1 | // Copyright (C) 2002, International Business Machines |
---|
2 | // Corporation and others. All Rights Reserved. |
---|
3 | |
---|
4 | |
---|
5 | /* Notes on implementation of dual simplex algorithm. |
---|
6 | |
---|
7 | When dual feasible: |
---|
8 | |
---|
9 | If primal feasible, we are optimal. Otherwise choose an infeasible |
---|
10 | basic variable to leave basis (normally going to nearest bound) (B). We |
---|
11 | now need to find an incoming variable which will leave problem |
---|
12 | dual feasible so we get the row of the tableau corresponding to |
---|
13 | the basic variable (with the correct sign depending if basic variable |
---|
14 | above or below feasibility region - as that affects whether reduced |
---|
15 | cost on outgoing variable has to be positive or negative). |
---|
16 | |
---|
17 | We now perform a ratio test to determine which incoming variable will |
---|
18 | preserve dual feasibility (C). If no variable found then problem |
---|
19 | is infeasible (in primal sense). If there is a variable, we then |
---|
20 | perform pivot and repeat. Trivial? |
---|
21 | |
---|
22 | ------------------------------------------- |
---|
23 | |
---|
24 | A) How do we get dual feasible? If all variables have bounds then |
---|
25 | it is trivial to get feasible by putting non-basic variables to |
---|
26 | correct bounds. OSL did not have a phase 1/phase 2 approach but |
---|
27 | instead effectively put fake bounds on variables and this is the |
---|
28 | approach here, although I had hoped to make it cleaner. |
---|
29 | |
---|
30 | If there is a weight of X on getting dual feasible: |
---|
31 | Non-basic variables with negative reduced costs are put to |
---|
32 | lesser of their upper bound and their lower bound + X. |
---|
33 | Similarly, mutatis mutandis, for positive reduced costs. |
---|
34 | |
---|
35 | Free variables should normally be in basis, otherwise I have |
---|
36 | coding which may be able to come out (and may not be correct). |
---|
37 | |
---|
38 | In OSL, this weight was changed heuristically, here at present |
---|
39 | it is only increased if problem looks finished. If problem is |
---|
40 | feasible I check for unboundedness. If not unbounded we |
---|
41 | could play with going into primal. As long as weights increase |
---|
42 | any algorithm would be finite. |
---|
43 | |
---|
44 | B) Which outgoing variable to choose is a virtual base class. |
---|
45 | For difficult problems steepest edge is preferred while for |
---|
46 | very easy (large) problems we will need partial scan. |
---|
47 | |
---|
48 | C) Sounds easy, but this is hardest part of algorithm. |
---|
49 | 1) Instead of stopping at first choice, we may be able |
---|
50 | to flip that variable to other bound and if objective |
---|
51 | still improving choose again. These mini iterations can |
---|
52 | increase speed by orders of magnitude but we may need to |
---|
53 | go to more of a bucket choice of variable rather than looking |
---|
54 | at them one by one (for speed). |
---|
55 | 2) Accuracy. Reduced costs may be of wrong sign but less than |
---|
56 | tolerance. Pivoting on these makes objective go backwards. |
---|
57 | OSL modified cost so a zero move was made, Gill et al |
---|
58 | (in primal analogue) modified so a strictly positive move was |
---|
59 | made. It is not quite as neat in dual but that is what we |
---|
60 | try and do. The two problems are that re-factorizations can |
---|
61 | change reduced costs above and below tolerances and that when |
---|
62 | finished we need to reset costs and try again. |
---|
63 | 3) Degeneracy. Gill et al helps but may not be enough. We |
---|
64 | may need more. Also it can improve speed a lot if we perturb |
---|
65 | the costs significantly. |
---|
66 | |
---|
67 | References: |
---|
68 | Forrest and Goldfarb, Steepest-edge simplex algorithms for |
---|
69 | linear programming - Mathematical Programming 1992 |
---|
70 | Forrest and Tomlin, Implementing the simplex method for |
---|
71 | the Optimization Subroutine Library - IBM Systems Journal 1992 |
---|
72 | Gill, Murray, Saunders, Wright A Practical Anti-Cycling |
---|
73 | Procedure for Linear and Nonlinear Programming SOL report 1988 |
---|
74 | |
---|
75 | |
---|
76 | TODO: |
---|
77 | |
---|
78 | a) Better recovery procedures. At present I never check on forward |
---|
79 | progress. There is checkpoint/restart with reducing |
---|
80 | re-factorization frequency, but this is only on singular |
---|
81 | factorizations. |
---|
82 | b) Fast methods for large easy problems (and also the option for |
---|
83 | the code to automatically choose which method). |
---|
84 | c) We need to be able to stop in various ways for OSI - this |
---|
85 | is fairly easy. |
---|
86 | |
---|
87 | */ |
---|
88 | |
---|
89 | #if defined(_MSC_VER) |
---|
90 | // Turn off compiler warning about long names |
---|
91 | # pragma warning(disable:4786) |
---|
92 | #endif |
---|
93 | |
---|
94 | #include <math.h> |
---|
95 | |
---|
96 | #include "CoinHelperFunctions.hpp" |
---|
97 | #include "ClpSimplexDual.hpp" |
---|
98 | #include "ClpFactorization.hpp" |
---|
99 | #include "CoinPackedMatrix.hpp" |
---|
100 | #include "CoinIndexedVector.hpp" |
---|
101 | #include "CoinWarmStartBasis.hpp" |
---|
102 | #include "ClpDualRowDantzig.hpp" |
---|
103 | #include "ClpMessage.hpp" |
---|
104 | #include <cfloat> |
---|
105 | #include <cassert> |
---|
106 | #include <string> |
---|
107 | #include <stdio.h> |
---|
108 | #include <iostream> |
---|
109 | // This returns a non const array filled with input from scalar |
---|
110 | // or actual array |
---|
111 | template <class T> inline T* |
---|
112 | copyOfArray( const T * array, const int size, T value) |
---|
113 | { |
---|
114 | T * arrayNew = new T[size]; |
---|
115 | if (array) |
---|
116 | CoinDisjointCopyN(array,size,arrayNew); |
---|
117 | else |
---|
118 | CoinFillN ( arrayNew, size,value); |
---|
119 | return arrayNew; |
---|
120 | } |
---|
121 | |
---|
122 | // This returns a non const array filled with actual array (or NULL) |
---|
123 | template <class T> inline T* |
---|
124 | copyOfArray( const T * array, const int size) |
---|
125 | { |
---|
126 | if (array) { |
---|
127 | T * arrayNew = new T[size]; |
---|
128 | CoinDisjointCopyN(array,size,arrayNew); |
---|
129 | return arrayNew; |
---|
130 | } else { |
---|
131 | return NULL; |
---|
132 | } |
---|
133 | } |
---|
134 | // dual |
---|
135 | int ClpSimplexDual::dual ( ) |
---|
136 | { |
---|
137 | |
---|
138 | /* *** Method |
---|
139 | This is a vanilla version of dual simplex. |
---|
140 | |
---|
141 | It tries to be a single phase approach with a weight of 1.0 being |
---|
142 | given to getting optimal and a weight of dualBound_ being |
---|
143 | given to getting dual feasible. In this version I have used the |
---|
144 | idea that this weight can be thought of as a fake bound. If the |
---|
145 | distance between the lower and upper bounds on a variable is less |
---|
146 | than the feasibility weight then we are always better off flipping |
---|
147 | to other bound to make dual feasible. If the distance is greater |
---|
148 | then we make up a fake bound dualBound_ away from one bound. |
---|
149 | If we end up optimal or primal infeasible, we check to see if |
---|
150 | bounds okay. If so we have finished, if not we increase dualBound_ |
---|
151 | and continue (after checking if unbounded). I am undecided about |
---|
152 | free variables - there is coding but I am not sure about it. At |
---|
153 | present I put them in basis anyway. |
---|
154 | |
---|
155 | The code is designed to take advantage of sparsity so arrays are |
---|
156 | seldom zeroed out from scratch or gone over in their entirety. |
---|
157 | The only exception is a full scan to find outgoing variable. This |
---|
158 | will be changed to keep an updated list of infeasibilities (or squares |
---|
159 | if steepest edge). Also on easy problems we don't need full scan - just |
---|
160 | pick first reasonable. |
---|
161 | |
---|
162 | One problem is how to tackle degeneracy and accuracy. At present |
---|
163 | I am using the modification of costs which I put in OSL and which was |
---|
164 | extended by Gill et al. I am still not sure of the exact details. |
---|
165 | |
---|
166 | The flow of dual is three while loops as follows: |
---|
167 | |
---|
168 | while (not finished) { |
---|
169 | |
---|
170 | while (not clean solution) { |
---|
171 | |
---|
172 | Factorize and/or clean up solution by flipping variables so |
---|
173 | dual feasible. If looks finished check fake dual bounds. |
---|
174 | Repeat until status is iterating (-1) or finished (0,1,2) |
---|
175 | |
---|
176 | } |
---|
177 | |
---|
178 | while (status==-1) { |
---|
179 | |
---|
180 | Iterate until no pivot in or out or time to re-factorize. |
---|
181 | |
---|
182 | Flow is: |
---|
183 | |
---|
184 | choose pivot row (outgoing variable). if none then |
---|
185 | we are primal feasible so looks as if done but we need to |
---|
186 | break and check bounds etc. |
---|
187 | |
---|
188 | Get pivot row in tableau |
---|
189 | |
---|
190 | Choose incoming column. If we don't find one then we look |
---|
191 | primal infeasible so break and check bounds etc. (Also the |
---|
192 | pivot tolerance is larger after any iterations so that may be |
---|
193 | reason) |
---|
194 | |
---|
195 | If we do find incoming column, we may have to adjust costs to |
---|
196 | keep going forwards (anti-degeneracy). Check pivot will be stable |
---|
197 | and if unstable throw away iteration (we will need to implement |
---|
198 | flagging of basic variables sometime) and break to re-factorize. |
---|
199 | If minor error re-factorize after iteration. |
---|
200 | |
---|
201 | Update everything (this may involve flipping variables to stay |
---|
202 | dual feasible. |
---|
203 | |
---|
204 | } |
---|
205 | |
---|
206 | } |
---|
207 | |
---|
208 | At present we never check we are going forwards. I overdid that in |
---|
209 | OSL so will try and make a last resort. |
---|
210 | |
---|
211 | Needs partial scan pivot out option. |
---|
212 | Needs dantzig, uninitialized and full steepest edge options (can still |
---|
213 | use partial scan) |
---|
214 | |
---|
215 | May need other anti-degeneracy measures, especially if we try and use |
---|
216 | loose tolerances as a way to solve in fewer iterations. |
---|
217 | |
---|
218 | I like idea of dynamic scaling. This gives opportunity to decouple |
---|
219 | different implications of scaling for accuracy, iteration count and |
---|
220 | feasibility tolerance. |
---|
221 | |
---|
222 | */ |
---|
223 | |
---|
224 | |
---|
225 | // sanity check |
---|
226 | assert (numberRows_==matrix_->getNumRows()); |
---|
227 | assert (numberColumns_==matrix_->getNumCols()); |
---|
228 | // for moment all arrays must exist |
---|
229 | assert(columnLower_); |
---|
230 | assert(columnUpper_); |
---|
231 | assert(rowLower_); |
---|
232 | assert(rowUpper_); |
---|
233 | |
---|
234 | #ifdef CLP_DEBUG |
---|
235 | int debugIteration=-1; |
---|
236 | #endif |
---|
237 | |
---|
238 | algorithm_ = -1; |
---|
239 | dualTolerance_=dblParam_[ClpDualTolerance]; |
---|
240 | primalTolerance_=dblParam_[ClpPrimalTolerance]; |
---|
241 | |
---|
242 | // put in standard form (and make row copy) |
---|
243 | // create modifiable copies of model rim and do optional scaling |
---|
244 | createRim(7+8+16,true); |
---|
245 | |
---|
246 | // save dual bound |
---|
247 | double saveDualBound_ = dualBound_; |
---|
248 | // save perturbation |
---|
249 | int savePerturbation = perturbation_; |
---|
250 | // save if sparse factorization wanted |
---|
251 | int saveSparse = factorization_->sparseThreshold(); |
---|
252 | |
---|
253 | if (sanityCheck()) { |
---|
254 | // Problem looks okay |
---|
255 | int iRow,iColumn; |
---|
256 | // Do initial factorization |
---|
257 | // and set certain stuff |
---|
258 | // We can either set increasing rows so ...IsBasic gives pivot row |
---|
259 | // or we can just increment iBasic one by one |
---|
260 | // for now let ...iBasic give pivot row |
---|
261 | factorization_->increasingRows(2); |
---|
262 | // row activities have negative sign |
---|
263 | factorization_->slackValue(-1.0); |
---|
264 | factorization_->zeroTolerance(1.0e-13); |
---|
265 | |
---|
266 | int factorizationStatus = internalFactorize(0); |
---|
267 | if (factorizationStatus<0) |
---|
268 | return 1; // some error |
---|
269 | else if (factorizationStatus) |
---|
270 | handler_->message(CLP_SINGULARITIES,messages_) |
---|
271 | <<factorizationStatus |
---|
272 | <<CoinMessageEol; |
---|
273 | |
---|
274 | // If user asked for perturbation - do it |
---|
275 | |
---|
276 | if (perturbation_<100) |
---|
277 | perturb(); |
---|
278 | |
---|
279 | double objectiveChange; |
---|
280 | // for dual we will change bounds using dualBound_ |
---|
281 | // for this we need clean basis so it is after factorize |
---|
282 | gutsOfSolution(rowActivityWork_,columnActivityWork_); |
---|
283 | |
---|
284 | numberFake_ =0; // Number of variables at fake bounds |
---|
285 | changeBounds(true,NULL,objectiveChange); |
---|
286 | |
---|
287 | problemStatus_ = -1; |
---|
288 | numberIterations_=0; |
---|
289 | |
---|
290 | int lastCleaned=0; // last time objective or bounds cleaned up |
---|
291 | |
---|
292 | // number of times we have declared optimality |
---|
293 | numberTimesOptimal_=0; |
---|
294 | |
---|
295 | // Progress indicator |
---|
296 | ClpSimplexProgress progress(this); |
---|
297 | |
---|
298 | // This says whether to restore things etc |
---|
299 | int factorType=0; |
---|
300 | /* |
---|
301 | Status of problem: |
---|
302 | 0 - optimal |
---|
303 | 1 - infeasible |
---|
304 | 2 - unbounded |
---|
305 | -1 - iterating |
---|
306 | -2 - factorization wanted |
---|
307 | -3 - redo checking without factorization |
---|
308 | -4 - looks infeasible |
---|
309 | */ |
---|
310 | while (problemStatus_<0) { |
---|
311 | // clear |
---|
312 | for (iRow=0;iRow<4;iRow++) { |
---|
313 | rowArray_[iRow]->clear(); |
---|
314 | } |
---|
315 | |
---|
316 | for (iColumn=0;iColumn<2;iColumn++) { |
---|
317 | columnArray_[iColumn]->clear(); |
---|
318 | } |
---|
319 | |
---|
320 | // give matrix (and model costs and bounds a chance to be |
---|
321 | // refreshed (normally null) |
---|
322 | matrix_->refresh(this); |
---|
323 | // If getting nowhere - why not give it a kick |
---|
324 | #if 0 |
---|
325 | // does not seem to work too well - do some more work |
---|
326 | if (perturbation_<101&&numberIterations_>2*(numberRows_+numberColumns_)) |
---|
327 | perturb(); |
---|
328 | #endif |
---|
329 | // may factorize, checks if problem finished |
---|
330 | statusOfProblemInDual(lastCleaned,factorType,progress); |
---|
331 | |
---|
332 | // Say good factorization |
---|
333 | factorType=1; |
---|
334 | if (saveSparse) { |
---|
335 | // use default at present |
---|
336 | factorization_->sparseThreshold(0); |
---|
337 | factorization_->goSparse(); |
---|
338 | } |
---|
339 | |
---|
340 | // Do iterations |
---|
341 | whileIterating(); |
---|
342 | } |
---|
343 | } |
---|
344 | |
---|
345 | //assert(!numberFake_||problemStatus_); // all bounds should be okay |
---|
346 | // at present we are leaving factorization around |
---|
347 | // maybe we should empty it |
---|
348 | deleteRim(); |
---|
349 | handler_->message(CLP_SIMPLEX_FINISHED+problemStatus_,messages_) |
---|
350 | <<objectiveValue() |
---|
351 | <<CoinMessageEol; |
---|
352 | // Restore any saved stuff |
---|
353 | perturbation_ = savePerturbation; |
---|
354 | factorization_->sparseThreshold(saveSparse); |
---|
355 | dualBound_ = saveDualBound_; |
---|
356 | return problemStatus_; |
---|
357 | } |
---|
358 | /* Reasons to come out: |
---|
359 | -1 iterations etc |
---|
360 | -2 inaccuracy |
---|
361 | -3 slight inaccuracy (and done iterations) |
---|
362 | +0 looks optimal (might be unbounded - but we will investigate) |
---|
363 | +1 looks infeasible |
---|
364 | +3 max iterations |
---|
365 | */ |
---|
366 | int |
---|
367 | ClpSimplexDual::whileIterating() |
---|
368 | { |
---|
369 | // status stays at -1 while iterating, >=0 finished, -2 to invert |
---|
370 | // status -3 to go to top without an invert |
---|
371 | int returnCode = -1; |
---|
372 | while (problemStatus_==-1) { |
---|
373 | #ifdef CLP_DEBUG |
---|
374 | { |
---|
375 | int i; |
---|
376 | for (i=0;i<4;i++) { |
---|
377 | rowArray_[i]->checkClear(); |
---|
378 | } |
---|
379 | for (i=0;i<2;i++) { |
---|
380 | columnArray_[i]->checkClear(); |
---|
381 | } |
---|
382 | } |
---|
383 | #endif |
---|
384 | #if CLP_DEBUG>2 |
---|
385 | // very expensive |
---|
386 | if (numberIterations_>0&&numberIterations_<-801) { |
---|
387 | handler_->setLogLevel(63); |
---|
388 | double saveValue = objectiveValue_; |
---|
389 | double * saveRow1 = new double[numberRows_]; |
---|
390 | double * saveRow2 = new double[numberRows_]; |
---|
391 | memcpy(saveRow1,rowReducedCost_,numberRows_*sizeof(double)); |
---|
392 | memcpy(saveRow2,rowActivityWork_,numberRows_*sizeof(double)); |
---|
393 | double * saveColumn1 = new double[numberColumns_]; |
---|
394 | double * saveColumn2 = new double[numberColumns_]; |
---|
395 | memcpy(saveColumn1,reducedCostWork_,numberColumns_*sizeof(double)); |
---|
396 | memcpy(saveColumn2,columnActivityWork_,numberColumns_*sizeof(double)); |
---|
397 | gutsOfSolution(rowActivityWork_,columnActivityWork_); |
---|
398 | printf("xxx %d old obj %g, recomputed %g, sum dual inf %g\n", |
---|
399 | numberIterations_, |
---|
400 | saveValue,objectiveValue_,sumDualInfeasibilities_); |
---|
401 | if (saveValue>objectiveValue_+1.0e-2) |
---|
402 | printf("**bad**\n"); |
---|
403 | memcpy(rowReducedCost_,saveRow1,numberRows_*sizeof(double)); |
---|
404 | memcpy(rowActivityWork_,saveRow2,numberRows_*sizeof(double)); |
---|
405 | memcpy(reducedCostWork_,saveColumn1,numberColumns_*sizeof(double)); |
---|
406 | memcpy(columnActivityWork_,saveColumn2,numberColumns_*sizeof(double)); |
---|
407 | delete [] saveRow1; |
---|
408 | delete [] saveRow2; |
---|
409 | delete [] saveColumn1; |
---|
410 | delete [] saveColumn2; |
---|
411 | objectiveValue_=saveValue; |
---|
412 | } |
---|
413 | #endif |
---|
414 | #ifdef CLP_DEBUG |
---|
415 | { |
---|
416 | int iSequence, number=numberRows_+numberColumns_; |
---|
417 | for (iSequence=0;iSequence<number;iSequence++) { |
---|
418 | double lowerValue=lower_[iSequence]; |
---|
419 | double upperValue=upper_[iSequence]; |
---|
420 | double value=solution_[iSequence]; |
---|
421 | if(getStatus(iSequence)!=ClpSimplex::basic) { |
---|
422 | assert(lowerValue>-1.0e20); |
---|
423 | assert(upperValue<1.0e20); |
---|
424 | } |
---|
425 | switch(getStatus(iSequence)) { |
---|
426 | |
---|
427 | case ClpSimplex::basic: |
---|
428 | break; |
---|
429 | case ClpSimplex::isFree: |
---|
430 | case ClpSimplex::superBasic: |
---|
431 | break; |
---|
432 | case ClpSimplex::atUpperBound: |
---|
433 | assert (fabs(value-upperValue)<=primalTolerance_) ; |
---|
434 | break; |
---|
435 | case ClpSimplex::atLowerBound: |
---|
436 | assert (fabs(value-lowerValue)<=primalTolerance_) ; |
---|
437 | break; |
---|
438 | } |
---|
439 | } |
---|
440 | } |
---|
441 | if(numberIterations_==debugIteration) { |
---|
442 | printf("dodgy iteration coming up\n"); |
---|
443 | } |
---|
444 | #endif |
---|
445 | // choose row to go out |
---|
446 | dualRow(); |
---|
447 | if (pivotRow_>=0) { |
---|
448 | // we found a pivot row |
---|
449 | handler_->message(CLP_SIMPLEX_PIVOTROW,messages_) |
---|
450 | <<pivotRow_ |
---|
451 | <<CoinMessageEol; |
---|
452 | // check accuracy of weights |
---|
453 | dualRowPivot_->checkAccuracy(); |
---|
454 | // get sign for finding row of tableau |
---|
455 | rowArray_[0]->insert(pivotRow_,directionOut_); |
---|
456 | factorization_->updateColumnTranspose(rowArray_[1],rowArray_[0]); |
---|
457 | // put row of tableau in rowArray[0] and columnArray[0] |
---|
458 | matrix_->transposeTimes(this,-1.0, |
---|
459 | rowArray_[0],columnArray_[1],columnArray_[0]); |
---|
460 | // rowArray has pi equivalent |
---|
461 | // do ratio test |
---|
462 | dualColumn(rowArray_[0],columnArray_[0],columnArray_[1], |
---|
463 | rowArray_[3]); |
---|
464 | if (sequenceIn_>=0) { |
---|
465 | // normal iteration |
---|
466 | // update the incoming column |
---|
467 | unpack(rowArray_[1]); |
---|
468 | factorization_->updateColumn(rowArray_[2],rowArray_[1],true); |
---|
469 | // and update dual weights (can do in parallel - with extra array) |
---|
470 | dualRowPivot_->updateWeights(rowArray_[0],rowArray_[2], |
---|
471 | rowArray_[1]); |
---|
472 | // see if update stable |
---|
473 | double btranAlpha = -alpha_*directionOut_; // for check |
---|
474 | alpha_=(*rowArray_[1])[pivotRow_]; |
---|
475 | #ifdef CLP_DEBUG |
---|
476 | if ((handler_->logLevel()&32)) |
---|
477 | printf("btran alpha %g, ftran alpha %g\n",btranAlpha,alpha_); |
---|
478 | #endif |
---|
479 | if (fabs(btranAlpha)<1.0e-12||fabs(alpha_)<1.0e-12|| |
---|
480 | fabs(btranAlpha-alpha_)>1.0e-7*(1.0+fabs(alpha_))) { |
---|
481 | handler_->message(CLP_DUAL_CHECK,messages_) |
---|
482 | <<btranAlpha |
---|
483 | <<alpha_ |
---|
484 | <<CoinMessageEol; |
---|
485 | dualRowPivot_->unrollWeights(); |
---|
486 | if (factorization_->pivots()) { |
---|
487 | problemStatus_=-2; // factorize now |
---|
488 | rowArray_[0]->clear(); |
---|
489 | rowArray_[1]->clear(); |
---|
490 | columnArray_[0]->clear(); |
---|
491 | returnCode=-2; |
---|
492 | break; |
---|
493 | } else { |
---|
494 | // take on more relaxed criterion |
---|
495 | if (fabs(btranAlpha)<1.0e-12||fabs(alpha_)<1.0e-12|| |
---|
496 | fabs(btranAlpha-alpha_)>1.0e-4*(1.0+fabs(alpha_))) { |
---|
497 | // need to reject something |
---|
498 | char x = isColumn(sequenceOut_) ? 'C' :'R'; |
---|
499 | handler_->message(CLP_SIMPLEX_FLAG,messages_) |
---|
500 | <<x<<sequenceWithin(sequenceOut_) |
---|
501 | <<CoinMessageEol; |
---|
502 | setFlagged(sequenceOut_); |
---|
503 | lastBadIteration_ = numberIterations_; // say be more cautious |
---|
504 | rowArray_[0]->clear(); |
---|
505 | rowArray_[1]->clear(); |
---|
506 | columnArray_[0]->clear(); |
---|
507 | continue; |
---|
508 | } |
---|
509 | } |
---|
510 | } |
---|
511 | // update duals BEFORE replaceColumn so can do updateColumn |
---|
512 | double objectiveChange=0.0; |
---|
513 | // do duals first as variables may flip bounds |
---|
514 | // rowArray_[0] and columnArray_[0] may have flips |
---|
515 | // so use rowArray_[3] for work array from here on |
---|
516 | int nswapped = |
---|
517 | updateDualsInDual(rowArray_[0],columnArray_[0],rowArray_[2],theta_, |
---|
518 | objectiveChange); |
---|
519 | // which will change basic solution |
---|
520 | if (nswapped) { |
---|
521 | #ifdef CLP_DEBUG |
---|
522 | if ((handler_->logLevel()&16)) |
---|
523 | printf("old dualOut_ %g, v %g, l %g, u %g - new ", |
---|
524 | dualOut_,valueOut_,lowerOut_,upperOut_); |
---|
525 | double oldOut=dualOut_; |
---|
526 | #endif |
---|
527 | factorization_->updateColumn(rowArray_[3],rowArray_[2],false); |
---|
528 | dualRowPivot_->updatePrimalSolution(rowArray_[2], |
---|
529 | 1.0,objectiveChange); |
---|
530 | |
---|
531 | // recompute dualOut_ |
---|
532 | valueOut_ = solution_[sequenceOut_]; |
---|
533 | if (directionOut_<0) { |
---|
534 | dualOut_ = valueOut_ - upperOut_; |
---|
535 | } else { |
---|
536 | dualOut_ = lowerOut_ - valueOut_; |
---|
537 | } |
---|
538 | #ifdef CLP_DEBUG |
---|
539 | if ((handler_->logLevel()&16)) |
---|
540 | printf("%g\n",dualOut_); |
---|
541 | assert(dualOut_<=oldOut); |
---|
542 | #endif |
---|
543 | if(dualOut_<0.0&&factorization_->pivots()) { |
---|
544 | // going backwards - factorize |
---|
545 | dualRowPivot_->unrollWeights(); |
---|
546 | problemStatus_=-2; // factorize now |
---|
547 | returnCode=-2; |
---|
548 | break; |
---|
549 | } |
---|
550 | } |
---|
551 | // amount primal will move |
---|
552 | double movement = -dualOut_*directionOut_/alpha_; |
---|
553 | // so objective should increase by fabs(dj)*movement |
---|
554 | // but we already have objective change - so check will be good |
---|
555 | if (objectiveChange+fabs(movement*dualIn_)<-1.0e-5) { |
---|
556 | #ifdef CLP_DEBUG |
---|
557 | if (handler_->logLevel()&32) |
---|
558 | printf("movement %g, swap change %g, rest %g * %g\n", |
---|
559 | objectiveChange+fabs(movement*dualIn_), |
---|
560 | objectiveChange,movement,dualIn_); |
---|
561 | #endif |
---|
562 | if(factorization_->pivots()>5) { |
---|
563 | // going backwards - factorize |
---|
564 | dualRowPivot_->unrollWeights(); |
---|
565 | problemStatus_=-2; // factorize now |
---|
566 | returnCode=-2; |
---|
567 | break; |
---|
568 | } |
---|
569 | } |
---|
570 | // if stable replace in basis |
---|
571 | int updateStatus = factorization_->replaceColumn(rowArray_[2], |
---|
572 | pivotRow_, |
---|
573 | alpha_); |
---|
574 | if (updateStatus==1) { |
---|
575 | // slight error |
---|
576 | if (factorization_->pivots()>5) { |
---|
577 | problemStatus_=-2; // factorize now |
---|
578 | returnCode=-3; |
---|
579 | } |
---|
580 | } else if (updateStatus==2) { |
---|
581 | // major error |
---|
582 | dualRowPivot_->unrollWeights(); |
---|
583 | // later we may need to unwind more e.g. fake bounds |
---|
584 | if (factorization_->pivots()) { |
---|
585 | problemStatus_=-2; // factorize now |
---|
586 | returnCode=-2; |
---|
587 | break; |
---|
588 | } else { |
---|
589 | // need to reject something |
---|
590 | char x = isColumn(sequenceOut_) ? 'C' :'R'; |
---|
591 | handler_->message(CLP_SIMPLEX_FLAG,messages_) |
---|
592 | <<x<<sequenceWithin(sequenceOut_) |
---|
593 | <<CoinMessageEol; |
---|
594 | setFlagged(sequenceOut_); |
---|
595 | lastBadIteration_ = numberIterations_; // say be more cautious |
---|
596 | rowArray_[0]->clear(); |
---|
597 | rowArray_[1]->clear(); |
---|
598 | columnArray_[0]->clear(); |
---|
599 | continue; |
---|
600 | } |
---|
601 | } else if (updateStatus==3) { |
---|
602 | // out of memory |
---|
603 | // increase space if not many iterations |
---|
604 | if (factorization_->pivots()< |
---|
605 | 0.5*factorization_->maximumPivots()&& |
---|
606 | factorization_->pivots()<200) |
---|
607 | factorization_->areaFactor( |
---|
608 | factorization_->areaFactor() * 1.1); |
---|
609 | problemStatus_=-2; // factorize now |
---|
610 | } |
---|
611 | // update primal solution |
---|
612 | if (theta_<0.0) { |
---|
613 | #ifdef CLP_DEBUG |
---|
614 | if (handler_->logLevel()&32) |
---|
615 | printf("negative theta %g\n",theta_); |
---|
616 | #endif |
---|
617 | theta_=0.0; |
---|
618 | } |
---|
619 | // do actual flips |
---|
620 | flipBounds(rowArray_[0],columnArray_[0],theta_); |
---|
621 | dualRowPivot_->updatePrimalSolution(rowArray_[1], |
---|
622 | movement, |
---|
623 | objectiveChange); |
---|
624 | #ifdef CLP_DEBUG |
---|
625 | double oldobj=objectiveValue_; |
---|
626 | #endif |
---|
627 | int whatNext=housekeeping(objectiveChange); |
---|
628 | // and set bounds correctly |
---|
629 | originalBound(sequenceIn_); |
---|
630 | changeBound(sequenceOut_); |
---|
631 | #ifdef CLP_DEBUG |
---|
632 | if (objectiveValue_<oldobj-1.0e-5&&(handler_->logLevel()&16)) |
---|
633 | printf("obj backwards %g %g\n",objectiveValue_,oldobj); |
---|
634 | #endif |
---|
635 | if (whatNext==1) { |
---|
636 | problemStatus_ =-2; // refactorize |
---|
637 | } else if (whatNext==2) { |
---|
638 | // maximum iterations or equivalent |
---|
639 | problemStatus_= 3; |
---|
640 | returnCode=3; |
---|
641 | break; |
---|
642 | } |
---|
643 | } else { |
---|
644 | // no incoming column is valid |
---|
645 | #ifdef CLP_DEBUG |
---|
646 | if (handler_->logLevel()&32) |
---|
647 | printf("** no column pivot\n"); |
---|
648 | #endif |
---|
649 | if (factorization_->pivots()<5) { |
---|
650 | problemStatus_=-4; //say looks infeasible |
---|
651 | // create ray anyway |
---|
652 | delete [] ray_; |
---|
653 | ray_ = new double [ numberRows_]; |
---|
654 | CoinDisjointCopyN(rowArray_[0]->denseVector(),numberRows_,ray_); |
---|
655 | } |
---|
656 | rowArray_[0]->clear(); |
---|
657 | columnArray_[0]->clear(); |
---|
658 | returnCode=1; |
---|
659 | break; |
---|
660 | } |
---|
661 | } else { |
---|
662 | // no pivot row |
---|
663 | #ifdef CLP_DEBUG |
---|
664 | if (handler_->logLevel()&32) |
---|
665 | printf("** no row pivot\n"); |
---|
666 | #endif |
---|
667 | if (!factorization_->pivots()) { |
---|
668 | // may have crept through - so may be optimal |
---|
669 | //problemStatus_=-5; //say looks unbounded |
---|
670 | problemStatus_=0; |
---|
671 | // check any flagged variables |
---|
672 | int iRow; |
---|
673 | for (iRow=0;iRow<numberRows_;iRow++) { |
---|
674 | int iPivot=pivotVariable_[iRow]; |
---|
675 | if (flagged(iPivot)) |
---|
676 | break; |
---|
677 | } |
---|
678 | if (iRow<numberRows_) { |
---|
679 | #ifdef CLP_DEBUG |
---|
680 | std::cerr<<"Flagged variables at end - infeasible?"<<std::endl; |
---|
681 | #endif |
---|
682 | problemStatus_=-4; //say looks infeasible |
---|
683 | // create ray anyway |
---|
684 | delete [] ray_; |
---|
685 | ray_ = new double [ numberRows_]; |
---|
686 | CoinDisjointCopyN(rowArray_[0]->denseVector(),numberRows_,ray_); |
---|
687 | } |
---|
688 | } |
---|
689 | returnCode=0; |
---|
690 | break; |
---|
691 | } |
---|
692 | } |
---|
693 | return returnCode; |
---|
694 | } |
---|
695 | /* The duals are updated by the given arrays. |
---|
696 | Returns number of infeasibilities. |
---|
697 | rowArray and columnarray will have flipped |
---|
698 | The output vector has movement (row length array) */ |
---|
699 | int |
---|
700 | ClpSimplexDual::updateDualsInDual(CoinIndexedVector * rowArray, |
---|
701 | CoinIndexedVector * columnArray, |
---|
702 | CoinIndexedVector * outputArray, |
---|
703 | double theta, |
---|
704 | double & objectiveChange) |
---|
705 | { |
---|
706 | |
---|
707 | outputArray->clear(); |
---|
708 | |
---|
709 | double * work; |
---|
710 | int number; |
---|
711 | int * which; |
---|
712 | |
---|
713 | int numberInfeasibilities=0; |
---|
714 | int numberRowInfeasibilities=0; |
---|
715 | |
---|
716 | // see whether we will be doing full recompute |
---|
717 | bool fullRecompute= (rowArray->getNumElements()==numberRows_&& |
---|
718 | columnArray->getNumElements()==numberColumns_); |
---|
719 | int numberAtFake=0; |
---|
720 | |
---|
721 | // use a tighter tolerance except for all being okay |
---|
722 | double tolerance = dualTolerance_; |
---|
723 | |
---|
724 | double changeObj=0.0; |
---|
725 | |
---|
726 | int iSection; |
---|
727 | |
---|
728 | for (iSection=0;iSection<2;iSection++) { |
---|
729 | int i; |
---|
730 | double * solution = solutionRegion(iSection); |
---|
731 | double * reducedCost = djRegion(iSection); |
---|
732 | double * lower = lowerRegion(iSection); |
---|
733 | double * upper = upperRegion(iSection); |
---|
734 | double * cost = costRegion(iSection); |
---|
735 | int addSequence; |
---|
736 | if (!iSection) { |
---|
737 | addSequence = numberColumns_; |
---|
738 | work = rowArray->denseVector(); |
---|
739 | number = rowArray->getNumElements(); |
---|
740 | which = rowArray->getIndices(); |
---|
741 | } else { |
---|
742 | // set number of infeasibilities in row array |
---|
743 | addSequence=0; |
---|
744 | numberRowInfeasibilities=numberInfeasibilities; |
---|
745 | rowArray->setNumElements(numberInfeasibilities); |
---|
746 | numberInfeasibilities=0; |
---|
747 | work = columnArray->denseVector(); |
---|
748 | number = columnArray->getNumElements(); |
---|
749 | which = columnArray->getIndices(); |
---|
750 | } |
---|
751 | |
---|
752 | for (i=0;i<number;i++) { |
---|
753 | int iSequence = which[i]; |
---|
754 | double alphaI = work[iSequence]; |
---|
755 | double value = reducedCost[iSequence]-theta*alphaI; |
---|
756 | work[iSequence]=0.0; |
---|
757 | reducedCost[iSequence]=value; |
---|
758 | |
---|
759 | if (!fixed(iSequence+addSequence)) { |
---|
760 | double movement=0.0; |
---|
761 | FakeBound bound = getFakeBound(iSequence+addSequence); |
---|
762 | Status status = getStatus(iSequence+addSequence); |
---|
763 | |
---|
764 | switch(status) { |
---|
765 | |
---|
766 | case ClpSimplex::basic: |
---|
767 | case ClpSimplex::superBasic: |
---|
768 | break; |
---|
769 | case ClpSimplex::isFree: |
---|
770 | if (fabs(value)>tolerance) { |
---|
771 | #ifdef CLP_DEBUG |
---|
772 | if (handler_->logLevel()&32) |
---|
773 | printf("%d %d, free has dj of %g, alpha %g\n", |
---|
774 | iSection,iSequence,value,alphaI); |
---|
775 | #endif |
---|
776 | } |
---|
777 | break; |
---|
778 | case ClpSimplex::atUpperBound: |
---|
779 | if (value>tolerance) { |
---|
780 | // to lower bound (if swap) |
---|
781 | // put back alpha |
---|
782 | which[numberInfeasibilities++]=iSequence; |
---|
783 | work[iSequence]=alphaI; |
---|
784 | movement = lower[iSequence]-upper[iSequence]; |
---|
785 | #ifdef CLP_DEBUG |
---|
786 | if ((handler_->logLevel()&32)) |
---|
787 | printf("%d %d, new dj %g, alpha %g, movement %g\n", |
---|
788 | iSection,iSequence,value,alphaI,movement); |
---|
789 | #endif |
---|
790 | changeObj += movement*cost[iSequence]; |
---|
791 | if (bound==ClpSimplexDual::bothFake|| |
---|
792 | bound==ClpSimplexDual::lowerFake) |
---|
793 | numberAtFake++; |
---|
794 | } else if (fullRecompute) { |
---|
795 | // at correct bound |
---|
796 | if (bound==ClpSimplexDual::bothFake|| |
---|
797 | bound==ClpSimplexDual::upperFake) { |
---|
798 | // but flip if dj would allow |
---|
799 | if (bound==ClpSimplexDual::upperFake&& |
---|
800 | value>=-tolerance) { |
---|
801 | movement = lower[iSequence]-upper[iSequence]; |
---|
802 | setStatus(iSequence+addSequence,ClpSimplex::atLowerBound); |
---|
803 | solution[iSequence] = lower[iSequence]; |
---|
804 | changeObj += movement*cost[iSequence]; |
---|
805 | } else { |
---|
806 | numberAtFake++; |
---|
807 | } |
---|
808 | } |
---|
809 | } |
---|
810 | break; |
---|
811 | case ClpSimplex::atLowerBound: |
---|
812 | if (value<-tolerance) { |
---|
813 | // to upper bound |
---|
814 | // put back alpha |
---|
815 | which[numberInfeasibilities++]=iSequence; |
---|
816 | work[iSequence]=alphaI; |
---|
817 | movement = upper[iSequence] - lower[iSequence]; |
---|
818 | #ifdef CLP_DEBUG |
---|
819 | if ((handler_->logLevel()&32)) |
---|
820 | printf("%d %d, new dj %g, alpha %g, movement %g\n", |
---|
821 | iSection,iSequence,value,alphaI,movement); |
---|
822 | #endif |
---|
823 | changeObj += movement*cost[iSequence]; |
---|
824 | if (bound==ClpSimplexDual::bothFake|| |
---|
825 | bound==ClpSimplexDual::upperFake) |
---|
826 | numberAtFake++; |
---|
827 | } else if (fullRecompute) { |
---|
828 | // at correct bound |
---|
829 | if (bound==ClpSimplexDual::bothFake|| |
---|
830 | bound==ClpSimplexDual::lowerFake) { |
---|
831 | // but flip if dj would allow |
---|
832 | if (bound==ClpSimplexDual::lowerFake&& |
---|
833 | value<=tolerance) { |
---|
834 | movement = upper[iSequence] - lower[iSequence]; |
---|
835 | setStatus(iSequence+addSequence,ClpSimplex::atUpperBound); |
---|
836 | solution[iSequence] = upper[iSequence]; |
---|
837 | changeObj += movement*cost[iSequence]; |
---|
838 | } else { |
---|
839 | numberAtFake++; |
---|
840 | } |
---|
841 | } |
---|
842 | } |
---|
843 | break; |
---|
844 | } |
---|
845 | if (!fullRecompute) { |
---|
846 | if (movement) { |
---|
847 | if (!iSection) { |
---|
848 | // row (sign ?) |
---|
849 | outputArray->quickAdd(iSequence,-movement); |
---|
850 | } else { |
---|
851 | matrix_->add(this,outputArray,iSequence,movement); |
---|
852 | } |
---|
853 | } |
---|
854 | } |
---|
855 | } |
---|
856 | } |
---|
857 | } |
---|
858 | #ifdef CLP_DEBUG |
---|
859 | if (fullRecompute&&numberAtFake&&(handler_->logLevel()&16)!=0) |
---|
860 | printf("%d fake after full update\n",numberAtFake); |
---|
861 | #endif |
---|
862 | outputArray->stopQuickAdd(); |
---|
863 | // set number of infeasibilities |
---|
864 | columnArray->setNumElements(numberInfeasibilities); |
---|
865 | numberInfeasibilities += numberRowInfeasibilities; |
---|
866 | if (fullRecompute) { |
---|
867 | // do actual flips |
---|
868 | flipBounds(rowArray,columnArray,theta); |
---|
869 | numberFake_ = numberAtFake; |
---|
870 | } |
---|
871 | objectiveChange += changeObj; |
---|
872 | return numberInfeasibilities; |
---|
873 | } |
---|
874 | /* |
---|
875 | Chooses dual pivot row |
---|
876 | Would be faster with separate region to scan |
---|
877 | and will have this (with square of infeasibility) when steepest |
---|
878 | For easy problems we can just choose one of the first rows we look at |
---|
879 | */ |
---|
880 | void |
---|
881 | ClpSimplexDual::dualRow() |
---|
882 | { |
---|
883 | // get pivot row using whichever method it is |
---|
884 | pivotRow_=dualRowPivot_->pivotRow(); |
---|
885 | if (pivotRow_>=0) { |
---|
886 | int iPivot=pivotVariable_[pivotRow_]; |
---|
887 | sequenceOut_ = iPivot; |
---|
888 | if (iPivot>=numberColumns_) { |
---|
889 | // slack |
---|
890 | iPivot-=numberColumns_; |
---|
891 | valueOut_=rowActivityWork_[iPivot]; |
---|
892 | lowerOut_=rowLowerWork_[iPivot]; |
---|
893 | upperOut_=rowUpperWork_[iPivot]; |
---|
894 | } else { |
---|
895 | // column |
---|
896 | valueOut_=columnActivityWork_[iPivot]; |
---|
897 | lowerOut_=columnLowerWork_[iPivot]; |
---|
898 | upperOut_=columnUpperWork_[iPivot]; |
---|
899 | } |
---|
900 | // if we have problems we could try other way and hope we get a |
---|
901 | // zero pivot? |
---|
902 | if (valueOut_>upperOut_) { |
---|
903 | directionOut_ = -1; |
---|
904 | dualOut_ = valueOut_ - upperOut_; |
---|
905 | } else { |
---|
906 | directionOut_ = 1; |
---|
907 | dualOut_ = lowerOut_ - valueOut_; |
---|
908 | } |
---|
909 | #ifdef CLP_DEBUG |
---|
910 | assert(dualOut_>=0.0); |
---|
911 | #endif |
---|
912 | } |
---|
913 | return ; |
---|
914 | } |
---|
915 | // Checks if any fake bounds active - if so returns number and modifies |
---|
916 | // dualBound_ and everything. |
---|
917 | // Free variables will be left as free |
---|
918 | // Returns number of bounds changed if >=0 |
---|
919 | // Returns -1 if not initialize and no effect |
---|
920 | // Fills in changeVector which can be used to see if unbounded |
---|
921 | // and cost of change vector |
---|
922 | int |
---|
923 | ClpSimplexDual::changeBounds(bool initialize, |
---|
924 | CoinIndexedVector * outputArray, |
---|
925 | double & changeCost) |
---|
926 | { |
---|
927 | if (!initialize) { |
---|
928 | int numberInfeasibilities; |
---|
929 | double newBound; |
---|
930 | newBound = 5.0*dualBound_; |
---|
931 | numberInfeasibilities=0; |
---|
932 | changeCost=0.0; |
---|
933 | // put back original bounds and then check |
---|
934 | createRim(3); |
---|
935 | int iSequence; |
---|
936 | // bounds will get bigger - just look at ones at bounds |
---|
937 | for (iSequence=0;iSequence<numberRows_+numberColumns_;iSequence++) { |
---|
938 | double lowerValue=lower_[iSequence]; |
---|
939 | double upperValue=upper_[iSequence]; |
---|
940 | double value=solution_[iSequence]; |
---|
941 | setFakeBound(iSequence,ClpSimplexDual::noFake); |
---|
942 | switch(getStatus(iSequence)) { |
---|
943 | |
---|
944 | case ClpSimplex::basic: |
---|
945 | break; |
---|
946 | case ClpSimplex::isFree: |
---|
947 | case ClpSimplex::superBasic: |
---|
948 | break; |
---|
949 | case ClpSimplex::atUpperBound: |
---|
950 | if (fabs(value-upperValue)>primalTolerance_) |
---|
951 | numberInfeasibilities++; |
---|
952 | break; |
---|
953 | case ClpSimplex::atLowerBound: |
---|
954 | if (fabs(value-lowerValue)>primalTolerance_) |
---|
955 | numberInfeasibilities++; |
---|
956 | break; |
---|
957 | } |
---|
958 | } |
---|
959 | if (numberInfeasibilities) { |
---|
960 | int iSequence; |
---|
961 | for (iSequence=0;iSequence<numberRows_+numberColumns_;iSequence++) { |
---|
962 | double lowerValue=lower_[iSequence]; |
---|
963 | double upperValue=upper_[iSequence]; |
---|
964 | double newLowerValue; |
---|
965 | double newUpperValue; |
---|
966 | Status status = getStatus(iSequence); |
---|
967 | if (status==ClpSimplex::atUpperBound|| |
---|
968 | status==ClpSimplex::atLowerBound) { |
---|
969 | double value = solution_[iSequence]; |
---|
970 | if (value-lowerValue<=upperValue-value) { |
---|
971 | newLowerValue = max(lowerValue,value-0.666667*newBound); |
---|
972 | newUpperValue = min(upperValue,newLowerValue+newBound); |
---|
973 | } else { |
---|
974 | newUpperValue = min(upperValue,value+0.666667*newBound); |
---|
975 | newLowerValue = max(lowerValue,newUpperValue-newBound); |
---|
976 | } |
---|
977 | lower_[iSequence]=newLowerValue; |
---|
978 | upper_[iSequence]=newUpperValue; |
---|
979 | if (newLowerValue > lowerValue) { |
---|
980 | if (newUpperValue < upperValue) |
---|
981 | setFakeBound(iSequence,ClpSimplexDual::bothFake); |
---|
982 | else |
---|
983 | setFakeBound(iSequence,ClpSimplexDual::lowerFake); |
---|
984 | } else { |
---|
985 | if (newUpperValue < upperValue) |
---|
986 | setFakeBound(iSequence,ClpSimplexDual::upperFake); |
---|
987 | } |
---|
988 | if (status==ClpSimplex::atUpperBound) |
---|
989 | solution_[iSequence] = newUpperValue; |
---|
990 | else |
---|
991 | solution_[iSequence] = newLowerValue; |
---|
992 | double movement = solution_[iSequence] - value; |
---|
993 | if (movement&&outputArray) { |
---|
994 | if (iSequence>=numberColumns_) { |
---|
995 | outputArray->quickAdd(iSequence,-movement); |
---|
996 | changeCost += movement*cost_[iSequence]; |
---|
997 | } else { |
---|
998 | matrix_->add(this,outputArray,iSequence,movement); |
---|
999 | changeCost += movement*cost_[iSequence]; |
---|
1000 | } |
---|
1001 | } |
---|
1002 | } |
---|
1003 | } |
---|
1004 | dualBound_ = newBound; |
---|
1005 | if (outputArray) |
---|
1006 | outputArray->stopQuickAdd(); |
---|
1007 | } else { |
---|
1008 | numberInfeasibilities=-1; |
---|
1009 | } |
---|
1010 | return numberInfeasibilities; |
---|
1011 | } else { |
---|
1012 | int iSequence; |
---|
1013 | |
---|
1014 | for (iSequence=0;iSequence<numberRows_+numberColumns_;iSequence++) { |
---|
1015 | Status status = getStatus(iSequence); |
---|
1016 | if (status==ClpSimplex::atUpperBound|| |
---|
1017 | status==ClpSimplex::atLowerBound) { |
---|
1018 | double lowerValue=lower_[iSequence]; |
---|
1019 | double upperValue=upper_[iSequence]; |
---|
1020 | double value = solution_[iSequence]; |
---|
1021 | if (lowerValue>-largeValue_||upperValue<largeValue_) { |
---|
1022 | if (lowerValue-value>-0.5*dualBound_|| |
---|
1023 | upperValue-value<0.5*dualBound_) { |
---|
1024 | if (fabs(lowerValue-value)<=fabs(upperValue-value)) { |
---|
1025 | if (upperValue > lowerValue + dualBound_) { |
---|
1026 | upper_[iSequence]=lowerValue+dualBound_; |
---|
1027 | setFakeBound(iSequence,ClpSimplexDual::upperFake); |
---|
1028 | } |
---|
1029 | } else { |
---|
1030 | if (lowerValue < upperValue - dualBound_) { |
---|
1031 | lower_[iSequence]=upperValue-dualBound_; |
---|
1032 | setFakeBound(iSequence,ClpSimplexDual::lowerFake); |
---|
1033 | } |
---|
1034 | } |
---|
1035 | } else { |
---|
1036 | lower_[iSequence]=-0.5*dualBound_; |
---|
1037 | upper_[iSequence]= 0.5*dualBound_; |
---|
1038 | setFakeBound(iSequence,ClpSimplexDual::bothFake); |
---|
1039 | } |
---|
1040 | } |
---|
1041 | } |
---|
1042 | } |
---|
1043 | return 1; |
---|
1044 | } |
---|
1045 | } |
---|
1046 | /* |
---|
1047 | Row array has row part of pivot row (as duals so sign may be switched) |
---|
1048 | Column array has column part. |
---|
1049 | This chooses pivot column. |
---|
1050 | Spare array will be needed when we start getting clever. |
---|
1051 | We will check for basic so spare array will never overflow. |
---|
1052 | If necessary will modify costs |
---|
1053 | */ |
---|
1054 | void |
---|
1055 | ClpSimplexDual::dualColumn(CoinIndexedVector * rowArray, |
---|
1056 | CoinIndexedVector * columnArray, |
---|
1057 | CoinIndexedVector * spareArray, |
---|
1058 | CoinIndexedVector * spareArray2) |
---|
1059 | { |
---|
1060 | double * work; |
---|
1061 | int number; |
---|
1062 | int * which; |
---|
1063 | double * reducedCost; |
---|
1064 | |
---|
1065 | int iSection; |
---|
1066 | |
---|
1067 | sequenceIn_=-1; |
---|
1068 | int numberPossiblySwapped=0; |
---|
1069 | int numberRemaining=0; |
---|
1070 | |
---|
1071 | double totalThru=0.0; // for when variables flip |
---|
1072 | double acceptablePivot=1.0e-7; |
---|
1073 | if (factorization_->pivots()) |
---|
1074 | acceptablePivot=1.0e-5; // if we have iterated be more strict |
---|
1075 | double bestEverPivot=acceptablePivot; |
---|
1076 | int lastSequence = -1; |
---|
1077 | double lastPivot=0.0; |
---|
1078 | double upperTheta; |
---|
1079 | double newTolerance = dualTolerance_; |
---|
1080 | // will we need to increase tolerance |
---|
1081 | bool thisIncrease=false; |
---|
1082 | // If we think we need to modify costs (not if something from broad sweep) |
---|
1083 | bool modifyCosts=false; |
---|
1084 | // Increase in objective due to swapping bounds (may be negative) |
---|
1085 | double increaseInObjective=0.0; |
---|
1086 | |
---|
1087 | // use spareArrays to put ones looked at in |
---|
1088 | // we are going to flip flop between |
---|
1089 | int iFlip = 0; |
---|
1090 | // Possible list of pivots |
---|
1091 | int interesting[2]; |
---|
1092 | // where possible swapped ones are |
---|
1093 | int swapped[2]; |
---|
1094 | // for zeroing out arrays after |
---|
1095 | int marker[2][2]; |
---|
1096 | // pivot elements |
---|
1097 | double * array[2], * spare, * spare2; |
---|
1098 | // indices |
---|
1099 | int * indices[2], * index, * index2; |
---|
1100 | spareArray->clear(); |
---|
1101 | spareArray2->clear(); |
---|
1102 | array[0] = spareArray->denseVector(); |
---|
1103 | indices[0] = spareArray->getIndices(); |
---|
1104 | spare = array[0]; |
---|
1105 | index = indices[0]; |
---|
1106 | array[1] = spareArray2->denseVector(); |
---|
1107 | indices[1] = spareArray2->getIndices(); |
---|
1108 | int i; |
---|
1109 | double * lower; |
---|
1110 | double * upper; |
---|
1111 | |
---|
1112 | // initialize lists |
---|
1113 | for (i=0;i<2;i++) { |
---|
1114 | interesting[i]=0; |
---|
1115 | swapped[i]=numberColumns_; |
---|
1116 | marker[i][0]=0; |
---|
1117 | marker[i][1]=numberColumns_; |
---|
1118 | } |
---|
1119 | |
---|
1120 | /* |
---|
1121 | First we get a list of possible pivots. We can also see if the |
---|
1122 | problem looks infeasible or whether we want to pivot in free variable. |
---|
1123 | This may make objective go backwards but can only happen a finite |
---|
1124 | number of times and I do want free variables basic. |
---|
1125 | |
---|
1126 | Then we flip back and forth. At the start of each iteration |
---|
1127 | interesting[iFlip] should have possible candidates and swapped[iFlip] |
---|
1128 | will have pivots if we decide to take a previous pivot. |
---|
1129 | At end of each iteration interesting[1-iFlip] should have |
---|
1130 | candidates if we go through this theta and swapped[1-iFlip] |
---|
1131 | pivots if we don't go through. |
---|
1132 | |
---|
1133 | At first we increase theta and see what happens. We start |
---|
1134 | theta at a reasonable guess. If in right area then we do bit by bit. |
---|
1135 | |
---|
1136 | */ |
---|
1137 | |
---|
1138 | // do first pass to get possibles |
---|
1139 | // We can also see if infeasible or pivoting on free |
---|
1140 | double tentativeTheta = 1.0e22; |
---|
1141 | upperTheta = 1.0e31; |
---|
1142 | double freePivot = acceptablePivot; |
---|
1143 | for (iSection=0;iSection<2;iSection++) { |
---|
1144 | |
---|
1145 | int addSequence; |
---|
1146 | |
---|
1147 | if (!iSection) { |
---|
1148 | lower = rowLowerWork_; |
---|
1149 | upper = rowUpperWork_; |
---|
1150 | work = rowArray->denseVector(); |
---|
1151 | number = rowArray->getNumElements(); |
---|
1152 | which = rowArray->getIndices(); |
---|
1153 | reducedCost = rowReducedCost_; |
---|
1154 | addSequence = numberColumns_; |
---|
1155 | } else { |
---|
1156 | lower = columnLowerWork_; |
---|
1157 | upper = columnUpperWork_; |
---|
1158 | work = columnArray->denseVector(); |
---|
1159 | number = columnArray->getNumElements(); |
---|
1160 | which = columnArray->getIndices(); |
---|
1161 | reducedCost = reducedCostWork_; |
---|
1162 | addSequence = 0; |
---|
1163 | } |
---|
1164 | |
---|
1165 | for (i=0;i<number;i++) { |
---|
1166 | int iSequence = which[i]; |
---|
1167 | double alpha = work[iSequence]; |
---|
1168 | if (fixed(iSequence+addSequence)||!alpha) |
---|
1169 | continue; // skip fixed ones or (zeroed out) |
---|
1170 | double oldValue = reducedCost[iSequence]; |
---|
1171 | double value = oldValue-tentativeTheta*alpha; |
---|
1172 | int keep = 0; |
---|
1173 | |
---|
1174 | switch(getStatus(iSequence+addSequence)) { |
---|
1175 | |
---|
1176 | case ClpSimplex::basic: |
---|
1177 | break; |
---|
1178 | case ClpSimplex::isFree: |
---|
1179 | case ClpSimplex::superBasic: |
---|
1180 | if (oldValue>dualTolerance_) { |
---|
1181 | if (value<-newTolerance) |
---|
1182 | keep = 2; |
---|
1183 | } else if (oldValue<-dualTolerance_) { |
---|
1184 | if (value>newTolerance) |
---|
1185 | keep = 2; |
---|
1186 | } else { |
---|
1187 | if (alpha>=acceptablePivot) |
---|
1188 | keep = 2; |
---|
1189 | else if (-alpha>=acceptablePivot) |
---|
1190 | keep = 2; |
---|
1191 | } |
---|
1192 | break; |
---|
1193 | case ClpSimplex::atUpperBound: |
---|
1194 | assert (oldValue<=dualTolerance_*1.0001); |
---|
1195 | if (value>newTolerance) { |
---|
1196 | keep = 1; |
---|
1197 | value = oldValue-upperTheta*alpha; |
---|
1198 | if (value>newTolerance && -alpha>=acceptablePivot) |
---|
1199 | upperTheta = (oldValue-newTolerance)/alpha; |
---|
1200 | } |
---|
1201 | break; |
---|
1202 | case ClpSimplex::atLowerBound: |
---|
1203 | assert (oldValue>=-dualTolerance_*1.0001); |
---|
1204 | if (value<-newTolerance) { |
---|
1205 | keep = 1; |
---|
1206 | value = oldValue-upperTheta*alpha; |
---|
1207 | if (value<-newTolerance && alpha>=acceptablePivot) |
---|
1208 | upperTheta = (oldValue+newTolerance)/alpha; |
---|
1209 | } |
---|
1210 | break; |
---|
1211 | } |
---|
1212 | if (keep) { |
---|
1213 | if (keep==2) { |
---|
1214 | // free - choose largest |
---|
1215 | if (fabs(alpha)>freePivot) { |
---|
1216 | freePivot=fabs(alpha); |
---|
1217 | sequenceIn_ = iSequence + addSequence; |
---|
1218 | theta_=oldValue/alpha; |
---|
1219 | } |
---|
1220 | } else { |
---|
1221 | // add to list |
---|
1222 | spare[numberRemaining]=alpha; |
---|
1223 | index[numberRemaining++]=iSequence+addSequence; |
---|
1224 | } |
---|
1225 | } |
---|
1226 | } |
---|
1227 | } |
---|
1228 | interesting[0]=numberRemaining; |
---|
1229 | marker[0][0] = numberRemaining; |
---|
1230 | |
---|
1231 | if (!numberRemaining) |
---|
1232 | return; // Looks infeasible |
---|
1233 | |
---|
1234 | if (sequenceIn_>=0) { |
---|
1235 | // free variable - always choose |
---|
1236 | } else { |
---|
1237 | |
---|
1238 | theta_=1.0e50; |
---|
1239 | // now flip flop between spare arrays until reasonable theta |
---|
1240 | tentativeTheta = max(10.0*upperTheta,1.0e-7); |
---|
1241 | |
---|
1242 | // loops increasing tentative theta until can't go through |
---|
1243 | |
---|
1244 | while (tentativeTheta < 1.0e22) { |
---|
1245 | double thruThis = 0.0; |
---|
1246 | |
---|
1247 | double bestPivot=acceptablePivot; |
---|
1248 | int bestSequence=-1; |
---|
1249 | |
---|
1250 | numberPossiblySwapped = numberColumns_; |
---|
1251 | numberRemaining = 0; |
---|
1252 | |
---|
1253 | upperTheta = 1.0e50; |
---|
1254 | |
---|
1255 | spare = array[iFlip]; |
---|
1256 | index = indices[iFlip]; |
---|
1257 | spare2 = array[1-iFlip]; |
---|
1258 | index2 = indices[1-iFlip]; |
---|
1259 | |
---|
1260 | // try 3 different ways |
---|
1261 | // 1 bias increase by ones with slightly wrong djs |
---|
1262 | // 2 bias by all |
---|
1263 | // 3 bias by all - tolerance (doesn't seem very good) |
---|
1264 | #define TRYBIAS 1 |
---|
1265 | |
---|
1266 | |
---|
1267 | double increaseInThis=0.0; //objective increase in this loop |
---|
1268 | |
---|
1269 | for (i=0;i<interesting[iFlip];i++) { |
---|
1270 | int iSequence = index[i]; |
---|
1271 | double alpha = spare[i]; |
---|
1272 | double oldValue = dj_[iSequence]; |
---|
1273 | double value = oldValue-tentativeTheta*alpha; |
---|
1274 | |
---|
1275 | if (alpha < 0.0) { |
---|
1276 | //at upper bound |
---|
1277 | if (value>newTolerance) { |
---|
1278 | double range = upper_[iSequence] - lower_[iSequence]; |
---|
1279 | thruThis -= range*alpha; |
---|
1280 | #if TRYBIAS==1 |
---|
1281 | if (oldValue>0.0) |
---|
1282 | increaseInThis -= oldValue*range; |
---|
1283 | #elif TRYBIAS==2 |
---|
1284 | increaseInThis -= oldValue*range; |
---|
1285 | #else |
---|
1286 | increaseInThis -= (oldValue+dualTolerance_)*range; |
---|
1287 | #endif |
---|
1288 | // goes on swapped list (also means candidates if too many) |
---|
1289 | spare2[--numberPossiblySwapped]=alpha; |
---|
1290 | index2[numberPossiblySwapped]=iSequence; |
---|
1291 | if (fabs(alpha)>bestPivot) { |
---|
1292 | bestPivot=fabs(alpha); |
---|
1293 | bestSequence=numberPossiblySwapped; |
---|
1294 | } |
---|
1295 | } else { |
---|
1296 | value = oldValue-upperTheta*alpha; |
---|
1297 | if (value>newTolerance && -alpha>=acceptablePivot) |
---|
1298 | upperTheta = (oldValue-newTolerance)/alpha; |
---|
1299 | spare2[numberRemaining]=alpha; |
---|
1300 | index2[numberRemaining++]=iSequence; |
---|
1301 | } |
---|
1302 | } else { |
---|
1303 | // at lower bound |
---|
1304 | if (value<-newTolerance) { |
---|
1305 | double range = upper_[iSequence] - lower_[iSequence]; |
---|
1306 | thruThis += range*alpha; |
---|
1307 | //?? is this correct - and should we look at good ones |
---|
1308 | #if TRYBIAS==1 |
---|
1309 | if (oldValue<0.0) |
---|
1310 | increaseInThis += oldValue*range; |
---|
1311 | #elif TRYBIAS==2 |
---|
1312 | increaseInThis += oldValue*range; |
---|
1313 | #else |
---|
1314 | increaseInThis += (oldValue-dualTolerance_)*range; |
---|
1315 | #endif |
---|
1316 | // goes on swapped list (also means candidates if too many) |
---|
1317 | spare2[--numberPossiblySwapped]=alpha; |
---|
1318 | index2[numberPossiblySwapped]=iSequence; |
---|
1319 | if (fabs(alpha)>bestPivot) { |
---|
1320 | bestPivot=fabs(alpha); |
---|
1321 | bestSequence=numberPossiblySwapped; |
---|
1322 | } |
---|
1323 | } else { |
---|
1324 | value = oldValue-upperTheta*alpha; |
---|
1325 | if (value<-newTolerance && alpha>=acceptablePivot) |
---|
1326 | upperTheta = (oldValue+newTolerance)/alpha; |
---|
1327 | spare2[numberRemaining]=alpha; |
---|
1328 | index2[numberRemaining++]=iSequence; |
---|
1329 | } |
---|
1330 | } |
---|
1331 | } |
---|
1332 | swapped[1-iFlip]=numberPossiblySwapped; |
---|
1333 | interesting[1-iFlip]=numberRemaining; |
---|
1334 | marker[1-iFlip][0]= max(marker[1-iFlip][0],numberRemaining); |
---|
1335 | marker[1-iFlip][1]= min(marker[1-iFlip][1],numberPossiblySwapped); |
---|
1336 | |
---|
1337 | if (totalThru+thruThis>=fabs(dualOut_)|| |
---|
1338 | increaseInObjective+increaseInThis<0.0) { |
---|
1339 | // We should be pivoting in this batch |
---|
1340 | // so compress down to this lot |
---|
1341 | numberRemaining=0; |
---|
1342 | for (i=numberColumns_-1;i>=swapped[1-iFlip];i--) { |
---|
1343 | spare[numberRemaining]=spare2[i]; |
---|
1344 | index[numberRemaining++]=index2[i]; |
---|
1345 | } |
---|
1346 | interesting[iFlip]=numberRemaining; |
---|
1347 | int iTry; |
---|
1348 | #define MAXTRY 100 |
---|
1349 | // first get ratio with tolerance |
---|
1350 | for (iTry=0;iTry<MAXTRY;iTry++) { |
---|
1351 | |
---|
1352 | upperTheta=1.0e50; |
---|
1353 | numberPossiblySwapped = numberColumns_; |
---|
1354 | numberRemaining = 0; |
---|
1355 | |
---|
1356 | increaseInThis=0.0; //objective increase in this loop |
---|
1357 | |
---|
1358 | thruThis=0.0; |
---|
1359 | |
---|
1360 | spare = array[iFlip]; |
---|
1361 | index = indices[iFlip]; |
---|
1362 | spare2 = array[1-iFlip]; |
---|
1363 | index2 = indices[1-iFlip]; |
---|
1364 | |
---|
1365 | for (i=0;i<interesting[iFlip];i++) { |
---|
1366 | int iSequence=index[i]; |
---|
1367 | double alpha=spare[i]; |
---|
1368 | double oldValue = dj_[iSequence]; |
---|
1369 | double value = oldValue-upperTheta*alpha; |
---|
1370 | |
---|
1371 | if (alpha < 0.0) { |
---|
1372 | //at upper bound |
---|
1373 | if (value>newTolerance) { |
---|
1374 | if (-alpha>=acceptablePivot) { |
---|
1375 | upperTheta = (oldValue-newTolerance)/alpha; |
---|
1376 | } |
---|
1377 | } |
---|
1378 | } else { |
---|
1379 | // at lower bound |
---|
1380 | if (value<-newTolerance) { |
---|
1381 | if (alpha>=acceptablePivot) { |
---|
1382 | upperTheta = (oldValue+newTolerance)/alpha; |
---|
1383 | } |
---|
1384 | } |
---|
1385 | } |
---|
1386 | } |
---|
1387 | bestPivot=acceptablePivot; |
---|
1388 | sequenceIn_=-1; |
---|
1389 | // now choose largest and sum all ones which will go through |
---|
1390 | #define MINIMUMTHETA 1.0e-12 |
---|
1391 | for (i=0;i<interesting[iFlip];i++) { |
---|
1392 | int iSequence=index[i]; |
---|
1393 | double alpha=spare[i]; |
---|
1394 | double value = dj_[iSequence]-upperTheta*alpha; |
---|
1395 | double badDj=0.0; |
---|
1396 | |
---|
1397 | bool addToSwapped=false; |
---|
1398 | |
---|
1399 | if (alpha < 0.0) { |
---|
1400 | //at upper bound |
---|
1401 | if (value>=0.0) { |
---|
1402 | addToSwapped=true; |
---|
1403 | #if TRYBIAS==1 |
---|
1404 | badDj = -max(dj_[iSequence],0.0); |
---|
1405 | #elif TRYBIAS==2 |
---|
1406 | badDj = -dj_[iSequence]; |
---|
1407 | #else |
---|
1408 | badDj = -dj_[iSequence]-dualTolerance_; |
---|
1409 | #endif |
---|
1410 | } |
---|
1411 | } else { |
---|
1412 | // at lower bound |
---|
1413 | if (value<=0.0) { |
---|
1414 | addToSwapped=true; |
---|
1415 | #if TRYBIAS==1 |
---|
1416 | badDj = min(dj_[iSequence],0.0); |
---|
1417 | #elif TRYBIAS==2 |
---|
1418 | badDj = dj_[iSequence]; |
---|
1419 | #else |
---|
1420 | badDj = dj_[iSequence]-dualTolerance_; |
---|
1421 | #endif |
---|
1422 | } |
---|
1423 | } |
---|
1424 | if (!addToSwapped) { |
---|
1425 | // add to list of remaining |
---|
1426 | spare2[numberRemaining]=alpha; |
---|
1427 | index2[numberRemaining++]=iSequence; |
---|
1428 | } else { |
---|
1429 | // add to list of swapped |
---|
1430 | spare2[--numberPossiblySwapped]=alpha; |
---|
1431 | index2[numberPossiblySwapped]=iSequence; |
---|
1432 | // select if largest pivot |
---|
1433 | if (fabs(alpha)>bestPivot) { |
---|
1434 | sequenceIn_ = numberPossiblySwapped; |
---|
1435 | bestPivot = fabs(alpha); |
---|
1436 | theta_ = dj_[iSequence]/alpha; |
---|
1437 | } |
---|
1438 | double range = upper[iSequence] - lower[iSequence]; |
---|
1439 | thruThis += range*fabs(alpha); |
---|
1440 | increaseInThis += badDj*range; |
---|
1441 | } |
---|
1442 | } |
---|
1443 | swapped[1-iFlip]=numberPossiblySwapped; |
---|
1444 | interesting[1-iFlip]=numberRemaining; |
---|
1445 | marker[1-iFlip][0]= max(marker[1-iFlip][0],numberRemaining); |
---|
1446 | marker[1-iFlip][1]= min(marker[1-iFlip][1],numberPossiblySwapped); |
---|
1447 | // If we stop now this will be increase in objective (I think) |
---|
1448 | double increase = (fabs(dualOut_)-totalThru)*theta_; |
---|
1449 | increase += increaseInObjective; |
---|
1450 | if (theta_<0.0) |
---|
1451 | thruThis += fabs(dualOut_); // force using this one |
---|
1452 | if (increaseInObjective<0.0&&increase<0.0&&lastSequence>=0) { |
---|
1453 | // back |
---|
1454 | // We may need to be more careful - we could do by |
---|
1455 | // switch so we always do fine grained? |
---|
1456 | bestPivot=0.0; |
---|
1457 | } else { |
---|
1458 | // add in |
---|
1459 | totalThru += thruThis; |
---|
1460 | increaseInObjective += increaseInThis; |
---|
1461 | } |
---|
1462 | if (bestPivot<0.1*bestEverPivot&& |
---|
1463 | bestEverPivot>1.0e-6&&bestPivot<1.0e-3) { |
---|
1464 | // back to previous one |
---|
1465 | sequenceIn_=lastSequence; |
---|
1466 | // swap regions |
---|
1467 | iFlip = 1-iFlip; |
---|
1468 | break; |
---|
1469 | } else if (sequenceIn_==-1&&upperTheta>largeValue_) { |
---|
1470 | if (lastPivot>acceptablePivot) { |
---|
1471 | // back to previous one |
---|
1472 | sequenceIn_=lastSequence; |
---|
1473 | // swap regions |
---|
1474 | iFlip = 1-iFlip; |
---|
1475 | } else { |
---|
1476 | // can only get here if all pivots too small |
---|
1477 | } |
---|
1478 | break; |
---|
1479 | } else if (totalThru>=fabs(dualOut_)) { |
---|
1480 | modifyCosts=true; // fine grain - we can modify costs |
---|
1481 | break; // no point trying another loop |
---|
1482 | } else { |
---|
1483 | lastSequence=sequenceIn_; |
---|
1484 | if (bestPivot>bestEverPivot) |
---|
1485 | bestEverPivot=bestPivot; |
---|
1486 | iFlip = 1 -iFlip; |
---|
1487 | modifyCosts=true; // fine grain - we can modify costs |
---|
1488 | } |
---|
1489 | } |
---|
1490 | if (iTry==MAXTRY) |
---|
1491 | iFlip = 1-iFlip; // flip back |
---|
1492 | break; |
---|
1493 | } else { |
---|
1494 | // skip this lot |
---|
1495 | if (bestPivot>1.0e-3||bestPivot>bestEverPivot) { |
---|
1496 | bestEverPivot=bestPivot; |
---|
1497 | lastSequence=bestSequence; |
---|
1498 | } else { |
---|
1499 | // keep old swapped |
---|
1500 | memcpy(array[1-iFlip]+swapped[iFlip], |
---|
1501 | array[iFlip]+swapped[iFlip], |
---|
1502 | (numberColumns_-swapped[iFlip])*sizeof(double)); |
---|
1503 | memcpy(indices[1-iFlip]+swapped[iFlip], |
---|
1504 | indices[iFlip]+swapped[iFlip], |
---|
1505 | (numberColumns_-swapped[iFlip])*sizeof(int)); |
---|
1506 | marker[1-iFlip][1] = min(marker[1-iFlip][1],swapped[iFlip]); |
---|
1507 | swapped[1-iFlip]=swapped[iFlip]; |
---|
1508 | } |
---|
1509 | increaseInObjective += increaseInThis; |
---|
1510 | iFlip = 1 - iFlip; // swap regions |
---|
1511 | tentativeTheta = 2.0*upperTheta; |
---|
1512 | totalThru += thruThis; |
---|
1513 | } |
---|
1514 | } |
---|
1515 | |
---|
1516 | // can get here without sequenceIn_ set but with lastSequence |
---|
1517 | if (sequenceIn_<0&&lastSequence>=0) { |
---|
1518 | // back to previous one |
---|
1519 | sequenceIn_=lastSequence; |
---|
1520 | // swap regions |
---|
1521 | iFlip = 1-iFlip; |
---|
1522 | } |
---|
1523 | |
---|
1524 | if (sequenceIn_>=0) { |
---|
1525 | // at this stage sequenceIn_ is just pointer into index array |
---|
1526 | // flip just so we can use iFlip |
---|
1527 | iFlip = 1 -iFlip; |
---|
1528 | spare = array[iFlip]; |
---|
1529 | index = indices[iFlip]; |
---|
1530 | double oldValue; |
---|
1531 | double alpha = spare[sequenceIn_]; |
---|
1532 | sequenceIn_ = indices[iFlip][sequenceIn_]; |
---|
1533 | oldValue = dj_[sequenceIn_]; |
---|
1534 | theta_ = oldValue/alpha; |
---|
1535 | #if 0 |
---|
1536 | if (numberIterations_>2000) |
---|
1537 | exit(99); |
---|
1538 | if (numberIterations_>2000-20) |
---|
1539 | handler_->setLogLevel(63); |
---|
1540 | if (numberIterations_>2000-20) |
---|
1541 | printf("theta %g oldValue %g tol %g %g\n",theta_,oldValue,dualTolerance_, |
---|
1542 | newTolerance); |
---|
1543 | #endif |
---|
1544 | if (theta_<MINIMUMTHETA) { |
---|
1545 | // can't pivot to zero |
---|
1546 | if (oldValue-MINIMUMTHETA*alpha>=-dualTolerance_) { |
---|
1547 | theta_=MINIMUMTHETA; |
---|
1548 | } else if (oldValue-MINIMUMTHETA*alpha>=-newTolerance) { |
---|
1549 | theta_=MINIMUMTHETA; |
---|
1550 | thisIncrease=true; |
---|
1551 | } else { |
---|
1552 | theta_=(oldValue+newTolerance)/alpha; |
---|
1553 | assert(theta_>=0.0); |
---|
1554 | thisIncrease=true; |
---|
1555 | } |
---|
1556 | } |
---|
1557 | // may need to adjust costs so all dual feasible AND pivoted is exactly 0 |
---|
1558 | if (modifyCosts) { |
---|
1559 | int i; |
---|
1560 | double * workRow = rowArray->denseVector(); |
---|
1561 | double * workColumn = columnArray->denseVector(); |
---|
1562 | for (i=numberColumns_-1;i>=swapped[iFlip];i--) { |
---|
1563 | int iSequence=index[i]; |
---|
1564 | double alpha; |
---|
1565 | if (iSequence>=numberColumns_) |
---|
1566 | alpha=workRow[iSequence-numberColumns_]; |
---|
1567 | else |
---|
1568 | alpha=workColumn[iSequence]; |
---|
1569 | double value = dj_[iSequence]-theta_*alpha; |
---|
1570 | #if 0 |
---|
1571 | if (numberIterations_>2000-20) |
---|
1572 | printf("%d alpha %g value %g\n",iSequence,alpha,value); |
---|
1573 | #endif |
---|
1574 | |
---|
1575 | // can't be free here |
---|
1576 | |
---|
1577 | if (alpha < 0.0) { |
---|
1578 | //at upper bound |
---|
1579 | if (value>dualTolerance_) { |
---|
1580 | thisIncrease=true; |
---|
1581 | #define MODIFYCOST 2 |
---|
1582 | #if MODIFYCOST |
---|
1583 | // modify cost to hit new tolerance |
---|
1584 | double modification = alpha*theta_-dj_[iSequence] |
---|
1585 | +newTolerance; |
---|
1586 | //modification = min(modification,dualTolerance_); |
---|
1587 | //assert (fabs(modification)<1.0e-7); |
---|
1588 | dj_[iSequence] += modification; |
---|
1589 | cost_[iSequence] += modification; |
---|
1590 | #if 0 |
---|
1591 | if (numberIterations_>2000-20) |
---|
1592 | printf("%d acost %g mod %g\n",iSequence,cost_[iSequence], |
---|
1593 | modification); |
---|
1594 | #endif |
---|
1595 | #endif |
---|
1596 | } |
---|
1597 | } else { |
---|
1598 | // at lower bound |
---|
1599 | if (-value>dualTolerance_) { |
---|
1600 | thisIncrease=true; |
---|
1601 | #if MODIFYCOST |
---|
1602 | // modify cost to hit new tolerance |
---|
1603 | double modification = alpha*theta_-dj_[iSequence] |
---|
1604 | -newTolerance; |
---|
1605 | //modification = max(modification,-dualTolerance_); |
---|
1606 | //assert (fabs(modification)<1.0e-7); |
---|
1607 | dj_[iSequence] += modification; |
---|
1608 | cost_[iSequence] += modification; |
---|
1609 | #if 0 |
---|
1610 | if (numberIterations_>2000-20) |
---|
1611 | printf("%d bcost %g mod %g\n",iSequence,cost_[iSequence], |
---|
1612 | modification); |
---|
1613 | #endif |
---|
1614 | #endif |
---|
1615 | } |
---|
1616 | } |
---|
1617 | } |
---|
1618 | } |
---|
1619 | } |
---|
1620 | } |
---|
1621 | |
---|
1622 | if (sequenceIn_>=0) { |
---|
1623 | if (sequenceIn_>=numberColumns_) { |
---|
1624 | //slack |
---|
1625 | alpha_ = rowArray->denseVector()[sequenceIn_-numberColumns_]; |
---|
1626 | } else { |
---|
1627 | // column |
---|
1628 | alpha_ = columnArray->denseVector()[sequenceIn_]; |
---|
1629 | } |
---|
1630 | lowerIn_ = lower_[sequenceIn_]; |
---|
1631 | upperIn_ = upper_[sequenceIn_]; |
---|
1632 | valueIn_ = solution_[sequenceIn_]; |
---|
1633 | dualIn_ = dj_[sequenceIn_]; |
---|
1634 | |
---|
1635 | if (numberTimesOptimal_) { |
---|
1636 | // can we adjust cost back closer to original |
---|
1637 | //*** add coding |
---|
1638 | } |
---|
1639 | #if MODIFYCOST>1 |
---|
1640 | // modify cost to hit zero exactly |
---|
1641 | // so (dualIn_+modification)==theta_*alpha_ |
---|
1642 | double modification = theta_*alpha_-dualIn_; |
---|
1643 | dualIn_ += modification; |
---|
1644 | dj_[sequenceIn_]=dualIn_; |
---|
1645 | cost_[sequenceIn_] += modification; |
---|
1646 | //assert (fabs(modification)<1.0e-6); |
---|
1647 | #ifdef CLP_DEBUG |
---|
1648 | if ((handler_->logLevel()&32)&&fabs(modification)>1.0e-15) |
---|
1649 | printf("exact %d new cost %g, change %g\n",sequenceIn_, |
---|
1650 | cost_[sequenceIn_],modification); |
---|
1651 | #endif |
---|
1652 | #endif |
---|
1653 | |
---|
1654 | if (alpha_<0.0) { |
---|
1655 | // as if from upper bound |
---|
1656 | directionIn_=-1; |
---|
1657 | upperIn_=valueIn_; |
---|
1658 | } else { |
---|
1659 | // as if from lower bound |
---|
1660 | directionIn_=1; |
---|
1661 | lowerIn_=valueIn_; |
---|
1662 | } |
---|
1663 | } |
---|
1664 | if (thisIncrease) { |
---|
1665 | newTolerance = dualTolerance_+1.0e-4*dblParam_[ClpDualTolerance]; |
---|
1666 | } |
---|
1667 | |
---|
1668 | // clear arrays |
---|
1669 | |
---|
1670 | for (i=0;i<2;i++) { |
---|
1671 | memset(array[i],0,marker[i][0]*sizeof(double)); |
---|
1672 | memset(array[i]+marker[i][1],0, |
---|
1673 | (numberColumns_-marker[i][1])*sizeof(double)); |
---|
1674 | } |
---|
1675 | } |
---|
1676 | /* Checks if tentative optimal actually means unbounded |
---|
1677 | Returns -3 if not, 2 if is unbounded */ |
---|
1678 | int |
---|
1679 | ClpSimplexDual::checkUnbounded(CoinIndexedVector * ray, |
---|
1680 | CoinIndexedVector * spare, |
---|
1681 | double changeCost) |
---|
1682 | { |
---|
1683 | int status=2; // say unbounded |
---|
1684 | factorization_->updateColumn(spare,ray); |
---|
1685 | // get reduced cost |
---|
1686 | int i; |
---|
1687 | int number=ray->getNumElements(); |
---|
1688 | int * index = ray->getIndices(); |
---|
1689 | double * array = ray->denseVector(); |
---|
1690 | for (i=0;i<number;i++) { |
---|
1691 | int iRow=index[i]; |
---|
1692 | int iPivot=pivotVariable_[iRow]; |
---|
1693 | changeCost -= cost(iPivot)*array[iRow]; |
---|
1694 | } |
---|
1695 | double way; |
---|
1696 | if (changeCost>0.0) { |
---|
1697 | //try going down |
---|
1698 | way=1.0; |
---|
1699 | } else if (changeCost<0.0) { |
---|
1700 | //try going up |
---|
1701 | way=-1.0; |
---|
1702 | } else { |
---|
1703 | #ifdef CLP_DEBUG |
---|
1704 | printf("can't decide on up or down\n"); |
---|
1705 | #endif |
---|
1706 | way=0.0; |
---|
1707 | status=-3; |
---|
1708 | } |
---|
1709 | double movement=1.0e10*way; // some largish number |
---|
1710 | double zeroTolerance = 1.0e-14*dualBound_; |
---|
1711 | for (i=0;i<number;i++) { |
---|
1712 | int iRow=index[i]; |
---|
1713 | int iPivot=pivotVariable_[iRow]; |
---|
1714 | double arrayValue = array[iRow]; |
---|
1715 | if (fabs(arrayValue)<zeroTolerance) |
---|
1716 | arrayValue=0.0; |
---|
1717 | double newValue=solution(iPivot)+movement*arrayValue; |
---|
1718 | if (newValue>upper(iPivot)+primalTolerance_|| |
---|
1719 | newValue<lower(iPivot)-primalTolerance_) |
---|
1720 | status=-3; // not unbounded |
---|
1721 | } |
---|
1722 | if (status==2) { |
---|
1723 | // create ray |
---|
1724 | delete [] ray_; |
---|
1725 | ray_ = new double [numberColumns_]; |
---|
1726 | CoinFillN(ray_,numberColumns_,0.0); |
---|
1727 | for (i=0;i<number;i++) { |
---|
1728 | int iRow=index[i]; |
---|
1729 | int iPivot=pivotVariable_[iRow]; |
---|
1730 | double arrayValue = array[iRow]; |
---|
1731 | if (iPivot<numberColumns_&&fabs(arrayValue)>=zeroTolerance) |
---|
1732 | ray_[iPivot] = way* array[iRow]; |
---|
1733 | } |
---|
1734 | } |
---|
1735 | ray->clear(); |
---|
1736 | return status; |
---|
1737 | } |
---|
1738 | /* Checks if finished. Updates status */ |
---|
1739 | void |
---|
1740 | ClpSimplexDual::statusOfProblemInDual(int & lastCleaned,int type, |
---|
1741 | ClpSimplexProgress &progress) |
---|
1742 | { |
---|
1743 | if (type==2) { |
---|
1744 | // trouble - restore solution |
---|
1745 | memcpy(status_ ,saveStatus_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
1746 | memcpy(rowActivityWork_,savedSolution_+numberColumns_ , |
---|
1747 | numberRows_*sizeof(double)); |
---|
1748 | memcpy(columnActivityWork_,savedSolution_ , |
---|
1749 | numberColumns_*sizeof(double)); |
---|
1750 | forceFactorization_=1; // a bit drastic but .. |
---|
1751 | changeMade_++; // say something changed |
---|
1752 | } |
---|
1753 | int tentativeStatus = problemStatus_; |
---|
1754 | double changeCost; |
---|
1755 | |
---|
1756 | if (problemStatus_>-3) { |
---|
1757 | // factorize |
---|
1758 | // later on we will need to recover from singularities |
---|
1759 | // also we could skip if first time |
---|
1760 | // save dual weights |
---|
1761 | dualRowPivot_->saveWeights(this,1); |
---|
1762 | if (type) { |
---|
1763 | // is factorization okay? |
---|
1764 | if (internalFactorize(1)) { |
---|
1765 | // no - restore previous basis |
---|
1766 | assert (type==1); |
---|
1767 | changeMade_++; // say something changed |
---|
1768 | memcpy(status_ ,saveStatus_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
1769 | memcpy(rowActivityWork_,savedSolution_+numberColumns_ , |
---|
1770 | numberRows_*sizeof(double)); |
---|
1771 | memcpy(columnActivityWork_,savedSolution_ , |
---|
1772 | numberColumns_*sizeof(double)); |
---|
1773 | // get correct bounds on all variables |
---|
1774 | double dummyChangeCost=0.0; |
---|
1775 | changeBounds(true,rowArray_[2],dummyChangeCost); |
---|
1776 | // throw away change |
---|
1777 | rowArray_[2]->clear(); |
---|
1778 | forceFactorization_=1; // a bit drastic but .. |
---|
1779 | type = 2; |
---|
1780 | assert (internalFactorize(1)==0); |
---|
1781 | } |
---|
1782 | } |
---|
1783 | problemStatus_=-3; |
---|
1784 | } |
---|
1785 | // at this stage status is -3 or -4 if looks infeasible |
---|
1786 | // get primal and dual solutions |
---|
1787 | gutsOfSolution(rowActivityWork_,columnActivityWork_); |
---|
1788 | // Check if looping |
---|
1789 | int loop = progress.looping(); |
---|
1790 | bool situationChanged=false; |
---|
1791 | if (loop>=0) { |
---|
1792 | problemStatus_ = loop; //exit if in loop |
---|
1793 | return; |
---|
1794 | } else if (loop<-1) { |
---|
1795 | // something may have changed |
---|
1796 | gutsOfSolution(rowActivityWork_,columnActivityWork_); |
---|
1797 | situationChanged=true; |
---|
1798 | } |
---|
1799 | progressFlag_ = 0; //reset progress flag |
---|
1800 | #ifdef CLP_DEBUG |
---|
1801 | if (!rowScale_&&(handler_->logLevel()&32)) { |
---|
1802 | double * objectiveSimplex |
---|
1803 | = copyOfArray(objective_,numberColumns_,0.0); |
---|
1804 | double * rowObjectiveSimplex |
---|
1805 | = copyOfArray(rowObjective_,numberRows_,0.0); |
---|
1806 | int i; |
---|
1807 | double largest; |
---|
1808 | largest=0.0; |
---|
1809 | for (i=0;i<numberRows_;i++) { |
---|
1810 | rowObjectiveSimplex[i] *= optimizationDirection_; |
---|
1811 | double difference = fabs(rowObjectiveWork_[i]-rowObjectiveSimplex[i]); |
---|
1812 | if (difference>largest) |
---|
1813 | largest=difference; |
---|
1814 | } |
---|
1815 | for (i=0;i<numberColumns_;i++) { |
---|
1816 | objectiveSimplex[i] *= optimizationDirection_; |
---|
1817 | double difference = fabs(objectiveWork_[i]-objectiveSimplex[i]); |
---|
1818 | if (difference>largest) |
---|
1819 | largest=difference; |
---|
1820 | } |
---|
1821 | if ((handler_->logLevel()&16)) |
---|
1822 | printf("difference in obj %g\n",largest); |
---|
1823 | delete [] objectiveSimplex; |
---|
1824 | delete [] rowObjectiveSimplex; |
---|
1825 | } |
---|
1826 | #endif |
---|
1827 | handler_->message(CLP_SIMPLEX_STATUS,messages_) |
---|
1828 | <<numberIterations_<<objectiveValue(); |
---|
1829 | handler_->printing(sumPrimalInfeasibilities_>0.0) |
---|
1830 | <<sumPrimalInfeasibilities_<<numberPrimalInfeasibilities_; |
---|
1831 | handler_->printing(sumDualInfeasibilities_>0.0) |
---|
1832 | <<sumDualInfeasibilities_<<numberDualInfeasibilities_; |
---|
1833 | handler_->printing(numberDualInfeasibilitiesWithoutFree_ |
---|
1834 | <numberDualInfeasibilities_) |
---|
1835 | <<numberDualInfeasibilities_- |
---|
1836 | numberDualInfeasibilitiesWithoutFree_; |
---|
1837 | handler_->message()<<CoinMessageEol; |
---|
1838 | |
---|
1839 | while (problemStatus_<=-3) { |
---|
1840 | bool cleanDuals=false; |
---|
1841 | int numberChangedBounds=0; |
---|
1842 | int doOriginalTolerance=0; |
---|
1843 | if ( lastCleaned==numberIterations_) |
---|
1844 | doOriginalTolerance=1; |
---|
1845 | // check optimal |
---|
1846 | // give code benefit of doubt |
---|
1847 | if (sumOfRelaxedDualInfeasibilities_ == 0.0&& |
---|
1848 | sumOfRelaxedPrimalInfeasibilities_ == 0.0) { |
---|
1849 | // say optimal (with these bounds etc) |
---|
1850 | numberDualInfeasibilities_ = 0; |
---|
1851 | sumDualInfeasibilities_ = 0.0; |
---|
1852 | numberPrimalInfeasibilities_ = 0; |
---|
1853 | sumPrimalInfeasibilities_ = 0.0; |
---|
1854 | } |
---|
1855 | if (dualFeasible()||problemStatus_==-4) { |
---|
1856 | if (primalFeasible()) { |
---|
1857 | // may be optimal - or may be bounds are wrong |
---|
1858 | handler_->message(CLP_DUAL_BOUNDS,messages_) |
---|
1859 | <<dualBound_ |
---|
1860 | <<CoinMessageEol; |
---|
1861 | // save solution in case unbounded |
---|
1862 | CoinDisjointCopyN(columnActivityWork_,numberColumns_, |
---|
1863 | columnArray_[0]->denseVector()); |
---|
1864 | CoinDisjointCopyN(rowActivityWork_,numberRows_, |
---|
1865 | rowArray_[2]->denseVector()); |
---|
1866 | numberChangedBounds=changeBounds(false,rowArray_[0],changeCost); |
---|
1867 | if (numberChangedBounds<=0) { |
---|
1868 | //looks optimal - do we need to reset tolerance |
---|
1869 | if (lastCleaned<numberIterations_&&numberTimesOptimal_<4) { |
---|
1870 | doOriginalTolerance=2; |
---|
1871 | numberTimesOptimal_++; |
---|
1872 | changeMade_++; // say something changed |
---|
1873 | if (numberTimesOptimal_==1) { |
---|
1874 | dualTolerance_ = min(dualTolerance_,1.0e-8); |
---|
1875 | // better to have small tolerance even if slower |
---|
1876 | factorization_->zeroTolerance(1.0e-15); |
---|
1877 | } |
---|
1878 | cleanDuals=true; |
---|
1879 | } else { |
---|
1880 | problemStatus_=0; // optimal |
---|
1881 | if (lastCleaned<numberIterations_) { |
---|
1882 | handler_->message(CLP_SIMPLEX_GIVINGUP,messages_) |
---|
1883 | <<CoinMessageEol; |
---|
1884 | } |
---|
1885 | } |
---|
1886 | } else { |
---|
1887 | cleanDuals=true; |
---|
1888 | if (doOriginalTolerance==1) { |
---|
1889 | // check unbounded |
---|
1890 | problemStatus_ = checkUnbounded(rowArray_[0],rowArray_[1], |
---|
1891 | changeCost); |
---|
1892 | if (problemStatus_==2) { |
---|
1893 | // it is unbounded - restore solution |
---|
1894 | // but first add in changes to non-basic |
---|
1895 | int iColumn; |
---|
1896 | double * original = columnArray_[0]->denseVector(); |
---|
1897 | for (iColumn=0;iColumn<numberColumns_;iColumn++) { |
---|
1898 | if(getColumnStatus(iColumn)!= ClpSimplex::basic) |
---|
1899 | ray_[iColumn] += |
---|
1900 | columnActivityWork_[iColumn]-original[iColumn]; |
---|
1901 | columnActivityWork_[iColumn] = original[iColumn]; |
---|
1902 | } |
---|
1903 | CoinDisjointCopyN(rowArray_[2]->denseVector(),numberRows_, |
---|
1904 | rowActivityWork_); |
---|
1905 | } |
---|
1906 | } else { |
---|
1907 | doOriginalTolerance=2; |
---|
1908 | rowArray_[0]->clear(); |
---|
1909 | } |
---|
1910 | } |
---|
1911 | CoinFillN(columnArray_[0]->denseVector(),numberColumns_,0.0); |
---|
1912 | CoinFillN(rowArray_[2]->denseVector(),numberRows_,0.0); |
---|
1913 | } |
---|
1914 | if (problemStatus_==-4) { |
---|
1915 | // may be infeasible - or may be bounds are wrong |
---|
1916 | handler_->message(CLP_DUAL_CHECKB,messages_) |
---|
1917 | <<dualBound_ |
---|
1918 | <<CoinMessageEol; |
---|
1919 | numberChangedBounds=changeBounds(false,NULL,changeCost); |
---|
1920 | if (numberChangedBounds<=0||dualBound_>1.0e20|| |
---|
1921 | (largestPrimalError_>1.0&&dualBound_>1.0e17)) { |
---|
1922 | problemStatus_=1; // infeasible |
---|
1923 | } else { |
---|
1924 | problemStatus_=-1; //iterate |
---|
1925 | cleanDuals=true; |
---|
1926 | doOriginalTolerance=2; |
---|
1927 | // and delete ray which has been created |
---|
1928 | delete [] ray_; |
---|
1929 | ray_ = NULL; |
---|
1930 | } |
---|
1931 | } |
---|
1932 | } else { |
---|
1933 | cleanDuals=true; |
---|
1934 | } |
---|
1935 | if (problemStatus_<0) { |
---|
1936 | if (doOriginalTolerance==2) { |
---|
1937 | // put back original tolerance |
---|
1938 | lastCleaned=numberIterations_; |
---|
1939 | handler_->message(CLP_DUAL_ORIGINAL,messages_) |
---|
1940 | <<CoinMessageEol; |
---|
1941 | |
---|
1942 | perturbation_=102; // stop any perturbations |
---|
1943 | createRim(4); |
---|
1944 | // make sure duals are current |
---|
1945 | computeDuals(); |
---|
1946 | // put back bounds as they were if was optimal |
---|
1947 | if (doOriginalTolerance==2) { |
---|
1948 | changeMade_++; // say something changed |
---|
1949 | changeBounds(true,NULL,changeCost); |
---|
1950 | cleanDuals=true; |
---|
1951 | } |
---|
1952 | } |
---|
1953 | if (cleanDuals) { |
---|
1954 | // make sure dual feasible |
---|
1955 | // look at all rows and columns |
---|
1956 | rowArray_[0]->clear(); |
---|
1957 | CoinIotaN(rowArray_[0]->getIndices(),numberRows_,0); |
---|
1958 | rowArray_[0]->setNumElements(numberRows_); |
---|
1959 | columnArray_[0]->clear(); |
---|
1960 | CoinIotaN(columnArray_[0]->getIndices(),numberColumns_,0); |
---|
1961 | columnArray_[0]->setNumElements(numberColumns_); |
---|
1962 | double objectiveChange=0.0; |
---|
1963 | updateDualsInDual(rowArray_[0],columnArray_[0],rowArray_[1], |
---|
1964 | 0.0,objectiveChange); |
---|
1965 | // for now - recompute all |
---|
1966 | gutsOfSolution(rowActivityWork_,columnActivityWork_); |
---|
1967 | assert(numberDualInfeasibilitiesWithoutFree_==0); |
---|
1968 | if (numberDualInfeasibilities_) { |
---|
1969 | // bad free variables |
---|
1970 | if (primalFeasible()) { |
---|
1971 | std::cerr<<"Free variable problem?"<<std::endl; |
---|
1972 | abort(); // what now |
---|
1973 | } |
---|
1974 | problemStatus_=-1; // carry on as normal |
---|
1975 | } |
---|
1976 | } else { |
---|
1977 | // iterate |
---|
1978 | problemStatus_=-1; |
---|
1979 | } |
---|
1980 | } |
---|
1981 | } |
---|
1982 | if (type==0||type==1) { |
---|
1983 | if (!type) { |
---|
1984 | // create save arrays |
---|
1985 | delete [] saveStatus_; |
---|
1986 | delete [] savedSolution_; |
---|
1987 | saveStatus_ = new unsigned char [numberRows_+numberColumns_]; |
---|
1988 | savedSolution_ = new double [numberRows_+numberColumns_]; |
---|
1989 | } |
---|
1990 | // save arrays |
---|
1991 | memcpy(saveStatus_,status_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
1992 | memcpy(savedSolution_+numberColumns_ ,rowActivityWork_, |
---|
1993 | numberRows_*sizeof(double)); |
---|
1994 | memcpy(savedSolution_ ,columnActivityWork_,numberColumns_*sizeof(double)); |
---|
1995 | } |
---|
1996 | |
---|
1997 | // restore weights (if saved) - also recompute infeasibility list |
---|
1998 | if (tentativeStatus>-3) |
---|
1999 | dualRowPivot_->saveWeights(this,(type <2) ? 2 : 4); |
---|
2000 | else |
---|
2001 | dualRowPivot_->saveWeights(this,3); |
---|
2002 | // unflag all variables (we may want to wait a bit?) |
---|
2003 | int iRow; |
---|
2004 | for (iRow=0;iRow<numberRows_;iRow++) { |
---|
2005 | int iPivot=pivotVariable_[iRow]; |
---|
2006 | clearFlagged(iPivot); |
---|
2007 | } |
---|
2008 | if (problemStatus_<0&&!changeMade_) { |
---|
2009 | problemStatus_=4; // unknown |
---|
2010 | } |
---|
2011 | |
---|
2012 | } |
---|
2013 | /* While updateDualsInDual sees what effect is of flip |
---|
2014 | this does actual flipping. |
---|
2015 | If change >0.0 then value in array >0.0 => from lower to upper |
---|
2016 | */ |
---|
2017 | void |
---|
2018 | ClpSimplexDual::flipBounds(CoinIndexedVector * rowArray, |
---|
2019 | CoinIndexedVector * columnArray, |
---|
2020 | double change) |
---|
2021 | { |
---|
2022 | double * work; |
---|
2023 | int number; |
---|
2024 | int * which; |
---|
2025 | |
---|
2026 | int iSection; |
---|
2027 | |
---|
2028 | for (iSection=0;iSection<2;iSection++) { |
---|
2029 | int i; |
---|
2030 | double * solution = solutionRegion(iSection); |
---|
2031 | double * lower = lowerRegion(iSection); |
---|
2032 | double * upper = upperRegion(iSection); |
---|
2033 | int addSequence; |
---|
2034 | if (!iSection) { |
---|
2035 | work = rowArray->denseVector(); |
---|
2036 | number = rowArray->getNumElements(); |
---|
2037 | which = rowArray->getIndices(); |
---|
2038 | addSequence = numberColumns_; |
---|
2039 | } else { |
---|
2040 | work = columnArray->denseVector(); |
---|
2041 | number = columnArray->getNumElements(); |
---|
2042 | which = columnArray->getIndices(); |
---|
2043 | addSequence = 0; |
---|
2044 | } |
---|
2045 | |
---|
2046 | for (i=0;i<number;i++) { |
---|
2047 | int iSequence = which[i]; |
---|
2048 | #ifndef NDEBUG |
---|
2049 | double value = work[iSequence]*change; |
---|
2050 | #endif |
---|
2051 | work[iSequence]=0.0; |
---|
2052 | Status status = getStatus(iSequence+addSequence); |
---|
2053 | |
---|
2054 | switch(status) { |
---|
2055 | |
---|
2056 | case ClpSimplex::basic: |
---|
2057 | case ClpSimplex::isFree: |
---|
2058 | case ClpSimplex::superBasic: |
---|
2059 | break; |
---|
2060 | case ClpSimplex::atUpperBound: |
---|
2061 | assert (value<=0.0); |
---|
2062 | // to lower bound |
---|
2063 | setStatus(iSequence+addSequence,ClpSimplex::atLowerBound); |
---|
2064 | solution[iSequence] = lower[iSequence]; |
---|
2065 | break; |
---|
2066 | case ClpSimplex::atLowerBound: |
---|
2067 | assert (value>=0.0); |
---|
2068 | // to upper bound |
---|
2069 | setStatus(iSequence+addSequence,ClpSimplex::atUpperBound); |
---|
2070 | solution[iSequence] = upper[iSequence]; |
---|
2071 | break; |
---|
2072 | } |
---|
2073 | } |
---|
2074 | } |
---|
2075 | rowArray->setNumElements(0); |
---|
2076 | columnArray->setNumElements(0); |
---|
2077 | } |
---|
2078 | // Restores bound to original bound |
---|
2079 | void |
---|
2080 | ClpSimplexDual::originalBound( int iSequence) |
---|
2081 | { |
---|
2082 | if (iSequence>=numberColumns_) { |
---|
2083 | // rows |
---|
2084 | int iRow = iSequence-numberColumns_; |
---|
2085 | rowLowerWork_[iRow]=rowLower_[iRow]; |
---|
2086 | rowUpperWork_[iRow]=rowUpper_[iRow]; |
---|
2087 | if (rowScale_) { |
---|
2088 | if (rowLowerWork_[iRow]>-1.0e50) |
---|
2089 | rowLowerWork_[iRow] *= rowScale_[iRow]; |
---|
2090 | if (rowUpperWork_[iRow]<1.0e50) |
---|
2091 | rowUpperWork_[iRow] *= rowScale_[iRow]; |
---|
2092 | } |
---|
2093 | } else { |
---|
2094 | // columns |
---|
2095 | columnLowerWork_[iSequence]=columnLower_[iSequence]; |
---|
2096 | columnUpperWork_[iSequence]=columnUpper_[iSequence]; |
---|
2097 | if (rowScale_) { |
---|
2098 | double multiplier = 1.0/columnScale_[iSequence]; |
---|
2099 | if (columnLowerWork_[iSequence]>-1.0e50) |
---|
2100 | columnLowerWork_[iSequence] *= multiplier; |
---|
2101 | if (columnUpperWork_[iSequence]<1.0e50) |
---|
2102 | columnUpperWork_[iSequence] *= multiplier; |
---|
2103 | } |
---|
2104 | } |
---|
2105 | setFakeBound(iSequence,ClpSimplex::noFake); |
---|
2106 | } |
---|
2107 | /* As changeBounds but just changes new bounds for a single variable. |
---|
2108 | Returns true if change */ |
---|
2109 | bool |
---|
2110 | ClpSimplexDual::changeBound( int iSequence) |
---|
2111 | { |
---|
2112 | // old values |
---|
2113 | double oldLower=lower_[iSequence]; |
---|
2114 | double oldUpper=upper_[iSequence]; |
---|
2115 | double value=solution_[iSequence]; |
---|
2116 | bool modified=false; |
---|
2117 | originalBound(iSequence); |
---|
2118 | // original values |
---|
2119 | double lowerValue=lower_[iSequence]; |
---|
2120 | double upperValue=upper_[iSequence]; |
---|
2121 | // back to altered values |
---|
2122 | lower_[iSequence] = oldLower; |
---|
2123 | upper_[iSequence] = oldUpper; |
---|
2124 | if (value==oldLower) { |
---|
2125 | if (upperValue > oldLower + dualBound_) { |
---|
2126 | upper_[iSequence]=oldLower+dualBound_; |
---|
2127 | setFakeBound(iSequence,ClpSimplex::upperFake); |
---|
2128 | modified=true; |
---|
2129 | } |
---|
2130 | } else if (value==oldUpper) { |
---|
2131 | if (lowerValue < oldUpper - dualBound_) { |
---|
2132 | lower_[iSequence]=oldUpper-dualBound_; |
---|
2133 | setFakeBound(iSequence,ClpSimplex::lowerFake); |
---|
2134 | modified=true; |
---|
2135 | } |
---|
2136 | } else { |
---|
2137 | assert(value==oldLower||value==oldUpper); |
---|
2138 | } |
---|
2139 | return modified; |
---|
2140 | } |
---|
2141 | // Perturbs problem |
---|
2142 | void |
---|
2143 | ClpSimplexDual::perturb() |
---|
2144 | { |
---|
2145 | if (perturbation_>100) |
---|
2146 | return; //perturbed already |
---|
2147 | int iRow,iColumn; |
---|
2148 | // dual perturbation |
---|
2149 | double perturbation=1.0e-20; |
---|
2150 | // maximum fraction of cost to perturb |
---|
2151 | double maximumFraction = 1.0e-4; |
---|
2152 | if (perturbation_==100) { |
---|
2153 | perturbation = 1.0e-4; |
---|
2154 | for (iRow=0;iRow<numberRows_;iRow++) { |
---|
2155 | double value = fabs(rowActivityWork_[iRow]*rowObjectiveWork_[iRow]); |
---|
2156 | perturbation = max(perturbation,value); |
---|
2157 | } |
---|
2158 | for (iColumn=0;iColumn<numberColumns_;iColumn++) { |
---|
2159 | double value = |
---|
2160 | fabs(columnActivityWork_[iColumn]*objectiveWork_[iColumn]); |
---|
2161 | perturbation = max(perturbation,value); |
---|
2162 | } |
---|
2163 | perturbation *= 1.0e-8; |
---|
2164 | } else if (perturbation_<100) { |
---|
2165 | perturbation = pow(10.0,perturbation_); |
---|
2166 | // user is in charge |
---|
2167 | maximumFraction = 1.0e100; |
---|
2168 | } |
---|
2169 | // modify costs |
---|
2170 | handler_->message(CLP_SIMPLEX_PERTURB,messages_) |
---|
2171 | <<perturbation |
---|
2172 | <<CoinMessageEol; |
---|
2173 | for (iRow=0;iRow<numberRows_;iRow++) { |
---|
2174 | double value = perturbation; |
---|
2175 | double currentValue = rowObjectiveWork_[iRow]; |
---|
2176 | value = min(value,maximumFraction*fabs(currentValue)+1.0e-6); |
---|
2177 | if (rowLowerWork_[iRow]>-largeValue_) { |
---|
2178 | if (fabs(rowLowerWork_[iRow])<fabs(rowUpperWork_[iRow])) |
---|
2179 | value *= drand48(); |
---|
2180 | else |
---|
2181 | value *= -drand48(); |
---|
2182 | } else if (rowUpperWork_[iRow]<largeValue_) { |
---|
2183 | value *= -drand48(); |
---|
2184 | } else { |
---|
2185 | value=0.0; |
---|
2186 | } |
---|
2187 | rowObjectiveWork_[iRow] += value; |
---|
2188 | } |
---|
2189 | for (iColumn=0;iColumn<numberColumns_;iColumn++) { |
---|
2190 | double value = perturbation; |
---|
2191 | double currentValue = objectiveWork_[iColumn]; |
---|
2192 | value = min(value,maximumFraction*fabs(currentValue)+1.0e-6); |
---|
2193 | if (columnLowerWork_[iColumn]>-largeValue_) { |
---|
2194 | if (fabs(columnLowerWork_[iColumn])< |
---|
2195 | fabs(columnUpperWork_[iColumn])) |
---|
2196 | value *= drand48(); |
---|
2197 | else |
---|
2198 | value *= -drand48(); |
---|
2199 | } else if (columnUpperWork_[iColumn]<largeValue_) { |
---|
2200 | value *= -drand48(); |
---|
2201 | } else { |
---|
2202 | value=0.0; |
---|
2203 | } |
---|
2204 | objectiveWork_[iColumn] += value; |
---|
2205 | } |
---|
2206 | // say perturbed |
---|
2207 | perturbation_=102; |
---|
2208 | |
---|
2209 | } |
---|
2210 | /* For strong branching. On input lower and upper are new bounds |
---|
2211 | while on output they are change in objective function values |
---|
2212 | (>1.0e50 infeasible). |
---|
2213 | Return code is 0 if nothing interesting, -1 if infeasible both |
---|
2214 | ways and +1 if infeasible one way (check values to see which one(s)) |
---|
2215 | */ |
---|
2216 | int ClpSimplexDual::strongBranching(int numberVariables,const int * variables, |
---|
2217 | double * newLower, double * newUpper, |
---|
2218 | bool stopOnFirstInfeasible, |
---|
2219 | bool alwaysFinish) |
---|
2220 | { |
---|
2221 | int i; |
---|
2222 | int returnCode=0; |
---|
2223 | double saveObjectiveValue = objectiveValue_; |
---|
2224 | #if 1 |
---|
2225 | algorithm_ = -1; |
---|
2226 | |
---|
2227 | //scaling(false); |
---|
2228 | |
---|
2229 | // put in standard form (and make row copy) |
---|
2230 | // create modifiable copies of model rim and do optional scaling |
---|
2231 | createRim(7+8+16,true); |
---|
2232 | |
---|
2233 | // change newLower and newUpper if scaled |
---|
2234 | |
---|
2235 | // Do initial factorization |
---|
2236 | // and set certain stuff |
---|
2237 | // We can either set increasing rows so ...IsBasic gives pivot row |
---|
2238 | // or we can just increment iBasic one by one |
---|
2239 | // for now let ...iBasic give pivot row |
---|
2240 | factorization_->increasingRows(2); |
---|
2241 | // row activities have negative sign |
---|
2242 | factorization_->slackValue(-1.0); |
---|
2243 | factorization_->zeroTolerance(1.0e-13); |
---|
2244 | // save if sparse factorization wanted |
---|
2245 | int saveSparse = factorization_->sparseThreshold(); |
---|
2246 | |
---|
2247 | int factorizationStatus = internalFactorize(0); |
---|
2248 | if (factorizationStatus<0) |
---|
2249 | return 1; // some error |
---|
2250 | else if (factorizationStatus) |
---|
2251 | handler_->message(CLP_SINGULARITIES,messages_) |
---|
2252 | <<factorizationStatus |
---|
2253 | <<CoinMessageEol; |
---|
2254 | if (saveSparse) { |
---|
2255 | // use default at present |
---|
2256 | factorization_->sparseThreshold(0); |
---|
2257 | factorization_->goSparse(); |
---|
2258 | } |
---|
2259 | |
---|
2260 | // save stuff |
---|
2261 | ClpFactorization saveFactorization(*factorization_); |
---|
2262 | // save basis and solution |
---|
2263 | double * saveSolution = new double[numberRows_+numberColumns_]; |
---|
2264 | memcpy(saveSolution,solution_, |
---|
2265 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2266 | unsigned char * saveStatus = |
---|
2267 | new unsigned char [numberRows_+numberColumns_]; |
---|
2268 | memcpy(saveStatus,status_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
2269 | // save bounds as createRim makes clean copies |
---|
2270 | double * saveLower = new double[numberRows_+numberColumns_]; |
---|
2271 | memcpy(saveLower,lower_, |
---|
2272 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2273 | double * saveUpper = new double[numberRows_+numberColumns_]; |
---|
2274 | memcpy(saveUpper,upper_, |
---|
2275 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2276 | double * saveObjective = new double[numberRows_+numberColumns_]; |
---|
2277 | memcpy(saveObjective,cost_, |
---|
2278 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2279 | int * savePivot = new int [numberRows_]; |
---|
2280 | memcpy(savePivot, pivotVariable_, numberRows_*sizeof(int)); |
---|
2281 | // need to save/restore weights. |
---|
2282 | |
---|
2283 | for (i=0;i<numberVariables;i++) { |
---|
2284 | int iColumn = variables[i]; |
---|
2285 | double objectiveChange; |
---|
2286 | double saveBound; |
---|
2287 | |
---|
2288 | // try down |
---|
2289 | |
---|
2290 | saveBound = columnUpper_[iColumn]; |
---|
2291 | // external view - in case really getting optimal |
---|
2292 | columnUpper_[iColumn] = newUpper[i]; |
---|
2293 | if (scalingFlag_<=0) |
---|
2294 | upper_[iColumn] = newUpper[i]; |
---|
2295 | else |
---|
2296 | upper_[iColumn] = newUpper[i]/columnScale_[iColumn]; // scale |
---|
2297 | // Start of fast iterations |
---|
2298 | int status = fastDual(alwaysFinish); |
---|
2299 | |
---|
2300 | // restore |
---|
2301 | memcpy(solution_,saveSolution, |
---|
2302 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2303 | memcpy(status_,saveStatus,(numberColumns_+numberRows_)*sizeof(char)); |
---|
2304 | memcpy(lower_,saveLower, |
---|
2305 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2306 | memcpy(upper_,saveUpper, |
---|
2307 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2308 | memcpy(cost_,saveObjective, |
---|
2309 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2310 | columnUpper_[iColumn] = saveBound; |
---|
2311 | memcpy(pivotVariable_, savePivot, numberRows_*sizeof(int)); |
---|
2312 | delete factorization_; |
---|
2313 | factorization_ = new ClpFactorization(saveFactorization); |
---|
2314 | |
---|
2315 | if (status||(problemStatus_==0&&!isDualObjectiveLimitReached())) { |
---|
2316 | objectiveChange = objectiveValue_-saveObjectiveValue; |
---|
2317 | } else { |
---|
2318 | objectiveChange = 1.0e100; |
---|
2319 | } |
---|
2320 | newUpper[i]=objectiveChange; |
---|
2321 | #ifdef CLP_DEBUG |
---|
2322 | printf("down on %d costs %g\n",iColumn,objectiveChange); |
---|
2323 | #endif |
---|
2324 | |
---|
2325 | // try up |
---|
2326 | |
---|
2327 | saveBound = columnLower_[iColumn]; |
---|
2328 | // external view - in case really getting optimal |
---|
2329 | columnLower_[iColumn] = newLower[i]; |
---|
2330 | if (scalingFlag_<=0) |
---|
2331 | lower_[iColumn] = newLower[i]; |
---|
2332 | else |
---|
2333 | lower_[iColumn] = newLower[i]/columnScale_[iColumn]; // scale |
---|
2334 | // Start of fast iterations |
---|
2335 | status = fastDual(alwaysFinish); |
---|
2336 | |
---|
2337 | // restore |
---|
2338 | memcpy(solution_,saveSolution, |
---|
2339 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2340 | memcpy(status_,saveStatus,(numberColumns_+numberRows_)*sizeof(char)); |
---|
2341 | memcpy(lower_,saveLower, |
---|
2342 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2343 | memcpy(upper_,saveUpper, |
---|
2344 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2345 | memcpy(cost_,saveObjective, |
---|
2346 | (numberRows_+numberColumns_)*sizeof(double)); |
---|
2347 | columnLower_[iColumn] = saveBound; |
---|
2348 | memcpy(pivotVariable_, savePivot, numberRows_*sizeof(int)); |
---|
2349 | delete factorization_; |
---|
2350 | factorization_ = new ClpFactorization(saveFactorization); |
---|
2351 | |
---|
2352 | if (status||(problemStatus_==0&&!isDualObjectiveLimitReached())) { |
---|
2353 | objectiveChange = objectiveValue_-saveObjectiveValue; |
---|
2354 | } else { |
---|
2355 | objectiveChange = 1.0e100; |
---|
2356 | } |
---|
2357 | newLower[i]=objectiveChange; |
---|
2358 | #ifdef CLP_DEBUG |
---|
2359 | printf("up on %d costs %g\n",iColumn,objectiveChange); |
---|
2360 | #endif |
---|
2361 | |
---|
2362 | /* Possibilities are: |
---|
2363 | Both sides feasible - store |
---|
2364 | Neither side feasible - set objective high and exit |
---|
2365 | One side feasible - change bounds and resolve |
---|
2366 | */ |
---|
2367 | if (newUpper[i]<1.0e100) { |
---|
2368 | if(newLower[i]<1.0e100) { |
---|
2369 | // feasible - no action |
---|
2370 | } else { |
---|
2371 | // up feasible, down infeasible |
---|
2372 | returnCode=1; |
---|
2373 | if (stopOnFirstInfeasible) |
---|
2374 | break; |
---|
2375 | } |
---|
2376 | } else { |
---|
2377 | if(newLower[i]<1.0e100) { |
---|
2378 | // down feasible, up infeasible |
---|
2379 | returnCode=1; |
---|
2380 | if (stopOnFirstInfeasible) |
---|
2381 | break; |
---|
2382 | } else { |
---|
2383 | // neither side feasible |
---|
2384 | returnCode=-1; |
---|
2385 | break; |
---|
2386 | } |
---|
2387 | } |
---|
2388 | } |
---|
2389 | delete [] saveSolution; |
---|
2390 | delete [] saveLower; |
---|
2391 | delete [] saveUpper; |
---|
2392 | delete [] saveObjective; |
---|
2393 | delete [] saveStatus; |
---|
2394 | delete [] savePivot; |
---|
2395 | |
---|
2396 | // at present we are leaving factorization around |
---|
2397 | // maybe we should empty it |
---|
2398 | deleteRim(); |
---|
2399 | factorization_->sparseThreshold(saveSparse); |
---|
2400 | #else |
---|
2401 | // save basis and solution |
---|
2402 | double * rowSolution = new double[numberRows_]; |
---|
2403 | memcpy(rowSolution,rowActivity_,numberRows_*sizeof(double)); |
---|
2404 | double * columnSolution = new double[numberColumns_]; |
---|
2405 | memcpy(columnSolution,columnActivity_,numberColumns_*sizeof(double)); |
---|
2406 | unsigned char * saveStatus = |
---|
2407 | new unsigned char [numberRows_+numberColumns_]; |
---|
2408 | if (!status_) { |
---|
2409 | // odd but anyway setup all slack basis |
---|
2410 | status_ = new unsigned char [numberColumns_+numberRows_]; |
---|
2411 | memset(status_,0,(numberColumns_+numberRows_)*sizeof(char)); |
---|
2412 | } |
---|
2413 | memcpy(saveStatus,status_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
2414 | for (i=0;i<numberVariables;i++) { |
---|
2415 | int iColumn = variables[i]; |
---|
2416 | double objectiveChange; |
---|
2417 | |
---|
2418 | // try down |
---|
2419 | |
---|
2420 | double saveUpper = columnUpper_[iColumn]; |
---|
2421 | columnUpper_[iColumn] = newUpper[i]; |
---|
2422 | dual(); |
---|
2423 | |
---|
2424 | // restore |
---|
2425 | columnUpper_[iColumn] = saveUpper; |
---|
2426 | memcpy(rowActivity_,rowSolution,numberRows_*sizeof(double)); |
---|
2427 | memcpy(columnActivity_,columnSolution,numberColumns_*sizeof(double)); |
---|
2428 | memcpy(status_,saveStatus,(numberColumns_+numberRows_)*sizeof(char)); |
---|
2429 | |
---|
2430 | if (problemStatus_==0&&!isDualObjectiveLimitReached()) { |
---|
2431 | objectiveChange = objectiveValue_-saveObjectiveValue; |
---|
2432 | } else { |
---|
2433 | objectiveChange = 1.0e100; |
---|
2434 | } |
---|
2435 | newUpper[i]=objectiveChange; |
---|
2436 | #ifdef CLP_DEBUG |
---|
2437 | printf("down on %d costs %g\n",iColumn,objectiveChange); |
---|
2438 | #endif |
---|
2439 | |
---|
2440 | // try up |
---|
2441 | |
---|
2442 | double saveLower = columnLower_[iColumn]; |
---|
2443 | columnLower_[iColumn] = newLower[i]; |
---|
2444 | dual(); |
---|
2445 | |
---|
2446 | // restore |
---|
2447 | columnLower_[iColumn] = saveLower; |
---|
2448 | memcpy(rowActivity_,rowSolution,numberRows_*sizeof(double)); |
---|
2449 | memcpy(columnActivity_,columnSolution,numberColumns_*sizeof(double)); |
---|
2450 | memcpy(status_,saveStatus,(numberColumns_+numberRows_)*sizeof(char)); |
---|
2451 | |
---|
2452 | if (problemStatus_==0&&!isDualObjectiveLimitReached()) { |
---|
2453 | objectiveChange = objectiveValue_-saveObjectiveValue; |
---|
2454 | } else { |
---|
2455 | objectiveChange = 1.0e100; |
---|
2456 | } |
---|
2457 | newLower[i]=objectiveChange; |
---|
2458 | #ifdef CLP_DEBUG |
---|
2459 | printf("up on %d costs %g\n",iColumn,objectiveChange); |
---|
2460 | #endif |
---|
2461 | |
---|
2462 | /* Possibilities are: |
---|
2463 | Both sides feasible - store |
---|
2464 | Neither side feasible - set objective high and exit |
---|
2465 | One side feasible - change bounds and resolve |
---|
2466 | */ |
---|
2467 | if (newUpper[i]<1.0e100) { |
---|
2468 | if(newLower[i]<1.0e100) { |
---|
2469 | // feasible - no action |
---|
2470 | } else { |
---|
2471 | // up feasible, down infeasible |
---|
2472 | returnCode=1; |
---|
2473 | if (stopOnFirstInfeasible) |
---|
2474 | break; |
---|
2475 | } |
---|
2476 | } else { |
---|
2477 | if(newLower[i]<1.0e100) { |
---|
2478 | // down feasible, up infeasible |
---|
2479 | returnCode=1; |
---|
2480 | if (stopOnFirstInfeasible) |
---|
2481 | break; |
---|
2482 | } else { |
---|
2483 | // neither side feasible |
---|
2484 | returnCode=-1; |
---|
2485 | break; |
---|
2486 | } |
---|
2487 | } |
---|
2488 | } |
---|
2489 | delete [] rowSolution; |
---|
2490 | delete [] columnSolution; |
---|
2491 | delete [] saveStatus; |
---|
2492 | #endif |
---|
2493 | objectiveValue_ = saveObjectiveValue; |
---|
2494 | return returnCode; |
---|
2495 | } |
---|
2496 | // treat no pivot as finished (unless interesting) |
---|
2497 | int ClpSimplexDual::fastDual(bool alwaysFinish) |
---|
2498 | { |
---|
2499 | algorithm_ = -1; |
---|
2500 | dualTolerance_=dblParam_[ClpDualTolerance]; |
---|
2501 | primalTolerance_=dblParam_[ClpPrimalTolerance]; |
---|
2502 | |
---|
2503 | // save dual bound |
---|
2504 | double saveDualBound = dualBound_; |
---|
2505 | |
---|
2506 | double objectiveChange; |
---|
2507 | // for dual we will change bounds using dualBound_ |
---|
2508 | // for this we need clean basis so it is after factorize |
---|
2509 | gutsOfSolution(rowActivityWork_,columnActivityWork_); |
---|
2510 | numberFake_ =0; // Number of variables at fake bounds |
---|
2511 | changeBounds(true,NULL,objectiveChange); |
---|
2512 | |
---|
2513 | problemStatus_ = -1; |
---|
2514 | numberIterations_=0; |
---|
2515 | |
---|
2516 | int lastCleaned=0; // last time objective or bounds cleaned up |
---|
2517 | |
---|
2518 | // number of times we have declared optimality |
---|
2519 | numberTimesOptimal_=0; |
---|
2520 | |
---|
2521 | // Progress indicator |
---|
2522 | ClpSimplexProgress progress(this); |
---|
2523 | |
---|
2524 | // This says whether to restore things etc |
---|
2525 | int factorType=0; |
---|
2526 | /* |
---|
2527 | Status of problem: |
---|
2528 | 0 - optimal |
---|
2529 | 1 - infeasible |
---|
2530 | 2 - unbounded |
---|
2531 | -1 - iterating |
---|
2532 | -2 - factorization wanted |
---|
2533 | -3 - redo checking without factorization |
---|
2534 | -4 - looks infeasible |
---|
2535 | |
---|
2536 | BUT also from whileIterating return code is: |
---|
2537 | |
---|
2538 | -1 iterations etc |
---|
2539 | -2 inaccuracy |
---|
2540 | -3 slight inaccuracy (and done iterations) |
---|
2541 | +0 looks optimal (might be unbounded - but we will investigate) |
---|
2542 | +1 looks infeasible |
---|
2543 | +3 max iterations |
---|
2544 | |
---|
2545 | */ |
---|
2546 | |
---|
2547 | int returnCode = 0; |
---|
2548 | |
---|
2549 | while (problemStatus_<0) { |
---|
2550 | int iRow,iColumn; |
---|
2551 | // clear |
---|
2552 | for (iRow=0;iRow<4;iRow++) { |
---|
2553 | rowArray_[iRow]->clear(); |
---|
2554 | } |
---|
2555 | |
---|
2556 | for (iColumn=0;iColumn<2;iColumn++) { |
---|
2557 | columnArray_[iColumn]->clear(); |
---|
2558 | } |
---|
2559 | |
---|
2560 | // give matrix (and model costs and bounds a chance to be |
---|
2561 | // refreshed (normally null) |
---|
2562 | matrix_->refresh(this); |
---|
2563 | // may factorize, checks if problem finished |
---|
2564 | // should be able to speed this up on first time |
---|
2565 | statusOfProblemInDual(lastCleaned,factorType,progress); |
---|
2566 | |
---|
2567 | // Say good factorization |
---|
2568 | factorType=1; |
---|
2569 | |
---|
2570 | // Do iterations |
---|
2571 | if (problemStatus_<0) { |
---|
2572 | returnCode = whileIterating(); |
---|
2573 | if (!alwaysFinish&&returnCode<1) { |
---|
2574 | double limit = 0.0; |
---|
2575 | getDblParam(ClpDualObjectiveLimit, limit); |
---|
2576 | if(objectiveValue()*optimizationDirection_<limit|| |
---|
2577 | numberAtFakeBound()) { |
---|
2578 | // can't say anything interesting - might as well return |
---|
2579 | #ifdef CLP_DEBUG |
---|
2580 | printf("returning from fastDual after %d iterations with code %d\n", |
---|
2581 | numberIterations_,returnCode); |
---|
2582 | #endif |
---|
2583 | returnCode=1; |
---|
2584 | break; |
---|
2585 | } |
---|
2586 | } |
---|
2587 | returnCode=0; |
---|
2588 | } |
---|
2589 | } |
---|
2590 | |
---|
2591 | assert(!numberFake_||returnCode||problemStatus_); // all bounds should be okay |
---|
2592 | dualBound_ = saveDualBound; |
---|
2593 | return returnCode; |
---|
2594 | } |
---|
2595 | /* Checks number of variables at fake bounds. This is used by fastDual |
---|
2596 | so can exit gracefully before end */ |
---|
2597 | int |
---|
2598 | ClpSimplexDual::numberAtFakeBound() |
---|
2599 | { |
---|
2600 | int iSequence; |
---|
2601 | int numberFake=0; |
---|
2602 | |
---|
2603 | for (iSequence=0;iSequence<numberRows_+numberColumns_;iSequence++) { |
---|
2604 | FakeBound bound = getFakeBound(iSequence); |
---|
2605 | switch(getStatus(iSequence)) { |
---|
2606 | |
---|
2607 | case ClpSimplex::basic: |
---|
2608 | break; |
---|
2609 | case ClpSimplex::isFree: |
---|
2610 | case ClpSimplex::superBasic: |
---|
2611 | break; |
---|
2612 | case ClpSimplex::atUpperBound: |
---|
2613 | if (bound==ClpSimplex::upperFake||bound==ClpSimplex::bothFake) |
---|
2614 | numberFake++; |
---|
2615 | break; |
---|
2616 | case ClpSimplex::atLowerBound: |
---|
2617 | if (bound==ClpSimplex::lowerFake||bound==ClpSimplex::bothFake) |
---|
2618 | numberFake++; |
---|
2619 | break; |
---|
2620 | } |
---|
2621 | } |
---|
2622 | numberFake_ = numberFake; |
---|
2623 | return numberFake; |
---|
2624 | } |
---|