iColumn]; |
---|
4892 | double value; |
---|
4893 | if (iTry==1) { |
---|
4894 | // try all way up |
---|
4895 | solver->setObjCoeff(iColumn,-1.0); |
---|
4896 | } else { |
---|
4897 | // try all way down |
---|
4898 | solver->setObjCoeff(iColumn,1.0); |
---|
4899 | } |
---|
4900 | solver->initialSolve(); |
---|
4901 | value = solver->getColSolution()[iColumn]; |
---|
4902 | bool change=false; |
---|
4903 | if (iTry==1) { |
---|
4904 | if (value<saveUpper-1.0e-4) { |
---|
4905 | if (solver->isInteger(iColumn)) { |
---|
4906 | value = floor(value+0.00001); |
---|
4907 | } else { |
---|
4908 | // relax a bit |
---|
4909 | value=CoinMin(saveUpper,value+1.0e-5*(fabs(saveUpper)+1)); |
---|
4910 | } |
---|
4911 | if (value-saveLower<1.0e-7) |
---|
4912 | value = saveLower; // make sure exactly same |
---|
4913 | solver->setColUpper(iColumn,value); |
---|
4914 | saveUpper=value; |
---|
4915 | change=true; |
---|
4916 | } |
---|
4917 | } else { |
---|
4918 | if (value>saveLower+1.0e-4) { |
---|
4919 | if (solver->isInteger(iColumn)) { |
---|
4920 | value = ceil(value-0.00001); |
---|
4921 | } else { |
---|
4922 | // relax a bit |
---|
4923 | value=CoinMax(saveLower,value-1.0e-5*(fabs(saveLower)+1)); |
---|
4924 | } |
---|
4925 | if (saveUpper-value<1.0e-7) |
---|
4926 | value = saveUpper; // make sure exactly same |
---|
4927 | solver->setColLower(iColumn,value); |
---|
4928 | saveLower=value; |
---|
4929 | change=true; |
---|
4930 | } |
---|
4931 | } |
---|
4932 | solver->setObjCoeff(iColumn,0.0); |
---|
4933 | if (change) { |
---|
4934 | if (saveUpper==saveLower) |
---|
4935 | numberFixed++; |
---|
4936 | else |
---|
4937 | numberTightened++; |
---|
4938 | int saveFixed=numberFixed; |
---|
4939 | |
---|
4940 | int jColumn; |
---|
4941 | if (generator) { |
---|
4942 | // Probing - return tight column bounds |
---|
4943 | cuts = OsiCuts(); |
---|
4944 | generator->generateCutsAndModify(*solver,cuts); |
---|
4945 | const double * tightLower = generator->tightLower(); |
---|
4946 | const double * lower = solver->getColLower(); |
---|
4947 | const double * tightUpper = generator->tightUpper(); |
---|
4948 | const double * upper = solver->getColUpper(); |
---|
4949 | for (jColumn=0;jColumn<numberColumns;jColumn++) { |
---|
4950 | double newUpper = tightUpper[jColumn]; |
---|
4951 | double newLower = tightLower[jColumn]; |
---|
4952 | if (newUpper<upper[jColumn]-1.0e-8*(fabs(upper[jColumn])+1)|| |
---|
4953 | newLower>lower[jColumn]+1.0e-8*(fabs(lower[jColumn])+1)) { |
---|
4954 | if (newUpper<newLower) { |
---|
4955 | fprintf(stderr,"Problem is infeasible\n"); |
---|
4956 | return false; |
---|
4957 | } |
---|
4958 | if (newUpper==newLower) { |
---|
4959 | numberFixed++; |
---|
4960 | numberFixedByProbing++; |
---|
4961 | solver->setColLower(jColumn,newLower); |
---|
4962 | solver->setColUpper(jColumn,newUpper); |
---|
4963 | } else if (vub[jColumn]) { |
---|
4964 | numberTightened++; |
---|
4965 | numberTightenedByProbing++; |
---|
4966 | if (!solver->isInteger(jColumn)) { |
---|
4967 | // relax |
---|
4968 | newLower=CoinMax(lower[jColumn], |
---|
4969 | newLower |
---|
4970 | -1.0e-5*(fabs(lower[jColumn])+1)); |
---|
4971 | newUpper=CoinMin(upper[jColumn], |
---|
4972 | newUpper |
---|
4973 | +1.0e-5*(fabs(upper[jColumn])+1)); |
---|
4974 | } |
---|
4975 | solver->setColLower(jColumn,newLower); |
---|
4976 | solver->setColUpper(jColumn,newUpper); |
---|
4977 | } |
---|
4978 | } |
---|
4979 | } |
---|
4980 | } |
---|
4981 | if (numberFixed>saveFixed) { |
---|
4982 | // original solution may not be feasible |
---|
4983 | // go back to true costs to solve if exists |
---|
4984 | if (objective) { |
---|
4985 | for (jColumn=0;jColumn<numberColumns;jColumn++) |
---|
4986 | solver->setObjCoeff(jColumn,objective[jColumn]); |
---|
4987 | } |
---|
4988 | solver->setColSolution(solution); |
---|
4989 | solver->setWarmStart(ws); |
---|
4990 | solver->resolve(); |
---|
4991 | if (!solver->isProvenOptimal()) { |
---|
4992 | fprintf(stderr,"Problem is infeasible\n"); |
---|
4993 | return false; |
---|
4994 | } |
---|
4995 | delete ws; |
---|
4996 | ws = solver->getWarmStart(); |
---|
4997 | memcpy(solution,solver->getColSolution(), |
---|
4998 | numberColumns*sizeof(double)); |
---|
4999 | for (jColumn=0;jColumn<numberColumns;jColumn++) |
---|
5000 | solver->setObjCoeff(jColumn,0.0); |
---|
5001 | } |
---|
5002 | } |
---|
5003 | solver->setColSolution(solution); |
---|
5004 | solver->setWarmStart(ws); |
---|
5005 | } |
---|
5006 | if (iVub%printFrequency==0) |
---|
5007 | handler_->message(CBC_VUB_PASS,messages_) |
---|
5008 | <<iVub+1<<numberFixed<<numberTightened |
---|
5009 | <<CoinMessageEol; |
---|
5010 | } |
---|
5011 | handler_->message(CBC_VUB_END,messages_) |
---|
5012 | <<numberFixed<<numberTightened |
---|
5013 | <<CoinMessageEol; |
---|
5014 | delete ws; |
---|
5015 | delete [] solution; |
---|
5016 | // go back to true costs to solve if exists |
---|
5017 | if (objective) { |
---|
5018 | for (iColumn=0;iColumn<numberColumns;iColumn++) |
---|
5019 | solver_->setObjCoeff(iColumn,objective[iColumn]); |
---|
5020 | delete [] objective; |
---|
5021 | } |
---|
5022 | delete [] vub; |
---|
5023 | if (generator) { |
---|
5024 | /*printf("Probing fixed %d and tightened %d\n", |
---|
5025 | numberFixedByProbing, |
---|
5026 | numberTightenedByProbing);*/ |
---|
5027 | if (generator_[iGen]->howOften()==-1&& |
---|
5028 | (numberFixedByProbing+numberTightenedByProbing)*5> |
---|
5029 | (numberFixed+numberTightened)) |
---|
5030 | generator_[iGen]->setHowOften(1000000+1); |
---|
5031 | generator->setMaxPass(save[0]); |
---|
5032 | generator->setMaxProbe(save[1]); |
---|
5033 | generator->setMaxLook(save[2]); |
---|
5034 | generator->setRowCuts(save[3]); |
---|
5035 | } |
---|
5036 | |
---|
5037 | if (solver!=solver_) { |
---|
5038 | // move bounds across |
---|
5039 | const double * lower = solver->getColLower(); |
---|
5040 | const double * upper = solver->getColUpper(); |
---|
5041 | const double * lowerOrig = solver_->getColLower(); |
---|
5042 | const double * upperOrig = solver_->getColUpper(); |
---|
5043 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
5044 | solver_->setColLower(iColumn,CoinMax(lower[iColumn],lowerOrig[iColumn])); |
---|
5045 | solver_->setColUpper(iColumn,CoinMin(upper[iColumn],upperOrig[iColumn])); |
---|
5046 | } |
---|
5047 | delete solver; |
---|
5048 | } |
---|
5049 | setCutoff(saveCutoff); |
---|
5050 | return true; |
---|
5051 | } |
---|
5052 | /* |
---|
5053 | Do Integer Presolve. Returns new model. |
---|
5054 | I have to work out cleanest way of getting solution to |
---|
5055 | original problem at end. So this is very preliminary. |
---|
5056 | */ |
---|
5057 | CbcModel * |
---|
5058 | CbcModel::integerPresolve(bool weak) |
---|
5059 | { |
---|
5060 | status_ = 0; |
---|
5061 | // solve LP |
---|
5062 | //solver_->writeMps("bad"); |
---|
5063 | bool feasible = resolve(); |
---|
5064 | |
---|
5065 | CbcModel * newModel = NULL; |
---|
5066 | if (feasible) { |
---|
5067 | |
---|
5068 | // get a new model |
---|
5069 | newModel = new CbcModel(*this); |
---|
5070 | newModel->messageHandler()->setLogLevel(messageHandler()->logLevel()); |
---|
5071 | |
---|
5072 | feasible = newModel->integerPresolveThisModel(solver_,weak); |
---|
5073 | } |
---|
5074 | if (!feasible) { |
---|
5075 | handler_->message(CBC_INFEAS,messages_) |
---|
5076 | <<CoinMessageEol; |
---|
5077 | status_ = 0; |
---|
5078 | secondaryStatus_ = 1; |
---|
5079 | delete newModel; |
---|
5080 | return NULL; |
---|
5081 | } else { |
---|
5082 | newModel->synchronizeModel(); // make sure everything that needs solver has it |
---|
5083 | return newModel; |
---|
5084 | } |
---|
5085 | } |
---|
5086 | /* |
---|
5087 | Do Integer Presolve - destroying current model |
---|
5088 | */ |
---|
5089 | bool |
---|
5090 | CbcModel::integerPresolveThisModel(OsiSolverInterface * originalSolver, |
---|
5091 | bool weak) |
---|
5092 | { |
---|
5093 | status_ = 0; |
---|
5094 | // solve LP |
---|
5095 | bool feasible = resolve(); |
---|
5096 | |
---|
5097 | bestObjective_=1.0e50; |
---|
5098 | numberSolutions_=0; |
---|
5099 | numberHeuristicSolutions_=0; |
---|
5100 | double cutoff = getCutoff() ; |
---|
5101 | double direction = solver_->getObjSense(); |
---|
5102 | if (cutoff < 1.0e20&&direction<0.0) |
---|
5103 | messageHandler()->message(CBC_CUTOFF_WARNING1, |
---|
5104 | messages()) |
---|
5105 | << cutoff << -cutoff << CoinMessageEol ; |
---|
5106 | if (cutoff > bestObjective_) |
---|
5107 | cutoff = bestObjective_ ; |
---|
5108 | setCutoff(cutoff) ; |
---|
5109 | int iColumn; |
---|
5110 | int numberColumns = getNumCols(); |
---|
5111 | int originalNumberColumns = numberColumns; |
---|
5112 | currentPassNumber_=0; |
---|
5113 | synchronizeModel(); // make sure everything that needs solver has it |
---|
5114 | // just point to solver_ |
---|
5115 | delete continuousSolver_; |
---|
5116 | continuousSolver_ = solver_; |
---|
5117 | // get a copy of original so we can fix bounds |
---|
5118 | OsiSolverInterface * cleanModel = originalSolver->clone(); |
---|
5119 | #ifdef CBC_DEBUG |
---|
5120 | std::string problemName; |
---|
5121 | cleanModel->getStrParam(OsiProbName,problemName); |
---|
5122 | printf("Problem name - %s\n",problemName.c_str()); |
---|
5123 | cleanModel->activateRowCutDebugger(problemName.c_str()); |
---|
5124 | const OsiRowCutDebugger * debugger = cleanModel->getRowCutDebugger(); |
---|
5125 | #endif |
---|
5126 | |
---|
5127 | // array which points from original columns to presolved |
---|
5128 | int * original = new int[numberColumns]; |
---|
5129 | // arrays giving bounds - only ones found by probing |
---|
5130 | // rest will be found by presolve |
---|
5131 | double * originalLower = new double[numberColumns]; |
---|
5132 | double * originalUpper = new double[numberColumns]; |
---|
5133 | { |
---|
5134 | const double * lower = getColLower(); |
---|
5135 | const double * upper = getColUpper(); |
---|
5136 | for (iColumn=0;iColumn<numberColumns;iColumn++) { |
---|
5137 | original[iColumn]=iColumn; |
---|
5138 | originalLower[iColumn] = lower[iColumn]; |
---|
5139 | originalUpper[iColumn] = upper[iColumn]; |
---|
5140 | } |
---|
5141 | } |
---|
5142 | findIntegers(true); |
---|
5143 | // save original integers |
---|
5144 | int * originalIntegers = new int[numberIntegers_]; |
---|
5145 | int originalNumberIntegers = numberIntegers_; |
---|
5146 | memcpy(originalIntegers,integerVariable_,numberIntegers_*sizeof(int)); |
---|
5147 | |
---|
5148 | int todo=20; |
---|
5149 | if (weak) |
---|
5150 | todo=1; |
---|
5151 | while (currentPassNumber_<todo) { |
---|
5152 | |
---|
5153 | currentPassNumber_++; |
---|
5154 | numberSolutions_=0; |
---|
5155 | // this will be set false to break out of loop with presolved problem |
---|
5156 | bool doIntegerPresolve=(currentPassNumber_!=20); |
---|
5157 | |
---|
5158 | // Current number of free integer variables |
---|
5159 | // Get increment in solutions |
---|
5160 | { |
---|
5161 | const double * objective = cleanModel->getObjCoefficients(); |
---|
5162 | const double * lower = cleanModel->getColLower(); |
---|
5163 | const double * upper = cleanModel->getColUpper(); |
---|
5164 | double maximumCost=0.0; |
---|
5165 | bool possibleMultiple=true; |
---|
5166 | int numberChanged=0; |
---|
5167 | for (iColumn=0;iColumn<originalNumberColumns;iColumn++) { |
---|
5168 | if (originalUpper[iColumn]>originalLower[iColumn]) { |
---|
5169 | if( cleanModel->isInteger(iColumn)) { |
---|
5170 | maximumCost = CoinMax(maximumCost,fabs(objective[iColumn])); |
---|
5171 | } else if (objective[iColumn]) { |
---|
5172 | possibleMultiple=false; |
---|
5173 | } |
---|
5174 | } |
---|
5175 | if (originalUpper[iColumn]<upper[iColumn]) { |
---|
5176 | #ifdef CBC_DEBUG |
---|
5177 | printf("Changing upper bound on %d from %g to %g\n", |
---|
5178 | iColumn,upper[iColumn],originalUpper[iColumn]); |
---|
5179 | #endif |
---|
5180 | cleanModel->setColUpper(iColumn,originalUpper[iColumn]); |
---|
5181 | numberChanged++; |
---|
5182 | } |
---|
5183 | if (originalLower[iColumn]>lower[iColumn]) { |
---|
5184 | #ifdef CBC_DEBUG |
---|
5185 | printf("Changing lower bound on %d from %g to %g\n", |
---|
5186 | iColumn,lower[iColumn],originalLower[iColumn]); |
---|
5187 | #endif |
---|
5188 | cleanModel->setColLower(iColumn,originalLower[iColumn]); |
---|
5189 | numberChanged++; |
---|
5190 | } |
---|
5191 | } |
---|
5192 | // if first pass - always try |
---|
5193 | if (currentPassNumber_==1) |
---|
5194 | numberChanged += 1; |
---|
5195 | if (possibleMultiple&&maximumCost) { |
---|
5196 | int increment=0; |
---|
5197 | double multiplier = 2520.0; |
---|
5198 | while (10.0*multiplier*maximumCost<1.0e8) |
---|
5199 | multiplier *= 10.0; |
---|
5200 | for (int j =0;j<originalNumberIntegers;j++) { |
---|
5201 | iColumn = originalIntegers[j]; |
---|
5202 | if (originalUpper[iColumn]>originalLower[iColumn]) { |
---|
5203 | if(objective[iColumn]) { |
---|
5204 | double value = fabs(objective[iColumn])*multiplier; |
---|
5205 | int nearest = (int) floor(value+0.5); |
---|
5206 | if (fabs(value-floor(value+0.5))>1.0e-8) { |
---|
5207 | increment=0; |
---|
5208 | break; // no good |
---|
5209 | } else if (!increment) { |
---|
5210 | // first |
---|
5211 | increment=nearest; |
---|
5212 | } else { |
---|
5213 | increment = gcd(increment,nearest); |
---|
5214 | } |
---|
5215 | } |
---|
5216 | } |
---|
5217 | } |
---|
5218 | if (increment) { |
---|
5219 | double value = increment; |
---|
5220 | value /= multiplier; |
---|
5221 | if (value*0.999>dblParam_[CbcCutoffIncrement]) { |
---|
5222 | messageHandler()->message(CBC_INTEGERINCREMENT,messages()) |
---|
5223 | <<value |
---|
5224 | <<CoinMessageEol; |
---|
5225 | dblParam_[CbcCutoffIncrement]=value*0.999; |
---|
5226 | } |
---|
5227 | } |
---|
5228 | } |
---|
5229 | if (!numberChanged) { |
---|
5230 | doIntegerPresolve=false; // not doing any better |
---|
5231 | } |
---|
5232 | } |
---|
5233 | #ifdef CBC_DEBUG |
---|
5234 | if (debugger) |
---|
5235 | assert(debugger->onOptimalPath(*cleanModel)); |
---|
5236 | #endif |
---|
5237 | #ifdef COIN_USE_CLP |
---|
5238 | // do presolve - for now just clp but easy to get osi interface |
---|
5239 | OsiClpSolverInterface * clpSolver |
---|
5240 | = dynamic_cast<OsiClpSolverInterface *> (cleanModel); |
---|
5241 | if (clpSolver) { |
---|
5242 | ClpSimplex * clp = clpSolver->getModelPtr(); |
---|
5243 | clp->messageHandler()->setLogLevel(cleanModel->messageHandler()->logLevel()); |
---|
5244 | ClpPresolve pinfo; |
---|
5245 | //printf("integerPresolve - temp switch off doubletons\n"); |
---|
5246 | //pinfo.setPresolveActions(4); |
---|
5247 | ClpSimplex * model2 = pinfo.presolvedModel(*clp,1.0e-8); |
---|
5248 | if (!model2) { |
---|
5249 | // presolve found to be infeasible |
---|
5250 | feasible=false; |
---|
5251 | } else { |
---|
5252 | // update original array |
---|
5253 | const int * originalColumns = pinfo.originalColumns(); |
---|
5254 | // just slot in new solver |
---|
5255 | OsiClpSolverInterface * temp = new OsiClpSolverInterface(model2,true); |
---|
5256 | numberColumns = temp->getNumCols(); |
---|
5257 | for (iColumn=0;iColumn<originalNumberColumns;iColumn++) |
---|
5258 | original[iColumn]=-1; |
---|
5259 | for (iColumn=0;iColumn<numberColumns;iColumn++) |
---|
5260 | original[originalColumns[iColumn]]=iColumn; |
---|
5261 | // copy parameters |
---|
5262 | temp->copyParameters(*solver_); |
---|
5263 | // and specialized ones |
---|
5264 | temp->setSpecialOptions(clpSolver->specialOptions()); |
---|
5265 | delete solver_; |
---|
5266 | solver_ = temp; |
---|
5267 | setCutoff(cutoff); |
---|
5268 | deleteObjects(); |
---|
5269 | if (!numberObjects_) { |
---|
5270 | // Nothing left |
---|
5271 | doIntegerPresolve=false; |
---|
5272 | weak=true; |
---|
5273 | break; |
---|
5274 | } |
---|
5275 | synchronizeModel(); // make sure everything that needs solver has it |
---|
5276 | // just point to solver_ |
---|
5277 | continuousSolver_ = solver_; |
---|
5278 | feasible=resolve(); |
---|
5279 | if (!feasible||!doIntegerPresolve||weak) break; |
---|
5280 | // see if we can get solution by heuristics |
---|
5281 | int found=-1; |
---|
5282 | int iHeuristic; |
---|
5283 | double * newSolution = new double [numberColumns]; |
---|
5284 | double heuristicValue=getCutoff(); |
---|
5285 | for (iHeuristic=0;iHeuristic<numberHeuristics_;iHeuristic++) { |
---|
5286 | double saveValue=heuristicValue; |
---|
5287 | int ifSol = heuristic_[iHeuristic]->solution(heuristicValue, |
---|
5288 | newSolution); |
---|
5289 | if (ifSol>0) { |
---|
5290 | // better solution found |
---|
5291 | found=iHeuristic; |
---|
5292 | incrementUsed(newSolution); |
---|
5293 | } else if (ifSol<0) { |
---|
5294 | heuristicValue = saveValue; |
---|
5295 | } |
---|
5296 | } |
---|
5297 | if (found >= 0) { |
---|
5298 | // We probably already have a current solution, but just in case ... |
---|
5299 | int numberColumns = getNumCols() ; |
---|
5300 | if (!currentSolution_) |
---|
5301 | currentSolution_ = new double[numberColumns] ; |
---|
5302 | testSolution_=currentSolution_; |
---|
5303 | // better solution save |
---|
5304 | setBestSolution(CBC_ROUNDING,heuristicValue, |
---|
5305 | newSolution); |
---|
5306 | lastHeuristic_ = heuristic_[found]; |
---|
5307 | // update cutoff |
---|
5308 | cutoff = getCutoff(); |
---|
5309 | } |
---|
5310 | delete [] newSolution; |
---|
5311 | // Space for type of cuts |
---|
5312 | int maximumWhich=1000; |
---|
5313 | int * whichGenerator = new int[maximumWhich]; |
---|
5314 | // save number of rows |
---|
5315 | numberRowsAtContinuous_ = getNumRows(); |
---|
5316 | maximumNumberCuts_=0; |
---|
5317 | currentNumberCuts_=0; |
---|
5318 | delete [] addedCuts_; |
---|
5319 | addedCuts_ = NULL; |
---|
5320 | |
---|
5321 | // maximum depth for tree walkback |
---|
5322 | maximumDepth_=10; |
---|
5323 | delete [] walkback_; |
---|
5324 | walkback_ = new CbcNodeInfo * [maximumDepth_]; |
---|
5325 | |
---|
5326 | OsiCuts cuts; |
---|
5327 | int numberOldActiveCuts=0; |
---|
5328 | int numberNewCuts = 0; |
---|
5329 | feasible = solveWithCuts(cuts,maximumCutPassesAtRoot_, |
---|
5330 | NULL,numberOldActiveCuts,numberNewCuts, |
---|
5331 | maximumWhich, whichGenerator); |
---|
5332 | currentNumberCuts_=numberNewCuts; |
---|
5333 | delete [] whichGenerator; |
---|
5334 | delete [] walkback_; |
---|
5335 | walkback_ = NULL; |
---|
5336 | delete [] addedCuts_; |
---|
5337 | addedCuts_=NULL; |
---|
5338 | if (feasible) { |
---|
5339 | // fix anything in original which integer presolve fixed |
---|
5340 | // for now just integers |
---|
5341 | const double * lower = solver_->getColLower(); |
---|
5342 | const double * upper = solver_->getColUpper(); |
---|
5343 | int i; |
---|
5344 | for (i=0;i<originalNumberIntegers;i++) { |
---|
5345 | iColumn = originalIntegers[i]; |
---|
5346 | int jColumn = original[iColumn]; |
---|
5347 | if (jColumn >= 0) { |
---|
5348 | if (upper[jColumn]<originalUpper[iColumn]) |
---|
5349 | originalUpper[iColumn] = upper[jColumn]; |
---|
5350 | if (lower[jColumn]>originalLower[iColumn]) |
---|
5351 | originalLower[iColumn] = lower[jColumn]; |
---|
5352 | } |
---|
5353 | } |
---|
5354 | } |
---|
5355 | } |
---|
5356 | } |
---|
5357 | #endif |
---|
5358 | if (!feasible||!doIntegerPresolve) { |
---|
5359 | break; |
---|
5360 | } |
---|
5361 | } |
---|
5362 | //solver_->writeMps("xx"); |
---|
5363 | delete cleanModel; |
---|
5364 | delete [] originalIntegers; |
---|
5365 | numberColumns = getNumCols(); |
---|
5366 | delete [] originalColumns_; |
---|
5367 | originalColumns_ = new int[numberColumns]; |
---|
5368 | numberColumns=0; |
---|
5369 | for (iColumn=0;iColumn<originalNumberColumns;iColumn++) { |
---|
5370 | int jColumn = original[iColumn]; |
---|
5371 | if (jColumn >= 0) |
---|
5372 | originalColumns_[numberColumns++]=iColumn; |
---|
5373 | } |
---|
5374 | delete [] original; |
---|
5375 | delete [] originalLower; |
---|
5376 | delete [] originalUpper; |
---|
5377 | |
---|
5378 | deleteObjects(); |
---|
5379 | synchronizeModel(); // make sure everything that needs solver has it |
---|
5380 | continuousSolver_=NULL; |
---|
5381 | currentNumberCuts_=0; |
---|
5382 | return feasible; |
---|
5383 | } |
---|
5384 | // Put back information into original model - after integerpresolve |
---|
5385 | void |
---|
5386 | CbcModel::originalModel(CbcModel * presolvedModel,bool weak) |
---|
5387 | { |
---|
5388 | solver_->copyParameters(*(presolvedModel->solver_)); |
---|
5389 | bestObjective_ = presolvedModel->bestObjective_; |
---|
5390 | delete [] bestSolution_; |
---|
5391 | findIntegers(true); |
---|
5392 | if (presolvedModel->bestSolution_) { |
---|
5393 | int numberColumns = getNumCols(); |
---|
5394 | int numberOtherColumns = presolvedModel->getNumCols(); |
---|
5395 | //bestSolution_ = new double[numberColumns]; |
---|
5396 | // set up map |
---|
5397 | int * back = new int[numberColumns]; |
---|
5398 | int i; |
---|
5399 | for (i=0;i<numberColumns;i++) |
---|
5400 | back[i]=-1; |
---|
5401 | for (i=0;i<numberOtherColumns;i++) |
---|
5402 | back[presolvedModel->originalColumns_[i]]=i; |
---|
5403 | int iColumn; |
---|
5404 | // set ones in presolved model to values |
---|
5405 | double * otherSolution = presolvedModel->bestSolution_; |
---|
5406 | //const double * lower = getColLower(); |
---|
5407 | for (i=0;i<numberIntegers_;i++) { |
---|
5408 | iColumn = integerVariable_[i]; |
---|
5409 | int jColumn = back[iColumn]; |
---|
5410 | //bestSolution_[iColumn]=lower[iColumn]; |
---|
5411 | if (jColumn >= 0) { |
---|
5412 | double value=floor(otherSolution[jColumn]+0.5); |
---|
5413 | solver_->setColLower(iColumn,value); |
---|
5414 | solver_->setColUpper(iColumn,value); |
---|
5415 | //bestSolution_[iColumn]=value; |
---|
5416 | } |
---|
5417 | } |
---|
5418 | delete [] back; |
---|
5419 | #if 0 |
---|
5420 | // ** looks as if presolve needs more intelligence |
---|
5421 | // do presolve - for now just clp but easy to get osi interface |
---|
5422 | OsiClpSolverInterface * clpSolver |
---|
5423 | = dynamic_cast<OsiClpSolverInterface *> (solver_); |
---|
5424 | assert (clpSolver); |
---|
5425 | ClpSimplex * clp = clpSolver->getModelPtr(); |
---|
5426 | Presolve pinfo; |
---|
5427 | ClpSimplex * model2 = pinfo.presolvedModel(*clp,1.0e-8); |
---|
5428 | model2->primal(1); |
---|
5429 | pinfo.postsolve(true); |
---|
5430 | const double * solution = solver_->getColSolution(); |
---|
5431 | for (i=0;i<numberIntegers_;i++) { |
---|
5432 | iColumn = integerVariable_[i]; |
---|
5433 | double value=floor(solution[iColumn]+0.5); |
---|
5434 | solver_->setColLower(iColumn,value); |
---|
5435 | solver_->setColUpper(iColumn,value); |
---|
5436 | } |
---|
5437 | #else |
---|
5438 | if (!weak) { |
---|
5439 | // for now give up |
---|
5440 | int save = numberCutGenerators_; |
---|
5441 | numberCutGenerators_=0; |
---|
5442 | bestObjective_=1.0e100; |
---|
5443 | branchAndBound(); |
---|
5444 | numberCutGenerators_=save; |
---|
5445 | } |
---|
5446 | #endif |
---|
5447 | if (bestSolution_) { |
---|
5448 | // solve problem |
---|
5449 | resolve(); |
---|
5450 | // should be feasible |
---|
5451 | int numberIntegerInfeasibilities; |
---|
5452 | int numberObjectInfeasibilities; |
---|
5453 | if (!currentSolution_) |
---|
5454 | currentSolution_ = new double[numberColumns] ; |
---|
5455 | testSolution_ = currentSolution_; |
---|
5456 | assert(feasibleSolution(numberIntegerInfeasibilities, |
---|
5457 | numberObjectInfeasibilities)); |
---|
5458 | } |
---|
5459 | } else { |
---|
5460 | bestSolution_=NULL; |
---|
5461 | } |
---|
5462 | numberSolutions_=presolvedModel->numberSolutions_; |
---|
5463 | numberHeuristicSolutions_=presolvedModel->numberHeuristicSolutions_; |
---|
5464 | numberNodes_ = presolvedModel->numberNodes_; |
---|
5465 | numberIterations_ = presolvedModel->numberIterations_; |
---|
5466 | status_ = presolvedModel->status_; |
---|
5467 | secondaryStatus_ = presolvedModel->secondaryStatus_; |
---|
5468 | synchronizeModel(); |
---|
5469 | } |
---|
5470 | // Pass in Message handler (not deleted at end) |
---|
5471 | void |
---|
5472 | CbcModel::passInMessageHandler(CoinMessageHandler * handler) |
---|
5473 | { |
---|
5474 | if (defaultHandler_) { |
---|
5475 | delete handler_; |
---|
5476 | handler_ = NULL; |
---|
5477 | } |
---|
5478 | defaultHandler_=false; |
---|
5479 | handler_=handler; |
---|
5480 | } |
---|
5481 | void |
---|
5482 | CbcModel::passInTreeHandler(CbcTree & tree) |
---|
5483 | { |
---|
5484 | delete tree_; |
---|
5485 | tree_ = tree.clone(); |
---|
5486 | } |
---|
5487 | // Make sure region there |
---|
5488 | void |
---|
5489 | CbcModel::reserveCurrentSolution(const double * solution) |
---|
5490 | { |
---|
5491 | int numberColumns = getNumCols() ; |
---|
5492 | if (!currentSolution_) |
---|
5493 | currentSolution_ = new double[numberColumns] ; |
---|
5494 | testSolution_=currentSolution_; |
---|
5495 | if (solution) |
---|
5496 | memcpy(currentSolution_,solution,numberColumns*sizeof(double)); |
---|
5497 | } |
---|
5498 | /* For passing in an CbcModel to do a sub Tree (with derived tree handlers). |
---|
5499 | Passed in model must exist for duration of branch and bound |
---|
5500 | */ |
---|
5501 | void |
---|
5502 | CbcModel::passInSubTreeModel(CbcModel & model) |
---|
5503 | { |
---|
5504 | subTreeModel_=&model; |
---|
5505 | } |
---|
5506 | // For retrieving a copy of subtree model with given OsiSolver or NULL |
---|
5507 | CbcModel * |
---|
5508 | CbcModel::subTreeModel(OsiSolverInterface * solver) const |
---|
5509 | { |
---|
5510 | const CbcModel * subModel=subTreeModel_; |
---|
5511 | if (!subModel) |
---|
5512 | subModel=this; |
---|
5513 | // Get new copy |
---|
5514 | CbcModel * newModel = new CbcModel(*subModel); |
---|
5515 | if (solver) |
---|
5516 | newModel->assignSolver(solver); |
---|
5517 | return newModel; |
---|
5518 | } |
---|
5519 | //############################################################################# |
---|
5520 | // Set/Get Application Data |
---|
5521 | // This is a pointer that the application can store into and retrieve |
---|
5522 | // from the solverInterface. |
---|
5523 | // This field is the application to optionally define and use. |
---|
5524 | //############################################################################# |
---|
5525 | |
---|
5526 | void CbcModel::setApplicationData(void * appData) |
---|
5527 | { |
---|
5528 | appData_ = appData; |
---|
5529 | } |
---|
5530 | //----------------------------------------------------------------------------- |
---|
5531 | void * CbcModel::getApplicationData() const |
---|
5532 | { |
---|
5533 | return appData_; |
---|
5534 | } |
---|
5535 | /* create a submodel from partially fixed problem |
---|
5536 | |
---|
5537 | The method creates a new clean model with given bounds. |
---|
5538 | */ |
---|
5539 | CbcModel * |
---|
5540 | CbcModel::cleanModel(const double * lower, const double * upper) |
---|
5541 | { |
---|
5542 | OsiSolverInterface * solver = continuousSolver_->clone(); |
---|
5543 | |
---|
5544 | int numberIntegers = numberIntegers_; |
---|
5545 | const int * integerVariable = integerVariable_; |
---|
5546 | |
---|
5547 | int i; |
---|
5548 | for (i=0;i<numberIntegers;i++) { |
---|
5549 | int iColumn=integerVariable[i]; |
---|
5550 | const CbcObject * object = object_[i]; |
---|
5551 | const CbcSimpleInteger * integerObject = |
---|
5552 | dynamic_cast<const CbcSimpleInteger *> (object); |
---|
5553 | assert(integerObject); |
---|
5554 | // get original bounds |
---|
5555 | double originalLower = integerObject->originalLowerBound(); |
---|
5556 | double originalUpper = integerObject->originalUpperBound(); |
---|
5557 | solver->setColLower(iColumn,CoinMax(lower[iColumn],originalLower)); |
---|
5558 | solver->setColUpper(iColumn,CoinMin(upper[iColumn],originalUpper)); |
---|
5559 | } |
---|
5560 | CbcModel * model = new CbcModel(*solver); |
---|
5561 | // off some messages |
---|
5562 | if (handler_->logLevel()<=1) { |
---|
5563 | model->messagesPointer()->setDetailMessage(3,9); |
---|
5564 | model->messagesPointer()->setDetailMessage(3,6); |
---|
5565 | model->messagesPointer()->setDetailMessage(3,4); |
---|
5566 | model->messagesPointer()->setDetailMessage(3,1); |
---|
5567 | model->messagesPointer()->setDetailMessage(3,13); |
---|
5568 | model->messagesPointer()->setDetailMessage(3,14); |
---|
5569 | model->messagesPointer()->setDetailMessage(3,3007); |
---|
5570 | } |
---|
5571 | // Cuts |
---|
5572 | for ( i = 0;i<numberCutGenerators_;i++) { |
---|
5573 | int howOften = generator_[i]->howOftenInSub(); |
---|
5574 | if (howOften>-100) { |
---|
5575 | CbcCutGenerator * generator = virginGenerator_[i]; |
---|
5576 | CglCutGenerator * cglGenerator = generator->generator(); |
---|
5577 | model->addCutGenerator(cglGenerator,howOften, |
---|
5578 | generator->cutGeneratorName(), |
---|
5579 | generator->normal(), |
---|
5580 | generator->atSolution(), |
---|
5581 | generator->whenInfeasible(), |
---|
5582 | -100, generator->whatDepthInSub(),-1); |
---|
5583 | } |
---|
5584 | } |
---|
5585 | double cutoff = getCutoff(); |
---|
5586 | model->setCutoff(cutoff); |
---|
5587 | return model; |
---|
5588 | } |
---|
5589 | /* Invoke the branch & cut algorithm on partially fixed problem |
---|
5590 | |
---|
5591 | The method uses a subModel created by cleanModel. The search |
---|
5592 | ends when the tree is exhausted or maximum nodes is reached. |
---|
5593 | |
---|
5594 | If better solution found then it is saved. |
---|
5595 | |
---|
5596 | Returns 0 if search completed and solution, 1 if not completed and solution, |
---|
5597 | 2 if completed and no solution, 3 if not completed and no solution. |
---|
5598 | |
---|
5599 | Normally okay to do subModel immediately followed by subBranchandBound |
---|
5600 | (== other form of subBranchAndBound) |
---|
5601 | but may need to get at model for advanced features. |
---|
5602 | |
---|
5603 | Deletes model |
---|
5604 | |
---|
5605 | */ |
---|
5606 | |
---|
5607 | int |
---|
5608 | CbcModel::subBranchAndBound(CbcModel * model, |
---|
5609 | CbcModel * presolvedModel, |
---|
5610 | int maximumNodes) |
---|
5611 | { |
---|
5612 | int i; |
---|
5613 | double cutoff=model->getCutoff(); |
---|
5614 | CbcModel * model2; |
---|
5615 | if (presolvedModel) |
---|
5616 | model2=presolvedModel; |
---|
5617 | else |
---|
5618 | model2=model; |
---|
5619 | // Do complete search |
---|
5620 | |
---|
5621 | for (i=0;i<numberHeuristics_;i++) { |
---|
5622 | model2->addHeuristic(heuristic_[i]); |
---|
5623 | model2->heuristic(i)->resetModel(model2); |
---|
5624 | } |
---|
5625 | // Definition of node choice |
---|
5626 | model2->setNodeComparison(nodeCompare_->clone()); |
---|
5627 | //model2->solver()->setHintParam(OsiDoReducePrint,true,OsiHintTry); |
---|
5628 | model2->messageHandler()->setLogLevel(CoinMax(0,handler_->logLevel()-1)); |
---|
5629 | //model2->solver()->messageHandler()->setLogLevel(2); |
---|
5630 | model2->setMaximumCutPassesAtRoot(maximumCutPassesAtRoot_); |
---|
5631 | model2->setPrintFrequency(50); |
---|
5632 | model2->setIntParam(CbcModel::CbcMaxNumNode,maximumNodes); |
---|
5633 | model2->branchAndBound(); |
---|
5634 | delete model2->nodeComparison(); |
---|
5635 | if (model2->getMinimizationObjValue()>cutoff) { |
---|
5636 | // no good |
---|
5637 | if (model!=model2) |
---|
5638 | delete model2; |
---|
5639 | delete model; |
---|
5640 | return 2; |
---|
5641 | } |
---|
5642 | if (model!=model2) { |
---|
5643 | // get back solution |
---|
5644 | model->originalModel(model2,false); |
---|
5645 | delete model2; |
---|
5646 | } |
---|
5647 | int status; |
---|
5648 | if (model->getMinimizationObjValue()<cutoff&&model->bestSolution()) { |
---|
5649 | double objValue = model->getObjValue(); |
---|
5650 | const double * solution = model->bestSolution(); |
---|
5651 | setBestSolution(CBC_TREE_SOL,objValue,solution); |
---|
5652 | status = 0; |
---|
5653 | } else { |
---|
5654 | status=2; |
---|
5655 | } |
---|
5656 | if (model->status()) |
---|
5657 | status ++ ; // not finished search |
---|
5658 | delete model; |
---|
5659 | return status; |
---|
5660 | } |
---|
5661 | /* Invoke the branch & cut algorithm on partially fixed problem |
---|
5662 | |
---|
5663 | The method creates a new model with given bounds, presolves it |
---|
5664 | then proceeds to explore the branch & cut search tree. The search |
---|
5665 | ends when the tree is exhausted or maximum nodes is reached. |
---|
5666 | Returns 0 if search completed and solution, 1 if not completed and solution, |
---|
5667 | 2 if completed and no solution, 3 if not completed and no solution. |
---|
5668 | */ |
---|
5669 | int |
---|
5670 | CbcModel::subBranchAndBound(const double * lower, const double * upper, |
---|
5671 | int maximumNodes) |
---|
5672 | { |
---|
5673 | OsiSolverInterface * solver = continuousSolver_->clone(); |
---|
5674 | |
---|
5675 | int numberIntegers = numberIntegers_; |
---|
5676 | const int * integerVariable = integerVariable_; |
---|
5677 | |
---|
5678 | int i; |
---|
5679 | for (i=0;i<numberIntegers;i++) { |
---|
5680 | int iColumn=integerVariable[i]; |
---|
5681 | const CbcObject * object = object_[i]; |
---|
5682 | const CbcSimpleInteger * integerObject = |
---|
5683 | dynamic_cast<const CbcSimpleInteger *> (object); |
---|
5684 | assert(integerObject); |
---|
5685 | // get original bounds |
---|
5686 | double originalLower = integerObject->originalLowerBound(); |
---|
5687 | double originalUpper = integerObject->originalUpperBound(); |
---|
5688 | solver->setColLower(iColumn,CoinMax(lower[iColumn],originalLower)); |
---|
5689 | solver->setColUpper(iColumn,CoinMin(upper[iColumn],originalUpper)); |
---|
5690 | } |
---|
5691 | CbcModel model(*solver); |
---|
5692 | // off some messages |
---|
5693 | if (handler_->logLevel()<=1) { |
---|
5694 | model.messagesPointer()->setDetailMessage(3,9); |
---|
5695 | model.messagesPointer()->setDetailMessage(3,6); |
---|
5696 | model.messagesPointer()->setDetailMessage(3,4); |
---|
5697 | model.messagesPointer()->setDetailMessage(3,1); |
---|
5698 | model.messagesPointer()->setDetailMessage(3,3007); |
---|
5699 | } |
---|
5700 | double cutoff = getCutoff(); |
---|
5701 | model.setCutoff(cutoff); |
---|
5702 | // integer presolve |
---|
5703 | CbcModel * model2 = model.integerPresolve(false); |
---|
5704 | if (!model2||!model2->getNumRows()) { |
---|
5705 | delete model2; |
---|
5706 | delete solver; |
---|
5707 | return 2; |
---|
5708 | } |
---|
5709 | if (handler_->logLevel()>1) |
---|
5710 | printf("Reduced model has %d rows and %d columns\n", |
---|
5711 | model2->getNumRows(),model2->getNumCols()); |
---|
5712 | // Do complete search |
---|
5713 | |
---|
5714 | // Cuts |
---|
5715 | for ( i = 0;i<numberCutGenerators_;i++) { |
---|
5716 | int howOften = generator_[i]->howOftenInSub(); |
---|
5717 | if (howOften>-100) { |
---|
5718 | CbcCutGenerator * generator = virginGenerator_[i]; |
---|
5719 | CglCutGenerator * cglGenerator = generator->generator(); |
---|
5720 | model2->addCutGenerator(cglGenerator,howOften, |
---|
5721 | generator->cutGeneratorName(), |
---|
5722 | generator->normal(), |
---|
5723 | generator->atSolution(), |
---|
5724 | generator->whenInfeasible(), |
---|
5725 | -100, generator->whatDepthInSub(),-1); |
---|
5726 | } |
---|
5727 | } |
---|
5728 | for (i=0;i<numberHeuristics_;i++) { |
---|
5729 | model2->addHeuristic(heuristic_[i]); |
---|
5730 | model2->heuristic(i)->resetModel(model2); |
---|
5731 | } |
---|
5732 | // Definition of node choice |
---|
5733 | model2->setNodeComparison(nodeCompare_->clone()); |
---|
5734 | //model2->solver()->setHintParam(OsiDoReducePrint,true,OsiHintTry); |
---|
5735 | model2->messageHandler()->setLogLevel(CoinMax(0,handler_->logLevel()-1)); |
---|
5736 | //model2->solver()->messageHandler()->setLogLevel(2); |
---|
5737 | model2->setMaximumCutPassesAtRoot(maximumCutPassesAtRoot_); |
---|
5738 | model2->setPrintFrequency(50); |
---|
5739 | model2->setIntParam(CbcModel::CbcMaxNumNode,maximumNodes); |
---|
5740 | model2->branchAndBound(); |
---|
5741 | delete model2->nodeComparison(); |
---|
5742 | if (model2->getMinimizationObjValue()>cutoff) { |
---|
5743 | // no good |
---|
5744 | delete model2; |
---|
5745 | delete solver; |
---|
5746 | return 2; |
---|
5747 | } |
---|
5748 | // get back solution |
---|
5749 | model.originalModel(model2,false); |
---|
5750 | delete model2; |
---|
5751 | int status; |
---|
5752 | if (model.getMinimizationObjValue()<cutoff&&model.bestSolution()) { |
---|
5753 | double objValue = model.getObjValue(); |
---|
5754 | const double * solution = model.bestSolution(); |
---|
5755 | setBestSolution(CBC_TREE_SOL,objValue,solution); |
---|
5756 | status = 0; |
---|
5757 | } else { |
---|
5758 | status=2; |
---|
5759 | } |
---|
5760 | if (model.status()) |
---|
5761 | status ++ ; // not finished search |
---|
5762 | delete solver; |
---|
5763 | return status; |
---|
5764 | } |
---|
5765 | // Set a pointer to a row cut which will be added instead of normal branching. |
---|
5766 | void |
---|
5767 | CbcModel::setNextRowCut(const OsiRowCut & cut) |
---|
5768 | { |
---|
5769 | nextRowCut_=new OsiRowCut(cut); |
---|
5770 | nextRowCut_->setEffectiveness(COIN_DBL_MAX); // mark so will always stay |
---|
5771 | } |
---|
5772 | /* Process root node and return a strengthened model |
---|
5773 | |
---|
5774 | The method assumes that initialSolve() has been called to solve the |
---|
5775 | LP relaxation. It processes the root node and then returns a pointer |
---|
5776 | to the strengthened model (or NULL if infeasible) |
---|
5777 | */ |
---|
5778 | OsiSolverInterface * |
---|
5779 | CbcModel::strengthenedModel() |
---|
5780 | { |
---|
5781 | /* |
---|
5782 | Assume we're done, and see if we're proven wrong. |
---|
5783 | */ |
---|
5784 | /* |
---|
5785 | Scan the variables, noting the integer variables. Create an |
---|
5786 | CbcSimpleInteger object for each integer variable. |
---|
5787 | */ |
---|
5788 | findIntegers(false) ; |
---|
5789 | /* |
---|
5790 | Ensure that objects on the lists of CbcObjects, heuristics, and cut |
---|
5791 | generators attached to this model all refer to this model. |
---|
5792 | */ |
---|
5793 | synchronizeModel() ; |
---|
5794 | |
---|
5795 | // Set so we can tell we are in initial phase in resolve |
---|
5796 | continuousObjective_ = -COIN_DBL_MAX ; |
---|
5797 | /* |
---|
5798 | Solve the relaxation. |
---|
5799 | |
---|
5800 | Apparently there are circumstances where this will be non-trivial --- i.e., |
---|
5801 | we've done something since initialSolve that's trashed the solution to the |
---|
5802 | continuous relaxation. |
---|
5803 | */ |
---|
5804 | bool feasible = resolve() ; |
---|
5805 | /* |
---|
5806 | If the linear relaxation of the root is infeasible, bail out now. Otherwise, |
---|
5807 | continue with processing the root node. |
---|
5808 | */ |
---|
5809 | if (!feasible) |
---|
5810 | { handler_->message(CBC_INFEAS,messages_)<< CoinMessageEol ; |
---|
5811 | return NULL; } |
---|
5812 | // Save objective (just so user can access it) |
---|
5813 | originalContinuousObjective_ = solver_->getObjValue(); |
---|
5814 | |
---|
5815 | /* |
---|
5816 | Begin setup to process a feasible root node. |
---|
5817 | */ |
---|
5818 | bestObjective_ = CoinMin(bestObjective_,1.0e50) ; |
---|
5819 | numberSolutions_ = 0 ; |
---|
5820 | numberHeuristicSolutions_ = 0 ; |
---|
5821 | // Everything is minimization |
---|
5822 | double cutoff=getCutoff() ; |
---|
5823 | double direction = solver_->getObjSense() ; |
---|
5824 | if (cutoff < 1.0e20&&direction<0.0) |
---|
5825 | messageHandler()->message(CBC_CUTOFF_WARNING1, |
---|
5826 | messages()) |
---|
5827 | << cutoff << -cutoff << CoinMessageEol ; |
---|
5828 | if (cutoff > bestObjective_) |
---|
5829 | cutoff = bestObjective_ ; |
---|
5830 | setCutoff(cutoff) ; |
---|
5831 | /* |
---|
5832 | We probably already have a current solution, but just in case ... |
---|
5833 | */ |
---|
5834 | int numberColumns = getNumCols() ; |
---|
5835 | if (!currentSolution_) |
---|
5836 | currentSolution_ = new double[numberColumns] ; |
---|
5837 | testSolution_=currentSolution_; |
---|
5838 | /* |
---|
5839 | Create a copy of the solver, thus capturing the original (root node) |
---|
5840 | constraint system (aka the continuous system). |
---|
5841 | */ |
---|
5842 | continuousSolver_ = solver_->clone() ; |
---|
5843 | numberRowsAtContinuous_ = getNumRows() ; |
---|
5844 | /* |
---|
5845 | Check the objective to see if we can deduce a nontrivial increment. If |
---|
5846 | it's better than the current value for CbcCutoffIncrement, it'll be |
---|
5847 | installed. |
---|
5848 | */ |
---|
5849 | analyzeObjective() ; |
---|
5850 | /* |
---|
5851 | Set up for cut generation. addedCuts_ holds the cuts which are relevant for |
---|
5852 | the active subproblem. whichGenerator will be used to record the generator |
---|
5853 | that produced a given cut. |
---|
5854 | */ |
---|
5855 | int maximumWhich = 1000 ; |
---|
5856 | int * whichGenerator = new int[maximumWhich] ; |
---|
5857 | maximumNumberCuts_ = 0 ; |
---|
5858 | currentNumberCuts_ = 0 ; |
---|
5859 | delete [] addedCuts_ ; |
---|
5860 | addedCuts_ = NULL ; |
---|
5861 | /* |
---|
5862 | Generate cuts at the root node and reoptimise. solveWithCuts does the heavy |
---|
5863 | lifting. It will iterate a generate/reoptimise loop (including reduced cost |
---|
5864 | fixing) until no cuts are generated, the change in objective falls off, or |
---|
5865 | the limit on the number of rounds of cut generation is exceeded. |
---|
5866 | |
---|
5867 | At the end of all this, any cuts will be recorded in cuts and also |
---|
5868 | installed in the solver's constraint system. We'll have reoptimised, and |
---|
5869 | removed any slack cuts (numberOldActiveCuts and numberNewCuts have been |
---|
5870 | adjusted accordingly). |
---|
5871 | |
---|
5872 | Tell cut generators they can be a bit more aggressive at root node |
---|
5873 | |
---|
5874 | */ |
---|
5875 | int iCutGenerator; |
---|
5876 | for (iCutGenerator = 0;iCutGenerator<numberCutGenerators_;iCutGenerator++) { |
---|
5877 | CglCutGenerator * generator = generator_[iCutGenerator]->generator(); |
---|
5878 | generator->setAggressiveness(generator->getAggressiveness()+100); |
---|
5879 | } |
---|
5880 | OsiCuts cuts ; |
---|
5881 | int numberOldActiveCuts = 0 ; |
---|
5882 | int numberNewCuts = 0 ; |
---|
5883 | { int iObject ; |
---|
5884 | int preferredWay ; |
---|
5885 | int numberUnsatisfied = 0 ; |
---|
5886 | memcpy(currentSolution_,solver_->getColSolution(), |
---|
5887 | numberColumns*sizeof(double)) ; |
---|
5888 | |
---|
5889 | for (iObject = 0 ; iObject < numberObjects_ ; iObject++) |
---|
5890 | { double infeasibility = |
---|
5891 | object_[iObject]->infeasibility(preferredWay) ; |
---|
5892 | if (infeasibility) numberUnsatisfied++ ; } |
---|
5893 | if (numberUnsatisfied) |
---|
5894 | { feasible = solveWithCuts(cuts,maximumCutPassesAtRoot_, |
---|
5895 | NULL,numberOldActiveCuts,numberNewCuts, |
---|
5896 | maximumWhich, whichGenerator) ; } } |
---|
5897 | /* |
---|
5898 | We've taken the continuous relaxation as far as we can. |
---|
5899 | */ |
---|
5900 | |
---|
5901 | OsiSolverInterface * newSolver=NULL; |
---|
5902 | if (feasible) { |
---|
5903 | // make copy of current solver |
---|
5904 | newSolver = solver_->clone(); |
---|
5905 | } |
---|
5906 | /* |
---|
5907 | Clean up dangling objects. continuousSolver_ may already be toast. |
---|
5908 | */ |
---|
5909 | delete [] whichGenerator ; |
---|
5910 | delete [] walkback_ ; |
---|
5911 | walkback_ = NULL ; |
---|
5912 | delete [] addedCuts_ ; |
---|
5913 | addedCuts_ = NULL ; |
---|
5914 | if (continuousSolver_) |
---|
5915 | { delete continuousSolver_ ; |
---|
5916 | continuousSolver_ = NULL ; } |
---|
5917 | /* |
---|
5918 | Destroy global cuts by replacing with an empty OsiCuts object. |
---|
5919 | */ |
---|
5920 | globalCuts_= OsiCuts() ; |
---|
5921 | |
---|
5922 | return newSolver; |
---|
5923 | } |
---|
5924 | // Just update objectiveValue |
---|
5925 | void CbcModel::setBestObjectiveValue( double objectiveValue) |
---|
5926 | { |
---|
5927 | bestObjective_=objectiveValue; |
---|
5928 | } |
---|
5929 | double |
---|
5930 | CbcModel::getBestPossibleObjValue() const |
---|
5931 | { |
---|
5932 | return CoinMin(bestPossibleObjective_,bestObjective_) * solver_->getObjSense() ; |
---|
5933 | } |
---|
5934 | // Make given rows (L or G) into global cuts and remove from lp |
---|
5935 | void |
---|
5936 | CbcModel::makeGlobalCuts(int number,const int * which) |
---|
5937 | { |
---|
5938 | const double * rowLower = solver_->getRowLower(); |
---|
5939 | const double * rowUpper = solver_->getRowUpper(); |
---|
5940 | |
---|
5941 | int numberRows = solver_->getNumRows(); |
---|
5942 | |
---|
5943 | // Row copy |
---|
5944 | const double * elementByRow = solver_->getMatrixByRow()->getElements(); |
---|
5945 | const int * column = solver_->getMatrixByRow()->getIndices(); |
---|
5946 | const CoinBigIndex * rowStart = solver_->getMatrixByRow()->getVectorStarts(); |
---|
5947 | const int * rowLength = solver_->getMatrixByRow()->getVectorLengths(); |
---|
5948 | |
---|
5949 | // Not all rows may be good so we need new array |
---|
5950 | int * whichDelete = new int[numberRows]; |
---|
5951 | int nDelete=0; |
---|
5952 | for (int i=0;i<number;i++) { |
---|
5953 | int iRow = which[i]; |
---|
5954 | if (iRow>=0 |
---|