1 | // Copyright (C) 2002, International Business Machines |
---|
2 | // Corporation and others. All Rights Reserved. |
---|
3 | |
---|
4 | |
---|
5 | /* Notes on implementation of primal simplex algorithm. |
---|
6 | |
---|
7 | When primal feasible(A): |
---|
8 | |
---|
9 | If dual feasible, we are optimal. Otherwise choose an infeasible |
---|
10 | basic variable to enter basis from a bound (B). We now need to find an |
---|
11 | outgoing variable which will leave problem primal feasible so we get |
---|
12 | the column of the tableau corresponding to the incoming variable |
---|
13 | (with the correct sign depending if variable will go up or down). |
---|
14 | |
---|
15 | We now perform a ratio test to determine which outgoing variable will |
---|
16 | preserve primal feasibility (C). If no variable found then problem |
---|
17 | is unbounded (in primal sense). If there is a variable, we then |
---|
18 | perform pivot and repeat. Trivial? |
---|
19 | |
---|
20 | ------------------------------------------- |
---|
21 | |
---|
22 | A) How do we get primal feasible? All variables have fake costs |
---|
23 | outside their feasible region so it is trivial to declare problem |
---|
24 | feasible. OSL did not have a phase 1/phase 2 approach but |
---|
25 | instead effectively put an extra cost on infeasible basic variables |
---|
26 | I am taking the same approach here, although it is generalized |
---|
27 | to allow for non-linear costs and dual information. |
---|
28 | |
---|
29 | In OSL, this weight was changed heuristically, here at present |
---|
30 | it is only increased if problem looks finished. If problem is |
---|
31 | feasible I check for unboundedness. If not unbounded we |
---|
32 | could play with going into dual. As long as weights increase |
---|
33 | any algorithm would be finite. |
---|
34 | |
---|
35 | B) Which incoming variable to choose is a virtual base class. |
---|
36 | For difficult problems steepest edge is preferred while for |
---|
37 | very easy (large) problems we will need partial scan. |
---|
38 | |
---|
39 | C) Sounds easy, but this is hardest part of algorithm. |
---|
40 | 1) Instead of stopping at first choice, we may be able |
---|
41 | to allow that variable to go through bound and if objective |
---|
42 | still improving choose again. These mini iterations can |
---|
43 | increase speed by orders of magnitude but we may need to |
---|
44 | go to more of a bucket choice of variable rather than looking |
---|
45 | at them one by one (for speed). |
---|
46 | 2) Accuracy. Basic infeasibilities may be less than |
---|
47 | tolerance. Pivoting on these makes objective go backwards. |
---|
48 | OSL modified cost so a zero move was made, Gill et al |
---|
49 | modified so a strictly positive move was made. |
---|
50 | The two problems are that re-factorizations can |
---|
51 | change rinfeasibilities above and below tolerances and that when |
---|
52 | finished we need to reset costs and try again. |
---|
53 | 3) Degeneracy. Gill et al helps but may not be enough. We |
---|
54 | may need more. Also it can improve speed a lot if we perturb |
---|
55 | the rhs and bounds significantly. |
---|
56 | |
---|
57 | References: |
---|
58 | Forrest and Goldfarb, Steepest-edge simplex algorithms for |
---|
59 | linear programming - Mathematical Programming 1992 |
---|
60 | Forrest and Tomlin, Implementing the simplex method for |
---|
61 | the Optimization Subroutine Library - IBM Systems Journal 1992 |
---|
62 | Gill, Murray, Saunders, Wright A Practical Anti-Cycling |
---|
63 | Procedure for Linear and Nonlinear Programming SOL report 1988 |
---|
64 | |
---|
65 | |
---|
66 | TODO: |
---|
67 | |
---|
68 | a) Better recovery procedures. At present I never check on forward |
---|
69 | progress. There is checkpoint/restart with reducing |
---|
70 | re-factorization frequency, but this is only on singular |
---|
71 | factorizations. |
---|
72 | b) Fast methods for large easy problems (and also the option for |
---|
73 | the code to automatically choose which method). |
---|
74 | c) We need to be able to stop in various ways for OSI - this |
---|
75 | is fairly easy. |
---|
76 | |
---|
77 | */ |
---|
78 | |
---|
79 | |
---|
80 | #include "CoinPragma.hpp" |
---|
81 | |
---|
82 | #include <math.h> |
---|
83 | |
---|
84 | #include "CoinHelperFunctions.hpp" |
---|
85 | #include "ClpSimplexPrimal.hpp" |
---|
86 | #include "ClpFactorization.hpp" |
---|
87 | #include "ClpNonLinearCost.hpp" |
---|
88 | #include "CoinPackedMatrix.hpp" |
---|
89 | #include "CoinIndexedVector.hpp" |
---|
90 | #include "CoinWarmStartBasis.hpp" |
---|
91 | #include "ClpPrimalColumnPivot.hpp" |
---|
92 | #include "ClpMessage.hpp" |
---|
93 | #include <cfloat> |
---|
94 | #include <cassert> |
---|
95 | #include <string> |
---|
96 | #include <stdio.h> |
---|
97 | #include <iostream> |
---|
98 | // primal |
---|
99 | int ClpSimplexPrimal::primal (int ifValuesPass ) |
---|
100 | { |
---|
101 | |
---|
102 | /* |
---|
103 | Method |
---|
104 | |
---|
105 | It tries to be a single phase approach with a weight of 1.0 being |
---|
106 | given to getting optimal and a weight of infeasibilityCost_ being |
---|
107 | given to getting primal feasible. In this version I have tried to |
---|
108 | be clever in a stupid way. The idea of fake bounds in dual |
---|
109 | seems to work so the primal analogue would be that of getting |
---|
110 | bounds on reduced costs (by a presolve approach) and using |
---|
111 | these for being above or below feasible region. I decided to waste |
---|
112 | memory and keep these explicitly. This allows for non-linear |
---|
113 | costs! |
---|
114 | |
---|
115 | The code is designed to take advantage of sparsity so arrays are |
---|
116 | seldom zeroed out from scratch or gone over in their entirety. |
---|
117 | The only exception is a full scan to find incoming variable for |
---|
118 | Dantzig row choice. For steepest edge we keep an updated list |
---|
119 | of dual infeasibilities (actually squares). |
---|
120 | On easy problems we don't need full scan - just |
---|
121 | pick first reasonable. |
---|
122 | |
---|
123 | One problem is how to tackle degeneracy and accuracy. At present |
---|
124 | I am using the modification of costs which I put in OSL and which was |
---|
125 | extended by Gill et al. I am still not sure of the exact details. |
---|
126 | |
---|
127 | The flow of primal is three while loops as follows: |
---|
128 | |
---|
129 | while (not finished) { |
---|
130 | |
---|
131 | while (not clean solution) { |
---|
132 | |
---|
133 | Factorize and/or clean up solution by changing bounds so |
---|
134 | primal feasible. If looks finished check fake primal bounds. |
---|
135 | Repeat until status is iterating (-1) or finished (0,1,2) |
---|
136 | |
---|
137 | } |
---|
138 | |
---|
139 | while (status==-1) { |
---|
140 | |
---|
141 | Iterate until no pivot in or out or time to re-factorize. |
---|
142 | |
---|
143 | Flow is: |
---|
144 | |
---|
145 | choose pivot column (incoming variable). if none then |
---|
146 | we are primal feasible so looks as if done but we need to |
---|
147 | break and check bounds etc. |
---|
148 | |
---|
149 | Get pivot column in tableau |
---|
150 | |
---|
151 | Choose outgoing row. If we don't find one then we look |
---|
152 | primal unbounded so break and check bounds etc. (Also the |
---|
153 | pivot tolerance is larger after any iterations so that may be |
---|
154 | reason) |
---|
155 | |
---|
156 | If we do find outgoing row, we may have to adjust costs to |
---|
157 | keep going forwards (anti-degeneracy). Check pivot will be stable |
---|
158 | and if unstable throw away iteration and break to re-factorize. |
---|
159 | If minor error re-factorize after iteration. |
---|
160 | |
---|
161 | Update everything (this may involve changing bounds on |
---|
162 | variables to stay primal feasible. |
---|
163 | |
---|
164 | } |
---|
165 | |
---|
166 | } |
---|
167 | |
---|
168 | At present we never check we are going forwards. I overdid that in |
---|
169 | OSL so will try and make a last resort. |
---|
170 | |
---|
171 | Needs partial scan pivot in option. |
---|
172 | |
---|
173 | May need other anti-degeneracy measures, especially if we try and use |
---|
174 | loose tolerances as a way to solve in fewer iterations. |
---|
175 | |
---|
176 | I like idea of dynamic scaling. This gives opportunity to decouple |
---|
177 | different implications of scaling for accuracy, iteration count and |
---|
178 | feasibility tolerance. |
---|
179 | |
---|
180 | */ |
---|
181 | |
---|
182 | algorithm_ = +1; |
---|
183 | //specialOptions_ |= 4; |
---|
184 | |
---|
185 | // save data |
---|
186 | ClpDataSave data = saveData(); |
---|
187 | |
---|
188 | // Save so can see if doing after dual |
---|
189 | int initialStatus=problemStatus_; |
---|
190 | // initialize - maybe values pass and algorithm_ is +1 |
---|
191 | if (!startup(ifValuesPass)) { |
---|
192 | |
---|
193 | // Set average theta |
---|
194 | nonLinearCost_->setAverageTheta(1.0e3); |
---|
195 | int lastCleaned=0; // last time objective or bounds cleaned up |
---|
196 | |
---|
197 | // Say no pivot has occurred (for steepest edge and updates) |
---|
198 | pivotRow_=-2; |
---|
199 | |
---|
200 | // This says whether to restore things etc |
---|
201 | int factorType=0; |
---|
202 | if (problemStatus_<0&&perturbation_<100) { |
---|
203 | perturb(0); |
---|
204 | // Can't get here if values pass |
---|
205 | assert (!ifValuesPass); |
---|
206 | gutsOfSolution(NULL,NULL); |
---|
207 | if (handler_->logLevel()>2) { |
---|
208 | handler_->message(CLP_SIMPLEX_STATUS,messages_) |
---|
209 | <<numberIterations_<<objectiveValue(); |
---|
210 | handler_->printing(sumPrimalInfeasibilities_>0.0) |
---|
211 | <<sumPrimalInfeasibilities_<<numberPrimalInfeasibilities_; |
---|
212 | handler_->printing(sumDualInfeasibilities_>0.0) |
---|
213 | <<sumDualInfeasibilities_<<numberDualInfeasibilities_; |
---|
214 | handler_->printing(numberDualInfeasibilitiesWithoutFree_ |
---|
215 | <numberDualInfeasibilities_) |
---|
216 | <<numberDualInfeasibilitiesWithoutFree_; |
---|
217 | handler_->message()<<CoinMessageEol; |
---|
218 | } |
---|
219 | } |
---|
220 | ClpSimplex * saveModel=NULL; |
---|
221 | int stopSprint=-1; |
---|
222 | int sprintPass=0; |
---|
223 | int reasonableSprintIteration=0; |
---|
224 | int lastSprintIteration=0; |
---|
225 | double lastObjectiveValue=COIN_DBL_MAX; |
---|
226 | // Start check for cycles |
---|
227 | progress_->startCheck(); |
---|
228 | /* |
---|
229 | Status of problem: |
---|
230 | 0 - optimal |
---|
231 | 1 - infeasible |
---|
232 | 2 - unbounded |
---|
233 | -1 - iterating |
---|
234 | -2 - factorization wanted |
---|
235 | -3 - redo checking without factorization |
---|
236 | -4 - looks infeasible |
---|
237 | -5 - looks unbounded |
---|
238 | */ |
---|
239 | while (problemStatus_<0) { |
---|
240 | int iRow,iColumn; |
---|
241 | // clear |
---|
242 | for (iRow=0;iRow<4;iRow++) { |
---|
243 | rowArray_[iRow]->clear(); |
---|
244 | } |
---|
245 | |
---|
246 | for (iColumn=0;iColumn<2;iColumn++) { |
---|
247 | columnArray_[iColumn]->clear(); |
---|
248 | } |
---|
249 | |
---|
250 | // give matrix (and model costs and bounds a chance to be |
---|
251 | // refreshed (normally null) |
---|
252 | matrix_->refresh(this); |
---|
253 | // If getting nowhere - why not give it a kick |
---|
254 | #if 1 |
---|
255 | if (perturbation_<101&&numberIterations_>2*(numberRows_+numberColumns_) |
---|
256 | &&initialStatus!=10) |
---|
257 | perturb(1); |
---|
258 | #endif |
---|
259 | // If we have done no iterations - special |
---|
260 | if (lastGoodIteration_==numberIterations_&&factorType) |
---|
261 | factorType=3; |
---|
262 | if (saveModel) { |
---|
263 | // Doing sprint |
---|
264 | if (sequenceIn_<0||numberIterations_>=stopSprint) { |
---|
265 | problemStatus_=-1; |
---|
266 | originalModel(saveModel); |
---|
267 | saveModel=NULL; |
---|
268 | if (sequenceIn_<0&&numberIterations_<reasonableSprintIteration&& |
---|
269 | sprintPass>100) |
---|
270 | primalColumnPivot_->switchOffSprint(); |
---|
271 | //lastSprintIteration=numberIterations_; |
---|
272 | printf("End small model\n"); |
---|
273 | } |
---|
274 | } |
---|
275 | |
---|
276 | // may factorize, checks if problem finished |
---|
277 | statusOfProblemInPrimal(lastCleaned,factorType,progress_,true,saveModel); |
---|
278 | // See if sprint says redo beacuse of problems |
---|
279 | if (numberDualInfeasibilities_==-776) { |
---|
280 | // Need new set of variables |
---|
281 | problemStatus_=-1; |
---|
282 | originalModel(saveModel); |
---|
283 | saveModel=NULL; |
---|
284 | //lastSprintIteration=numberIterations_; |
---|
285 | printf("End small model after\n"); |
---|
286 | statusOfProblemInPrimal(lastCleaned,factorType,progress_,true,saveModel); |
---|
287 | } |
---|
288 | int numberSprintIterations=0; |
---|
289 | int numberSprintColumns = primalColumnPivot_->numberSprintColumns(numberSprintIterations); |
---|
290 | if (problemStatus_==777) { |
---|
291 | // problems so do one pass with normal |
---|
292 | problemStatus_=-1; |
---|
293 | originalModel(saveModel); |
---|
294 | saveModel=NULL; |
---|
295 | // Skip factorization |
---|
296 | //statusOfProblemInPrimal(lastCleaned,factorType,progress_,false,saveModel); |
---|
297 | statusOfProblemInPrimal(lastCleaned,factorType,progress_,true,saveModel); |
---|
298 | } else if (problemStatus_<0&&!saveModel&&numberSprintColumns&&firstFree_<0) { |
---|
299 | int numberSort=0; |
---|
300 | int numberFixed=0; |
---|
301 | int numberBasic=0; |
---|
302 | reasonableSprintIteration = numberIterations_ + 100; |
---|
303 | int * whichColumns = new int[numberColumns_]; |
---|
304 | double * weight = new double[numberColumns_]; |
---|
305 | int numberNegative=0; |
---|
306 | double sumNegative = 0.0; |
---|
307 | // now massage weight so all basic in plus good djs |
---|
308 | for (iColumn=0;iColumn<numberColumns_;iColumn++) { |
---|
309 | double dj = dj_[iColumn]; |
---|
310 | switch(getColumnStatus(iColumn)) { |
---|
311 | |
---|
312 | case basic: |
---|
313 | dj = -1.0e50; |
---|
314 | numberBasic++; |
---|
315 | break; |
---|
316 | case atUpperBound: |
---|
317 | dj = -dj; |
---|
318 | break; |
---|
319 | case isFixed: |
---|
320 | dj=1.0e50; |
---|
321 | numberFixed++; |
---|
322 | break; |
---|
323 | case atLowerBound: |
---|
324 | dj = dj; |
---|
325 | break; |
---|
326 | case isFree: |
---|
327 | dj = -100.0*fabs(dj); |
---|
328 | break; |
---|
329 | case superBasic: |
---|
330 | dj = -100.0*fabs(dj); |
---|
331 | break; |
---|
332 | } |
---|
333 | if (dj<-dualTolerance_&&dj>-1.0e50) { |
---|
334 | numberNegative++; |
---|
335 | sumNegative -= dj; |
---|
336 | } |
---|
337 | weight[iColumn]=dj; |
---|
338 | whichColumns[iColumn] = iColumn; |
---|
339 | } |
---|
340 | handler_->message(CLP_SPRINT,messages_) |
---|
341 | <<sprintPass<<numberIterations_-lastSprintIteration<<objectiveValue()<<sumNegative |
---|
342 | <<numberNegative |
---|
343 | <<CoinMessageEol; |
---|
344 | sprintPass++; |
---|
345 | lastSprintIteration=numberIterations_; |
---|
346 | if (objectiveValue()*optimizationDirection_>lastObjectiveValue-1.0e-7&&sprintPass>5) { |
---|
347 | // switch off |
---|
348 | printf("Switching off sprint\n"); |
---|
349 | primalColumnPivot_->switchOffSprint(); |
---|
350 | } else { |
---|
351 | lastObjectiveValue = objectiveValue()*optimizationDirection_; |
---|
352 | // sort |
---|
353 | CoinSort_2(weight,weight+numberColumns_,whichColumns); |
---|
354 | numberSort = min(numberColumns_-numberFixed,numberBasic+numberSprintColumns); |
---|
355 | // Sort to make consistent ? |
---|
356 | std::sort(whichColumns,whichColumns+numberSort); |
---|
357 | saveModel = new ClpSimplex(this,numberSort,whichColumns); |
---|
358 | delete [] whichColumns; |
---|
359 | delete [] weight; |
---|
360 | // Skip factorization |
---|
361 | //statusOfProblemInPrimal(lastCleaned,factorType,progress_,false,saveModel); |
---|
362 | //statusOfProblemInPrimal(lastCleaned,factorType,progress_,true,saveModel); |
---|
363 | stopSprint = numberIterations_+numberSprintIterations; |
---|
364 | printf("Sprint with %d columns for %d iterations\n", |
---|
365 | numberSprintColumns,numberSprintIterations); |
---|
366 | } |
---|
367 | } |
---|
368 | |
---|
369 | // Say good factorization |
---|
370 | factorType=1; |
---|
371 | |
---|
372 | // Say no pivot has occurred (for steepest edge and updates) |
---|
373 | pivotRow_=-2; |
---|
374 | |
---|
375 | // exit if victory declared |
---|
376 | if (problemStatus_>=0) |
---|
377 | break; |
---|
378 | |
---|
379 | // Iterate |
---|
380 | whileIterating(ifValuesPass); |
---|
381 | } |
---|
382 | } |
---|
383 | // if infeasible get real values |
---|
384 | if (problemStatus_==1) { |
---|
385 | infeasibilityCost_=0.0; |
---|
386 | createRim(7); |
---|
387 | nonLinearCost_->checkInfeasibilities(0.0); |
---|
388 | sumPrimalInfeasibilities_=nonLinearCost_->sumInfeasibilities(); |
---|
389 | numberPrimalInfeasibilities_= nonLinearCost_->numberInfeasibilities(); |
---|
390 | // and get good feasible duals |
---|
391 | computeDuals(NULL); |
---|
392 | } |
---|
393 | // clean up |
---|
394 | unflag(); |
---|
395 | finish(); |
---|
396 | restoreData(data); |
---|
397 | return problemStatus_; |
---|
398 | } |
---|
399 | /* |
---|
400 | Reasons to come out: |
---|
401 | -1 iterations etc |
---|
402 | -2 inaccuracy |
---|
403 | -3 slight inaccuracy (and done iterations) |
---|
404 | -4 end of values pass and done iterations |
---|
405 | +0 looks optimal (might be infeasible - but we will investigate) |
---|
406 | +2 looks unbounded |
---|
407 | +3 max iterations |
---|
408 | */ |
---|
409 | int |
---|
410 | ClpSimplexPrimal::whileIterating(int valuesOption) |
---|
411 | { |
---|
412 | // Say if values pass |
---|
413 | int ifValuesPass=(firstFree_>=0) ? 1 : 0; |
---|
414 | int returnCode=-1; |
---|
415 | int superBasicType=1; |
---|
416 | if (valuesOption>1) |
---|
417 | superBasicType=3; |
---|
418 | // status stays at -1 while iterating, >=0 finished, -2 to invert |
---|
419 | // status -3 to go to top without an invert |
---|
420 | while (problemStatus_==-1) { |
---|
421 | //#define CLP_DEBUG 1 |
---|
422 | #ifdef CLP_DEBUG |
---|
423 | { |
---|
424 | int i; |
---|
425 | // not [1] as has information |
---|
426 | for (i=0;i<4;i++) { |
---|
427 | if (i!=1) |
---|
428 | rowArray_[i]->checkClear(); |
---|
429 | } |
---|
430 | for (i=0;i<2;i++) { |
---|
431 | columnArray_[i]->checkClear(); |
---|
432 | } |
---|
433 | } |
---|
434 | #endif |
---|
435 | #if CLP_DEBUG>2 |
---|
436 | // very expensive |
---|
437 | if (numberIterations_>0&&numberIterations_<-2534) { |
---|
438 | handler_->setLogLevel(63); |
---|
439 | double saveValue = objectiveValue_; |
---|
440 | double * saveRow1 = new double[numberRows_]; |
---|
441 | double * saveRow2 = new double[numberRows_]; |
---|
442 | memcpy(saveRow1,rowReducedCost_,numberRows_*sizeof(double)); |
---|
443 | memcpy(saveRow2,rowActivityWork_,numberRows_*sizeof(double)); |
---|
444 | double * saveColumn1 = new double[numberColumns_]; |
---|
445 | double * saveColumn2 = new double[numberColumns_]; |
---|
446 | memcpy(saveColumn1,reducedCostWork_,numberColumns_*sizeof(double)); |
---|
447 | memcpy(saveColumn2,columnActivityWork_,numberColumns_*sizeof(double)); |
---|
448 | createRim(7); |
---|
449 | gutsOfSolution(NULL,NULL); |
---|
450 | printf("xxx %d old obj %g, recomputed %g, sum primal inf %g\n", |
---|
451 | numberIterations_, |
---|
452 | saveValue,objectiveValue_,sumPrimalInfeasibilities_); |
---|
453 | memcpy(rowReducedCost_,saveRow1,numberRows_*sizeof(double)); |
---|
454 | memcpy(rowActivityWork_,saveRow2,numberRows_*sizeof(double)); |
---|
455 | memcpy(reducedCostWork_,saveColumn1,numberColumns_*sizeof(double)); |
---|
456 | memcpy(columnActivityWork_,saveColumn2,numberColumns_*sizeof(double)); |
---|
457 | delete [] saveRow1; |
---|
458 | delete [] saveRow2; |
---|
459 | delete [] saveColumn1; |
---|
460 | delete [] saveColumn2; |
---|
461 | objectiveValue_=saveValue; |
---|
462 | } |
---|
463 | #endif |
---|
464 | if (!ifValuesPass) { |
---|
465 | // choose column to come in |
---|
466 | // can use pivotRow_ to update weights |
---|
467 | // pass in list of cost changes so can do row updates (rowArray_[1]) |
---|
468 | // NOTE rowArray_[0] is used by computeDuals which is a |
---|
469 | // slow way of getting duals but might be used |
---|
470 | primalColumn(rowArray_[1],rowArray_[2],rowArray_[3], |
---|
471 | columnArray_[0],columnArray_[1]); |
---|
472 | } else { |
---|
473 | // in values pass |
---|
474 | int sequenceIn=nextSuperBasic(superBasicType,columnArray_[0]); |
---|
475 | if (valuesOption>1) |
---|
476 | superBasicType=2; |
---|
477 | if (sequenceIn<0) { |
---|
478 | // end of values pass - initialize weights etc |
---|
479 | handler_->message(CLP_END_VALUES_PASS,messages_) |
---|
480 | <<numberIterations_; |
---|
481 | primalColumnPivot_->saveWeights(this,5); |
---|
482 | problemStatus_=-2; // factorize now |
---|
483 | pivotRow_=-1; // say no weights update |
---|
484 | returnCode=-4; |
---|
485 | // Clean up |
---|
486 | int i; |
---|
487 | for (i=0;i<numberRows_+numberColumns_;i++) { |
---|
488 | if (getColumnStatus(i)==atLowerBound||getColumnStatus(i)==isFixed) |
---|
489 | solution_[i]=lower_[i]; |
---|
490 | else if (getColumnStatus(i)==atUpperBound) |
---|
491 | solution_[i]=upper_[i]; |
---|
492 | } |
---|
493 | break; |
---|
494 | } else { |
---|
495 | // normal |
---|
496 | sequenceIn_ = sequenceIn; |
---|
497 | valueIn_=solution_[sequenceIn_]; |
---|
498 | lowerIn_=lower_[sequenceIn_]; |
---|
499 | upperIn_=upper_[sequenceIn_]; |
---|
500 | dualIn_=dj_[sequenceIn_]; |
---|
501 | } |
---|
502 | } |
---|
503 | pivotRow_=-1; |
---|
504 | sequenceOut_=-1; |
---|
505 | rowArray_[1]->clear(); |
---|
506 | if (sequenceIn_>=0) { |
---|
507 | // we found a pivot column |
---|
508 | assert (!flagged(sequenceIn_)); |
---|
509 | #ifdef CLP_DEBUG |
---|
510 | if ((handler_->logLevel()&32)) { |
---|
511 | char x = isColumn(sequenceIn_) ? 'C' :'R'; |
---|
512 | std::cout<<"pivot column "<< |
---|
513 | x<<sequenceWithin(sequenceIn_)<<std::endl; |
---|
514 | } |
---|
515 | #endif |
---|
516 | // do second half of iteration |
---|
517 | returnCode = pivotResult(ifValuesPass); |
---|
518 | if (returnCode<-1&&returnCode>-5) { |
---|
519 | problemStatus_=-2; // |
---|
520 | } else if (returnCode==-5) { |
---|
521 | // something flagged - continue; |
---|
522 | } else if (returnCode==2) { |
---|
523 | problemStatus_=-5; // looks unbounded |
---|
524 | } else if (returnCode==4) { |
---|
525 | problemStatus_=-2; // looks unbounded but has iterated |
---|
526 | } else if (returnCode!=-1) { |
---|
527 | assert(returnCode==3); |
---|
528 | problemStatus_=3; |
---|
529 | } |
---|
530 | } else { |
---|
531 | // no pivot column |
---|
532 | #ifdef CLP_DEBUG |
---|
533 | if (handler_->logLevel()&32) |
---|
534 | printf("** no column pivot\n"); |
---|
535 | #endif |
---|
536 | if (nonLinearCost_->numberInfeasibilities()) |
---|
537 | problemStatus_=-4; // might be infeasible |
---|
538 | returnCode=0; |
---|
539 | break; |
---|
540 | } |
---|
541 | } |
---|
542 | if (valuesOption>1) |
---|
543 | columnArray_[0]->setNumElements(0); |
---|
544 | return returnCode; |
---|
545 | } |
---|
546 | /* Checks if finished. Updates status */ |
---|
547 | void |
---|
548 | ClpSimplexPrimal::statusOfProblemInPrimal(int & lastCleaned,int type, |
---|
549 | ClpSimplexProgress * progress, |
---|
550 | bool doFactorization, |
---|
551 | ClpSimplex * originalModel) |
---|
552 | { |
---|
553 | if (type==2) { |
---|
554 | // trouble - restore solution |
---|
555 | memcpy(status_ ,saveStatus_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
556 | memcpy(rowActivityWork_,savedSolution_+numberColumns_ , |
---|
557 | numberRows_*sizeof(double)); |
---|
558 | memcpy(columnActivityWork_,savedSolution_ , |
---|
559 | numberColumns_*sizeof(double)); |
---|
560 | forceFactorization_=1; // a bit drastic but .. |
---|
561 | pivotRow_=-1; // say no weights update |
---|
562 | changeMade_++; // say change made |
---|
563 | } |
---|
564 | int saveThreshold = factorization_->sparseThreshold(); |
---|
565 | int tentativeStatus = problemStatus_; |
---|
566 | if (problemStatus_>-3||problemStatus_==-4) { |
---|
567 | // factorize |
---|
568 | // later on we will need to recover from singularities |
---|
569 | // also we could skip if first time |
---|
570 | // do weights |
---|
571 | // This may save pivotRow_ for use |
---|
572 | if (doFactorization) |
---|
573 | primalColumnPivot_->saveWeights(this,1); |
---|
574 | |
---|
575 | if (type&&doFactorization) { |
---|
576 | // is factorization okay? |
---|
577 | if (internalFactorize(1)) { |
---|
578 | if (solveType_==2) { |
---|
579 | // say odd |
---|
580 | problemStatus_=5; |
---|
581 | return; |
---|
582 | } |
---|
583 | #if 1 |
---|
584 | // switch off dense |
---|
585 | int saveDense = factorization_->denseThreshold(); |
---|
586 | factorization_->setDenseThreshold(0); |
---|
587 | internalFactorize(2); |
---|
588 | factorization_->setDenseThreshold(saveDense); |
---|
589 | #else |
---|
590 | |
---|
591 | // no - restore previous basis |
---|
592 | assert (type==1); |
---|
593 | memcpy(status_ ,saveStatus_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
594 | memcpy(rowActivityWork_,savedSolution_+numberColumns_ , |
---|
595 | numberRows_*sizeof(double)); |
---|
596 | memcpy(columnActivityWork_,savedSolution_ , |
---|
597 | numberColumns_*sizeof(double)); |
---|
598 | forceFactorization_=1; // a bit drastic but .. |
---|
599 | type = 2; |
---|
600 | assert (internalFactorize(1)==0); |
---|
601 | #endif |
---|
602 | changeMade_++; // say change made |
---|
603 | } |
---|
604 | } |
---|
605 | if (problemStatus_!=-4) |
---|
606 | problemStatus_=-3; |
---|
607 | } |
---|
608 | // at this stage status is -3 or -5 if looks unbounded |
---|
609 | // get primal and dual solutions |
---|
610 | // put back original costs and then check |
---|
611 | createRim(4); |
---|
612 | gutsOfSolution(NULL,NULL,(firstFree_>=0)); |
---|
613 | // Double check reduced costs if no action |
---|
614 | if (progress->lastIterationNumber(0)==numberIterations_) { |
---|
615 | if (primalColumnPivot_->looksOptimal()) { |
---|
616 | numberDualInfeasibilities_ = 0; |
---|
617 | sumDualInfeasibilities_ = 0.0; |
---|
618 | } |
---|
619 | } |
---|
620 | // Check if looping |
---|
621 | int loop; |
---|
622 | if (type!=2) |
---|
623 | loop = progress->looping(); |
---|
624 | else |
---|
625 | loop=-1; |
---|
626 | if (loop>=0) { |
---|
627 | if (!problemStatus_) { |
---|
628 | // declaring victory |
---|
629 | numberPrimalInfeasibilities_ = 0; |
---|
630 | sumPrimalInfeasibilities_ = 0.0; |
---|
631 | } else { |
---|
632 | problemStatus_ = loop; //exit if in loop |
---|
633 | problemStatus_ = 10; // instead - try other algorithm |
---|
634 | } |
---|
635 | problemStatus_ = 10; // instead - try other algorithm |
---|
636 | return ; |
---|
637 | } else if (loop<-1) { |
---|
638 | // Is it time for drastic measures |
---|
639 | if (nonLinearCost_->numberInfeasibilities()&&progress->badTimes()>5&& |
---|
640 | progress->oddState()<10&&progress->oddState()>=0) { |
---|
641 | progress->newOddState(); |
---|
642 | nonLinearCost_->zapCosts(); |
---|
643 | } |
---|
644 | // something may have changed |
---|
645 | gutsOfSolution(NULL,NULL); |
---|
646 | } |
---|
647 | // If progress then reset costs |
---|
648 | if (loop==-1&&!nonLinearCost_->numberInfeasibilities()&&progress->oddState()<0) { |
---|
649 | createRim(4,false); // costs back |
---|
650 | delete nonLinearCost_; |
---|
651 | nonLinearCost_ = new ClpNonLinearCost(this); |
---|
652 | progress->endOddState(); |
---|
653 | gutsOfSolution(NULL,NULL); |
---|
654 | } |
---|
655 | // Flag to say whether to go to dual to clean up |
---|
656 | bool goToDual=false; |
---|
657 | // really for free variables in |
---|
658 | //if((progressFlag_&2)!=0) |
---|
659 | //problemStatus_=-1;; |
---|
660 | progressFlag_ = 0; //reset progress flag |
---|
661 | |
---|
662 | handler_->message(CLP_SIMPLEX_STATUS,messages_) |
---|
663 | <<numberIterations_<<nonLinearCost_->feasibleReportCost(); |
---|
664 | handler_->printing(nonLinearCost_->numberInfeasibilities()>0) |
---|
665 | <<nonLinearCost_->sumInfeasibilities()<<nonLinearCost_->numberInfeasibilities(); |
---|
666 | handler_->printing(sumDualInfeasibilities_>0.0) |
---|
667 | <<sumDualInfeasibilities_<<numberDualInfeasibilities_; |
---|
668 | handler_->printing(numberDualInfeasibilitiesWithoutFree_ |
---|
669 | <numberDualInfeasibilities_) |
---|
670 | <<numberDualInfeasibilitiesWithoutFree_; |
---|
671 | handler_->message()<<CoinMessageEol; |
---|
672 | if (!primalFeasible()) { |
---|
673 | nonLinearCost_->checkInfeasibilities(primalTolerance_); |
---|
674 | gutsOfSolution(NULL,NULL); |
---|
675 | } |
---|
676 | // we may wish to say it is optimal even if infeasible |
---|
677 | bool alwaysOptimal = (specialOptions_&1)!=0; |
---|
678 | // give code benefit of doubt |
---|
679 | if (sumOfRelaxedDualInfeasibilities_ == 0.0&& |
---|
680 | sumOfRelaxedPrimalInfeasibilities_ == 0.0) { |
---|
681 | // say optimal (with these bounds etc) |
---|
682 | numberDualInfeasibilities_ = 0; |
---|
683 | sumDualInfeasibilities_ = 0.0; |
---|
684 | numberPrimalInfeasibilities_ = 0; |
---|
685 | sumPrimalInfeasibilities_ = 0.0; |
---|
686 | // But check if in sprint |
---|
687 | if (originalModel) { |
---|
688 | // Carry on and re-do |
---|
689 | numberDualInfeasibilities_ = -776; |
---|
690 | } |
---|
691 | } |
---|
692 | // had ||(type==3&&problemStatus_!=-5) -- ??? why ???? |
---|
693 | if (dualFeasible()||problemStatus_==-4) { |
---|
694 | if (nonLinearCost_->numberInfeasibilities()&&!alwaysOptimal) { |
---|
695 | //may need infeasiblity cost changed |
---|
696 | // we can see if we can construct a ray |
---|
697 | // make up a new objective |
---|
698 | double saveWeight = infeasibilityCost_; |
---|
699 | // save nonlinear cost as we are going to switch off costs |
---|
700 | ClpNonLinearCost * nonLinear = nonLinearCost_; |
---|
701 | // do twice to make sure Primal solution has settled |
---|
702 | // put non-basics to bounds in case tolerance moved |
---|
703 | // put back original costs |
---|
704 | createRim(4); |
---|
705 | nonLinearCost_->checkInfeasibilities(0.0); |
---|
706 | gutsOfSolution(NULL,NULL); |
---|
707 | |
---|
708 | infeasibilityCost_=1.0e100; |
---|
709 | // put back original costs |
---|
710 | createRim(4); |
---|
711 | nonLinearCost_->checkInfeasibilities(primalTolerance_); |
---|
712 | // may have fixed infeasibilities - double check |
---|
713 | if (nonLinearCost_->numberInfeasibilities()==0) { |
---|
714 | // carry on |
---|
715 | problemStatus_ = -1; |
---|
716 | infeasibilityCost_=saveWeight; |
---|
717 | nonLinearCost_->checkInfeasibilities(primalTolerance_); |
---|
718 | } else { |
---|
719 | nonLinearCost_=NULL; |
---|
720 | // scale |
---|
721 | int i; |
---|
722 | for (i=0;i<numberRows_+numberColumns_;i++) |
---|
723 | cost_[i] *= 1.0e-95; |
---|
724 | gutsOfSolution(NULL,NULL); |
---|
725 | nonLinearCost_=nonLinear; |
---|
726 | infeasibilityCost_=saveWeight; |
---|
727 | if ((infeasibilityCost_>=1.0e18|| |
---|
728 | numberDualInfeasibilities_==0)&&perturbation_==101) { |
---|
729 | goToDual=unPerturb(); // stop any further perturbation |
---|
730 | if (nonLinearCost_->sumInfeasibilities()>1.0e-1) |
---|
731 | goToDual=false; |
---|
732 | nonLinearCost_->checkInfeasibilities(primalTolerance_); |
---|
733 | numberDualInfeasibilities_=1; // carry on |
---|
734 | problemStatus_=-1; |
---|
735 | } |
---|
736 | if (infeasibilityCost_>=1.0e20|| |
---|
737 | numberDualInfeasibilities_==0) { |
---|
738 | // we are infeasible - use as ray |
---|
739 | delete [] ray_; |
---|
740 | ray_ = new double [numberRows_]; |
---|
741 | memcpy(ray_,dual_,numberRows_*sizeof(double)); |
---|
742 | // and get feasible duals |
---|
743 | infeasibilityCost_=0.0; |
---|
744 | createRim(4); |
---|
745 | nonLinearCost_->checkInfeasibilities(primalTolerance_); |
---|
746 | gutsOfSolution(NULL,NULL); |
---|
747 | // so will exit |
---|
748 | infeasibilityCost_=1.0e30; |
---|
749 | // reset infeasibilities |
---|
750 | sumPrimalInfeasibilities_=nonLinearCost_->sumInfeasibilities();; |
---|
751 | numberPrimalInfeasibilities_= |
---|
752 | nonLinearCost_->numberInfeasibilities(); |
---|
753 | } |
---|
754 | if (infeasibilityCost_<1.0e20) { |
---|
755 | infeasibilityCost_ *= 5.0; |
---|
756 | changeMade_++; // say change made |
---|
757 | handler_->message(CLP_PRIMAL_WEIGHT,messages_) |
---|
758 | <<infeasibilityCost_ |
---|
759 | <<CoinMessageEol; |
---|
760 | // put back original costs and then check |
---|
761 | createRim(4); |
---|
762 | nonLinearCost_->checkInfeasibilities(0.0); |
---|
763 | gutsOfSolution(NULL,NULL); |
---|
764 | problemStatus_=-1; //continue |
---|
765 | goToDual=false; |
---|
766 | } else { |
---|
767 | // say infeasible |
---|
768 | problemStatus_ = 1; |
---|
769 | } |
---|
770 | } |
---|
771 | } else { |
---|
772 | // may be optimal |
---|
773 | if (perturbation_==101) { |
---|
774 | goToDual=unPerturb(); // stop any further perturbation |
---|
775 | lastCleaned=-1; // carry on |
---|
776 | } |
---|
777 | bool unflagged = unflag(); |
---|
778 | if ( lastCleaned!=numberIterations_||unflagged) { |
---|
779 | handler_->message(CLP_PRIMAL_OPTIMAL,messages_) |
---|
780 | <<primalTolerance_ |
---|
781 | <<CoinMessageEol; |
---|
782 | if (numberTimesOptimal_<4) { |
---|
783 | numberTimesOptimal_++; |
---|
784 | changeMade_++; // say change made |
---|
785 | if (numberTimesOptimal_==1) { |
---|
786 | // better to have small tolerance even if slower |
---|
787 | factorization_->zeroTolerance(1.0e-15); |
---|
788 | } |
---|
789 | lastCleaned=numberIterations_; |
---|
790 | if (primalTolerance_!=dblParam_[ClpPrimalTolerance]) |
---|
791 | handler_->message(CLP_PRIMAL_ORIGINAL,messages_) |
---|
792 | <<CoinMessageEol; |
---|
793 | double oldTolerance = primalTolerance_; |
---|
794 | primalTolerance_=dblParam_[ClpPrimalTolerance]; |
---|
795 | #if 0 |
---|
796 | double * xcost = new double[numberRows_+numberColumns_]; |
---|
797 | double * xlower = new double[numberRows_+numberColumns_]; |
---|
798 | double * xupper = new double[numberRows_+numberColumns_]; |
---|
799 | double * xdj = new double[numberRows_+numberColumns_]; |
---|
800 | double * xsolution = new double[numberRows_+numberColumns_]; |
---|
801 | memcpy(xcost,cost_,(numberRows_+numberColumns_)*sizeof(double)); |
---|
802 | memcpy(xlower,lower_,(numberRows_+numberColumns_)*sizeof(double)); |
---|
803 | memcpy(xupper,upper_,(numberRows_+numberColumns_)*sizeof(double)); |
---|
804 | memcpy(xdj,dj_,(numberRows_+numberColumns_)*sizeof(double)); |
---|
805 | memcpy(xsolution,solution_,(numberRows_+numberColumns_)*sizeof(double)); |
---|
806 | #endif |
---|
807 | // put back original costs and then check |
---|
808 | createRim(4); |
---|
809 | nonLinearCost_->checkInfeasibilities(oldTolerance); |
---|
810 | #if 0 |
---|
811 | int i; |
---|
812 | for (i=0;i<numberRows_+numberColumns_;i++) { |
---|
813 | if (cost_[i]!=xcost[i]) |
---|
814 | printf("** %d old cost %g new %g sol %g\n", |
---|
815 | i,xcost[i],cost_[i],solution_[i]); |
---|
816 | if (lower_[i]!=xlower[i]) |
---|
817 | printf("** %d old lower %g new %g sol %g\n", |
---|
818 | i,xlower[i],lower_[i],solution_[i]); |
---|
819 | if (upper_[i]!=xupper[i]) |
---|
820 | printf("** %d old upper %g new %g sol %g\n", |
---|
821 | i,xupper[i],upper_[i],solution_[i]); |
---|
822 | if (dj_[i]!=xdj[i]) |
---|
823 | printf("** %d old dj %g new %g sol %g\n", |
---|
824 | i,xdj[i],dj_[i],solution_[i]); |
---|
825 | if (solution_[i]!=xsolution[i]) |
---|
826 | printf("** %d old solution %g new %g sol %g\n", |
---|
827 | i,xsolution[i],solution_[i],solution_[i]); |
---|
828 | } |
---|
829 | delete [] xcost; |
---|
830 | delete [] xupper; |
---|
831 | delete [] xlower; |
---|
832 | delete [] xdj; |
---|
833 | delete [] xsolution; |
---|
834 | #endif |
---|
835 | gutsOfSolution(NULL,NULL); |
---|
836 | if (sumOfRelaxedDualInfeasibilities_ == 0.0&& |
---|
837 | sumOfRelaxedPrimalInfeasibilities_ == 0.0) { |
---|
838 | // say optimal (with these bounds etc) |
---|
839 | numberDualInfeasibilities_ = 0; |
---|
840 | sumDualInfeasibilities_ = 0.0; |
---|
841 | numberPrimalInfeasibilities_ = 0; |
---|
842 | sumPrimalInfeasibilities_ = 0.0; |
---|
843 | } |
---|
844 | if (dualFeasible()&&!nonLinearCost_->numberInfeasibilities()&&lastCleaned>=0) |
---|
845 | problemStatus_=0; |
---|
846 | else |
---|
847 | problemStatus_ = -1; |
---|
848 | } else { |
---|
849 | problemStatus_=0; // optimal |
---|
850 | if (lastCleaned<numberIterations_) { |
---|
851 | handler_->message(CLP_SIMPLEX_GIVINGUP,messages_) |
---|
852 | <<CoinMessageEol; |
---|
853 | } |
---|
854 | } |
---|
855 | } else { |
---|
856 | problemStatus_=0; // optimal |
---|
857 | } |
---|
858 | } |
---|
859 | } else { |
---|
860 | // see if looks unbounded |
---|
861 | if (problemStatus_==-5) { |
---|
862 | if (nonLinearCost_->numberInfeasibilities()) { |
---|
863 | if (infeasibilityCost_>1.0e18&&perturbation_==101) { |
---|
864 | // back off weight |
---|
865 | infeasibilityCost_ = 1.0e13; |
---|
866 | unPerturb(); // stop any further perturbation |
---|
867 | } |
---|
868 | //we need infeasiblity cost changed |
---|
869 | if (infeasibilityCost_<1.0e20) { |
---|
870 | infeasibilityCost_ *= 5.0; |
---|
871 | changeMade_++; // say change made |
---|
872 | handler_->message(CLP_PRIMAL_WEIGHT,messages_) |
---|
873 | <<infeasibilityCost_ |
---|
874 | <<CoinMessageEol; |
---|
875 | // put back original costs and then check |
---|
876 | createRim(4); |
---|
877 | gutsOfSolution(NULL,NULL); |
---|
878 | problemStatus_=-1; //continue |
---|
879 | } else { |
---|
880 | // say unbounded |
---|
881 | problemStatus_ = 2; |
---|
882 | } |
---|
883 | } else { |
---|
884 | // say unbounded |
---|
885 | problemStatus_ = 2; |
---|
886 | } |
---|
887 | } else { |
---|
888 | if(type==3&&problemStatus_!=-5) |
---|
889 | unflag(); // odd |
---|
890 | // carry on |
---|
891 | problemStatus_ = -1; |
---|
892 | } |
---|
893 | } |
---|
894 | if (type==0||type==1) { |
---|
895 | if (type!=1||!saveStatus_) { |
---|
896 | // create save arrays |
---|
897 | delete [] saveStatus_; |
---|
898 | delete [] savedSolution_; |
---|
899 | saveStatus_ = new unsigned char [numberRows_+numberColumns_]; |
---|
900 | savedSolution_ = new double [numberRows_+numberColumns_]; |
---|
901 | } |
---|
902 | // save arrays |
---|
903 | memcpy(saveStatus_,status_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
904 | memcpy(savedSolution_+numberColumns_ ,rowActivityWork_, |
---|
905 | numberRows_*sizeof(double)); |
---|
906 | memcpy(savedSolution_ ,columnActivityWork_,numberColumns_*sizeof(double)); |
---|
907 | } |
---|
908 | if (doFactorization) { |
---|
909 | // restore weights (if saved) - also recompute infeasibility list |
---|
910 | if (tentativeStatus>-3) |
---|
911 | primalColumnPivot_->saveWeights(this,(type <2) ? 2 : 4); |
---|
912 | else |
---|
913 | primalColumnPivot_->saveWeights(this,3); |
---|
914 | if (saveThreshold) { |
---|
915 | // use default at present |
---|
916 | factorization_->sparseThreshold(0); |
---|
917 | factorization_->goSparse(); |
---|
918 | } |
---|
919 | } |
---|
920 | if (problemStatus_<0&&!changeMade_) { |
---|
921 | problemStatus_=4; // unknown |
---|
922 | } |
---|
923 | lastGoodIteration_ = numberIterations_; |
---|
924 | if (goToDual) |
---|
925 | problemStatus_=10; // try dual |
---|
926 | #if 0 |
---|
927 | double thisObj = progress->lastObjective(0); |
---|
928 | double lastObj = progress->lastObjective(1); |
---|
929 | if (lastObj<thisObj-1.0e-7*max(fabs(thisObj),fabs(lastObj))-1.0e-8 |
---|
930 | &&firstFree_<0) { |
---|
931 | int maxFactor = factorization_->maximumPivots(); |
---|
932 | if (maxFactor>10) { |
---|
933 | if (forceFactorization_<0) |
---|
934 | forceFactorization_= maxFactor; |
---|
935 | forceFactorization_ = max (1,(forceFactorization_>>1)); |
---|
936 | printf("Reducing factorization frequency\n"); |
---|
937 | } |
---|
938 | } |
---|
939 | #endif |
---|
940 | } |
---|
941 | /* |
---|
942 | Row array has pivot column |
---|
943 | This chooses pivot row. |
---|
944 | For speed, we may need to go to a bucket approach when many |
---|
945 | variables go through bounds |
---|
946 | On exit rhsArray will have changes in costs of basic variables |
---|
947 | */ |
---|
948 | void |
---|
949 | ClpSimplexPrimal::primalRow(CoinIndexedVector * rowArray, |
---|
950 | CoinIndexedVector * rhsArray, |
---|
951 | CoinIndexedVector * spareArray, |
---|
952 | CoinIndexedVector * spareArray2, |
---|
953 | int valuesPass) |
---|
954 | { |
---|
955 | //rowArray->scanAndPack(); |
---|
956 | if (valuesPass) { |
---|
957 | dualIn_ = cost_[sequenceIn_]; |
---|
958 | |
---|
959 | double * work=rowArray->denseVector(); |
---|
960 | int number=rowArray->getNumElements(); |
---|
961 | int * which=rowArray->getIndices(); |
---|
962 | |
---|
963 | int iIndex; |
---|
964 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
965 | |
---|
966 | int iRow = which[iIndex]; |
---|
967 | double alpha = work[iIndex]; |
---|
968 | int iPivot=pivotVariable_[iRow]; |
---|
969 | dualIn_ -= alpha*cost_[iPivot]; |
---|
970 | } |
---|
971 | // determine direction here |
---|
972 | if (dualIn_<-dualTolerance_) { |
---|
973 | directionIn_=1; |
---|
974 | } else if (dualIn_>dualTolerance_) { |
---|
975 | directionIn_=-1; |
---|
976 | } else { |
---|
977 | // towards nearest bound |
---|
978 | if (valueIn_-lowerIn_<upperIn_-valueIn_) { |
---|
979 | directionIn_=-1; |
---|
980 | dualIn_=dualTolerance_; |
---|
981 | } else { |
---|
982 | directionIn_=1; |
---|
983 | dualIn_=-dualTolerance_; |
---|
984 | } |
---|
985 | } |
---|
986 | } |
---|
987 | |
---|
988 | // sequence stays as row number until end |
---|
989 | pivotRow_=-1; |
---|
990 | int numberRemaining=0; |
---|
991 | |
---|
992 | double totalThru=0.0; // for when variables flip |
---|
993 | double acceptablePivot=1.0e-7; |
---|
994 | if (factorization_->pivots()) |
---|
995 | acceptablePivot=1.0e-5; // if we have iterated be more strict |
---|
996 | double bestEverPivot=acceptablePivot; |
---|
997 | int lastPivotRow = -1; |
---|
998 | double lastPivot=0.0; |
---|
999 | double lastTheta=1.0e50; |
---|
1000 | |
---|
1001 | // use spareArrays to put ones looked at in |
---|
1002 | // First one is list of candidates |
---|
1003 | // We could compress if we really know we won't need any more |
---|
1004 | // Second array has current set of pivot candidates |
---|
1005 | // with a backup list saved in double * part of indexed vector |
---|
1006 | |
---|
1007 | // pivot elements |
---|
1008 | double * spare; |
---|
1009 | // indices |
---|
1010 | int * index; |
---|
1011 | spareArray->clear(); |
---|
1012 | spare = spareArray->denseVector(); |
---|
1013 | index = spareArray->getIndices(); |
---|
1014 | |
---|
1015 | // we also need somewhere for effective rhs |
---|
1016 | double * rhs=rhsArray->denseVector(); |
---|
1017 | //int * indexRhs = rhsArray->getIndices(); |
---|
1018 | //int numberFlip=0; // Those which may change if flips |
---|
1019 | |
---|
1020 | /* |
---|
1021 | First we get a list of possible pivots. We can also see if the |
---|
1022 | problem looks unbounded. |
---|
1023 | |
---|
1024 | At first we increase theta and see what happens. We start |
---|
1025 | theta at a reasonable guess. If in right area then we do bit by bit. |
---|
1026 | We save possible pivot candidates |
---|
1027 | |
---|
1028 | */ |
---|
1029 | |
---|
1030 | // do first pass to get possibles |
---|
1031 | // We can also see if unbounded |
---|
1032 | |
---|
1033 | double * work=rowArray->denseVector(); |
---|
1034 | int number=rowArray->getNumElements(); |
---|
1035 | int * which=rowArray->getIndices(); |
---|
1036 | |
---|
1037 | // we need to swap sign if coming in from ub |
---|
1038 | double way = directionIn_; |
---|
1039 | double maximumMovement; |
---|
1040 | if (way>0.0) |
---|
1041 | maximumMovement = min(1.0e30,upperIn_-valueIn_); |
---|
1042 | else |
---|
1043 | maximumMovement = min(1.0e30,valueIn_-lowerIn_); |
---|
1044 | |
---|
1045 | double averageTheta = nonLinearCost_->averageTheta(); |
---|
1046 | double tentativeTheta = min(10.0*averageTheta,maximumMovement); |
---|
1047 | double upperTheta = maximumMovement; |
---|
1048 | if (tentativeTheta>0.5*maximumMovement) |
---|
1049 | tentativeTheta=maximumMovement; |
---|
1050 | |
---|
1051 | double dualCheck = fabs(dualIn_); |
---|
1052 | // but make a bit more pessimistic |
---|
1053 | dualCheck=max(dualCheck-100.0*dualTolerance_,0.99*dualCheck); |
---|
1054 | |
---|
1055 | int iIndex; |
---|
1056 | bool gotList=false; |
---|
1057 | int pivotOne=-1; |
---|
1058 | while (!gotList) { |
---|
1059 | pivotOne=-1; |
---|
1060 | totalThru=0.0; |
---|
1061 | // We also re-compute reduced cost |
---|
1062 | numberRemaining=0; |
---|
1063 | dualIn_ = cost_[sequenceIn_]; |
---|
1064 | double tolerance = primalTolerance_*1.002; |
---|
1065 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
1066 | |
---|
1067 | int iRow = which[iIndex]; |
---|
1068 | double alpha = work[iIndex]; |
---|
1069 | int iPivot=pivotVariable_[iRow]; |
---|
1070 | dualIn_ -= alpha*cost_[iPivot]; |
---|
1071 | alpha *= way; |
---|
1072 | double oldValue = solution_[iPivot]; |
---|
1073 | // get where in bound sequence |
---|
1074 | if (alpha>0.0) { |
---|
1075 | // basic variable going towards lower bound |
---|
1076 | double bound = lower_[iPivot]; |
---|
1077 | oldValue -= bound; |
---|
1078 | } else { |
---|
1079 | // basic variable going towards upper bound |
---|
1080 | double bound = upper_[iPivot]; |
---|
1081 | oldValue = bound-oldValue; |
---|
1082 | } |
---|
1083 | |
---|
1084 | double value = oldValue-tentativeTheta*fabs(alpha); |
---|
1085 | assert (oldValue>=-tolerance); |
---|
1086 | if (value<=tolerance) { |
---|
1087 | value=oldValue-upperTheta*fabs(alpha); |
---|
1088 | if (value<-primalTolerance_&&fabs(alpha)>=acceptablePivot) { |
---|
1089 | upperTheta = (oldValue+primalTolerance_)/fabs(alpha); |
---|
1090 | pivotOne=numberRemaining; |
---|
1091 | } |
---|
1092 | // add to list |
---|
1093 | spare[numberRemaining]=alpha; |
---|
1094 | rhs[numberRemaining]=oldValue; |
---|
1095 | index[numberRemaining++]=iRow; |
---|
1096 | totalThru += fabs(alpha); |
---|
1097 | setActive(iRow); |
---|
1098 | //} else if (value<primalTolerance_*1.002) { |
---|
1099 | // May change if is a flip |
---|
1100 | //indexRhs[numberFlip++]=iRow; |
---|
1101 | } |
---|
1102 | } |
---|
1103 | if (upperTheta<maximumMovement&&totalThru*infeasibilityCost_>=1.0001*dualCheck) { |
---|
1104 | // Can pivot here |
---|
1105 | gotList=true; |
---|
1106 | } else if (tentativeTheta<maximumMovement) { |
---|
1107 | //printf("Going round with average theta of %g\n",averageTheta); |
---|
1108 | tentativeTheta=maximumMovement; |
---|
1109 | } else { |
---|
1110 | gotList=true; |
---|
1111 | } |
---|
1112 | } |
---|
1113 | totalThru=0.0; |
---|
1114 | // we need to keep where rhs non-zeros are |
---|
1115 | int numberInRhs=numberRemaining; |
---|
1116 | memcpy(rhsArray->getIndices(),index,numberInRhs*sizeof(int)); |
---|
1117 | rhsArray->setNumElements(numberInRhs); |
---|
1118 | rhsArray->setPacked(); |
---|
1119 | |
---|
1120 | theta_=maximumMovement; |
---|
1121 | |
---|
1122 | bool goBackOne = false; |
---|
1123 | |
---|
1124 | //printf("%d remain out of %d\n",numberRemaining,number); |
---|
1125 | int iTry=0; |
---|
1126 | #define MAXTRY 1000 |
---|
1127 | if (numberRemaining&&upperTheta<maximumMovement) { |
---|
1128 | // First check if previously chosen one will work |
---|
1129 | if (pivotOne>=0&&0) { |
---|
1130 | double thruCost = infeasibilityCost_*fabs(spare[pivotOne]); |
---|
1131 | if (thruCost>=0.99*fabs(dualIn_)) |
---|
1132 | printf("Could pivot on %d as change %g dj %g\n", |
---|
1133 | index[pivotOne],thruCost,dualIn_); |
---|
1134 | double alpha = spare[pivotOne]; |
---|
1135 | double oldValue = rhs[pivotOne]; |
---|
1136 | theta_ = oldValue/fabs(alpha); |
---|
1137 | pivotRow_=pivotOne; |
---|
1138 | // Stop loop |
---|
1139 | iTry=MAXTRY; |
---|
1140 | } |
---|
1141 | |
---|
1142 | // first get ratio with tolerance |
---|
1143 | for ( ;iTry<MAXTRY;iTry++) { |
---|
1144 | |
---|
1145 | upperTheta=maximumMovement; |
---|
1146 | int iBest=-1; |
---|
1147 | for (iIndex=0;iIndex<numberRemaining;iIndex++) { |
---|
1148 | |
---|
1149 | double alpha = fabs(spare[iIndex]); |
---|
1150 | double oldValue = rhs[iIndex]; |
---|
1151 | double value = oldValue-upperTheta*alpha; |
---|
1152 | |
---|
1153 | if (value<-primalTolerance_ && alpha>=acceptablePivot) { |
---|
1154 | upperTheta = (oldValue+primalTolerance_)/alpha; |
---|
1155 | iBest=iIndex; // just in case weird numbers |
---|
1156 | } |
---|
1157 | } |
---|
1158 | |
---|
1159 | // now look at best in this lot |
---|
1160 | double bestPivot=acceptablePivot; |
---|
1161 | pivotRow_=-1; |
---|
1162 | for (iIndex=0;iIndex<numberRemaining;iIndex++) { |
---|
1163 | |
---|
1164 | int iRow = index[iIndex]; |
---|
1165 | double alpha = spare[iIndex]; |
---|
1166 | double oldValue = rhs[iIndex]; |
---|
1167 | double value = oldValue-upperTheta*fabs(alpha); |
---|
1168 | |
---|
1169 | if (value<=0||iBest==iIndex) { |
---|
1170 | // how much would it cost to go thru and modify bound |
---|
1171 | totalThru += nonLinearCost_->changeInCost(pivotVariable_[iRow],alpha,rhs[iIndex]); |
---|
1172 | setActive(iRow); |
---|
1173 | if (fabs(alpha)>bestPivot) { |
---|
1174 | bestPivot=fabs(alpha); |
---|
1175 | theta_ = oldValue/bestPivot; |
---|
1176 | pivotRow_=iIndex; |
---|
1177 | } |
---|
1178 | } |
---|
1179 | } |
---|
1180 | if (bestPivot<0.1*bestEverPivot&& |
---|
1181 | bestEverPivot>1.0e-6&& bestPivot<1.0e-3) { |
---|
1182 | // back to previous one |
---|
1183 | goBackOne = true; |
---|
1184 | break; |
---|
1185 | } else if (pivotRow_==-1&&upperTheta>largeValue_) { |
---|
1186 | if (lastPivot>acceptablePivot) { |
---|
1187 | // back to previous one |
---|
1188 | goBackOne = true; |
---|
1189 | } else { |
---|
1190 | // can only get here if all pivots so far too small |
---|
1191 | } |
---|
1192 | break; |
---|
1193 | } else if (totalThru>=dualCheck) { |
---|
1194 | break; // no point trying another loop |
---|
1195 | } else { |
---|
1196 | lastPivotRow=pivotRow_; |
---|
1197 | lastTheta = theta_; |
---|
1198 | if (bestPivot>bestEverPivot) |
---|
1199 | bestEverPivot=bestPivot; |
---|
1200 | } |
---|
1201 | } |
---|
1202 | // can get here without pivotRow_ set but with lastPivotRow |
---|
1203 | if (goBackOne||(pivotRow_<0&&lastPivotRow>=0)) { |
---|
1204 | // back to previous one |
---|
1205 | pivotRow_=lastPivotRow; |
---|
1206 | theta_ = lastTheta; |
---|
1207 | } |
---|
1208 | } else { |
---|
1209 | // mark ones which may move |
---|
1210 | //for (int i=0;i<numberFlip;i++) { |
---|
1211 | //int iRow= indexRhs[i]; |
---|
1212 | //setActive(iRow); |
---|
1213 | //} |
---|
1214 | } |
---|
1215 | //if (iTry>50) |
---|
1216 | //printf("** %d tries\n",iTry); |
---|
1217 | if (pivotRow_>=0) { |
---|
1218 | int position=pivotRow_; // position in list |
---|
1219 | pivotRow_=index[position]; |
---|
1220 | //alpha_ = work[pivotRow_]; |
---|
1221 | alpha_ = way*spare[position]; |
---|
1222 | // translate to sequence |
---|
1223 | sequenceOut_ = pivotVariable_[pivotRow_]; |
---|
1224 | valueOut_ = solution(sequenceOut_); |
---|
1225 | lowerOut_=lower_[sequenceOut_]; |
---|
1226 | upperOut_=upper_[sequenceOut_]; |
---|
1227 | #define MINIMUMTHETA 1.0e-12 |
---|
1228 | // Movement should be minimum for anti-degeneracy - unless |
---|
1229 | // fixed variable out |
---|
1230 | double minimumTheta; |
---|
1231 | if (upperOut_>lowerOut_) |
---|
1232 | minimumTheta=MINIMUMTHETA; |
---|
1233 | else |
---|
1234 | minimumTheta=0.0; |
---|
1235 | // will we need to increase tolerance |
---|
1236 | //#define CLP_DEBUG |
---|
1237 | #ifdef CLP_DEBUG |
---|
1238 | bool found=false; |
---|
1239 | #endif |
---|
1240 | double largestInfeasibility = primalTolerance_; |
---|
1241 | if (theta_<minimumTheta&&(specialOptions_&4)==0&&!valuesPass) { |
---|
1242 | theta_=minimumTheta; |
---|
1243 | for (iIndex=0;iIndex<numberRemaining-numberRemaining;iIndex++) { |
---|
1244 | largestInfeasibility = max (largestInfeasibility, |
---|
1245 | -(rhs[iIndex]-fabs(spare[iIndex])*theta_)); |
---|
1246 | } |
---|
1247 | #define CLP_DEBUG |
---|
1248 | #ifdef CLP_DEBUG |
---|
1249 | if (largestInfeasibility>primalTolerance_&&(handler_->logLevel()&32)>-1) |
---|
1250 | printf("Primal tolerance increased from %g to %g\n", |
---|
1251 | primalTolerance_,largestInfeasibility); |
---|
1252 | #endif |
---|
1253 | #undef CLP_DEBUG |
---|
1254 | primalTolerance_ = max(primalTolerance_,largestInfeasibility); |
---|
1255 | } |
---|
1256 | // Need to look at all in some cases |
---|
1257 | if (theta_>tentativeTheta) { |
---|
1258 | for (iIndex=0;iIndex<number;iIndex++) |
---|
1259 | setActive(which[iIndex]); |
---|
1260 | } |
---|
1261 | if (way<0.0) |
---|
1262 | theta_ = - theta_; |
---|
1263 | double newValue = valueOut_ - theta_*alpha_; |
---|
1264 | // If 4 bit set - Force outgoing variables to exact bound (primal) |
---|
1265 | if (alpha_*way<0.0) { |
---|
1266 | directionOut_=-1; // to upper bound |
---|
1267 | if (fabs(theta_)>1.0e-6||(specialOptions_&4)!=0) { |
---|
1268 | upperOut_ = nonLinearCost_->nearest(sequenceOut_,newValue); |
---|
1269 | } else { |
---|
1270 | upperOut_ = newValue; |
---|
1271 | } |
---|
1272 | } else { |
---|
1273 | directionOut_=1; // to lower bound |
---|
1274 | if (fabs(theta_)>1.0e-6||(specialOptions_&4)!=0) { |
---|
1275 | lowerOut_ = nonLinearCost_->nearest(sequenceOut_,newValue); |
---|
1276 | } else { |
---|
1277 | lowerOut_ = newValue; |
---|
1278 | } |
---|
1279 | } |
---|
1280 | dualOut_ = reducedCost(sequenceOut_); |
---|
1281 | } else if (maximumMovement<1.0e20) { |
---|
1282 | // flip |
---|
1283 | pivotRow_ = -2; // so we can tell its a flip |
---|
1284 | sequenceOut_ = sequenceIn_; |
---|
1285 | valueOut_ = valueIn_; |
---|
1286 | dualOut_ = dualIn_; |
---|
1287 | lowerOut_ = lowerIn_; |
---|
1288 | upperOut_ = upperIn_; |
---|
1289 | alpha_ = 0.0; |
---|
1290 | if (way<0.0) { |
---|
1291 | directionOut_=1; // to lower bound |
---|
1292 | theta_ = lowerOut_ - valueOut_; |
---|
1293 | } else { |
---|
1294 | directionOut_=-1; // to upper bound |
---|
1295 | theta_ = upperOut_ - valueOut_; |
---|
1296 | } |
---|
1297 | } |
---|
1298 | |
---|
1299 | double theta1 = max(theta_,1.0e-12); |
---|
1300 | double theta2 = numberIterations_*nonLinearCost_->averageTheta(); |
---|
1301 | // Set average theta |
---|
1302 | nonLinearCost_->setAverageTheta((theta1+theta2)/((double) (numberIterations_+1))); |
---|
1303 | //if (numberIterations_%1000==0) |
---|
1304 | //printf("average theta is %g\n",nonLinearCost_->averageTheta()); |
---|
1305 | |
---|
1306 | // clear arrays |
---|
1307 | |
---|
1308 | memset(spare,0,numberRemaining*sizeof(double)); |
---|
1309 | |
---|
1310 | // put back original bounds etc |
---|
1311 | nonLinearCost_->goBackAll(rhsArray); |
---|
1312 | |
---|
1313 | rhsArray->clear(); |
---|
1314 | |
---|
1315 | } |
---|
1316 | /* |
---|
1317 | Chooses primal pivot column |
---|
1318 | updateArray has cost updates (also use pivotRow_ from last iteration) |
---|
1319 | Would be faster with separate region to scan |
---|
1320 | and will have this (with square of infeasibility) when steepest |
---|
1321 | For easy problems we can just choose one of the first columns we look at |
---|
1322 | */ |
---|
1323 | void |
---|
1324 | ClpSimplexPrimal::primalColumn(CoinIndexedVector * updates, |
---|
1325 | CoinIndexedVector * spareRow1, |
---|
1326 | CoinIndexedVector * spareRow2, |
---|
1327 | CoinIndexedVector * spareColumn1, |
---|
1328 | CoinIndexedVector * spareColumn2) |
---|
1329 | { |
---|
1330 | sequenceIn_ = primalColumnPivot_->pivotColumn(updates,spareRow1, |
---|
1331 | spareRow2,spareColumn1, |
---|
1332 | spareColumn2); |
---|
1333 | if (sequenceIn_>=0) { |
---|
1334 | valueIn_=solution_[sequenceIn_]; |
---|
1335 | dualIn_=dj_[sequenceIn_]; |
---|
1336 | if (nonLinearCost_->lookBothWays()) { |
---|
1337 | // double check |
---|
1338 | ClpSimplex::Status status = getStatus(sequenceIn_); |
---|
1339 | |
---|
1340 | switch(status) { |
---|
1341 | case ClpSimplex::atUpperBound: |
---|
1342 | if (dualIn_<0.0) { |
---|
1343 | // move to other side |
---|
1344 | printf("For %d U (%g, %g, %g) dj changed from %g", |
---|
1345 | sequenceIn_,lower_[sequenceIn_],solution_[sequenceIn_], |
---|
1346 | upper_[sequenceIn_],dualIn_); |
---|
1347 | dualIn_ -= nonLinearCost_->changeUpInCost(sequenceIn_); |
---|
1348 | printf(" to %g\n",dualIn_); |
---|
1349 | nonLinearCost_->setOne(sequenceIn_,upper_[sequenceIn_]+2.0*currentPrimalTolerance()); |
---|
1350 | setStatus(sequenceIn_,ClpSimplex::atLowerBound); |
---|
1351 | } |
---|
1352 | break; |
---|
1353 | case ClpSimplex::atLowerBound: |
---|
1354 | if (dualIn_>0.0) { |
---|
1355 | // move to other side |
---|
1356 | printf("For %d L (%g, %g, %g) dj changed from %g", |
---|
1357 | sequenceIn_,lower_[sequenceIn_],solution_[sequenceIn_], |
---|
1358 | upper_[sequenceIn_],dualIn_); |
---|
1359 | dualIn_ -= nonLinearCost_->changeDownInCost(sequenceIn_); |
---|
1360 | printf(" to %g\n",dualIn_); |
---|
1361 | nonLinearCost_->setOne(sequenceIn_,lower_[sequenceIn_]-2.0*currentPrimalTolerance()); |
---|
1362 | setStatus(sequenceIn_,ClpSimplex::atUpperBound); |
---|
1363 | } |
---|
1364 | break; |
---|
1365 | default: |
---|
1366 | break; |
---|
1367 | } |
---|
1368 | } |
---|
1369 | lowerIn_=lower_[sequenceIn_]; |
---|
1370 | upperIn_=upper_[sequenceIn_]; |
---|
1371 | if (dualIn_>0.0) |
---|
1372 | directionIn_ = -1; |
---|
1373 | else |
---|
1374 | directionIn_ = 1; |
---|
1375 | } else { |
---|
1376 | sequenceIn_ = -1; |
---|
1377 | } |
---|
1378 | } |
---|
1379 | /* The primals are updated by the given array. |
---|
1380 | Returns number of infeasibilities. |
---|
1381 | After rowArray will have list of cost changes |
---|
1382 | */ |
---|
1383 | int |
---|
1384 | ClpSimplexPrimal::updatePrimalsInPrimal(CoinIndexedVector * rowArray, |
---|
1385 | double theta, |
---|
1386 | double & objectiveChange, |
---|
1387 | int valuesPass) |
---|
1388 | { |
---|
1389 | // Cost on pivot row may change - may need to change dualIn |
---|
1390 | double oldCost=0.0; |
---|
1391 | if (pivotRow_>=0) |
---|
1392 | oldCost = cost_[sequenceOut_]; |
---|
1393 | //rowArray->scanAndPack(); |
---|
1394 | double * work=rowArray->denseVector(); |
---|
1395 | int number=rowArray->getNumElements(); |
---|
1396 | int * which=rowArray->getIndices(); |
---|
1397 | |
---|
1398 | int newNumber = 0; |
---|
1399 | int pivotPosition = -1; |
---|
1400 | nonLinearCost_->setChangeInCost(0.0); |
---|
1401 | int iIndex; |
---|
1402 | if (!valuesPass) { |
---|
1403 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
1404 | |
---|
1405 | int iRow = which[iIndex]; |
---|
1406 | double alpha = work[iIndex]; |
---|
1407 | work[iIndex]=0.0; |
---|
1408 | int iPivot=pivotVariable_[iRow]; |
---|
1409 | double change = theta*alpha; |
---|
1410 | double value = solution_[iPivot] - change; |
---|
1411 | solution_[iPivot]=value; |
---|
1412 | #ifndef NDEBUG |
---|
1413 | // check if not active then okay |
---|
1414 | if (!active(iRow)) { |
---|
1415 | // But make sure one going out is feasible |
---|
1416 | if (change>0.0) { |
---|
1417 | // going down |
---|
1418 | if (value<lower_[iPivot]+primalTolerance_) { |
---|
1419 | if (iPivot==sequenceOut_&&value>lower_[iPivot]-1.001*primalTolerance_) |
---|
1420 | value=lower_[iPivot]; |
---|
1421 | double difference = nonLinearCost_->setOne(iPivot,value); |
---|
1422 | assert (!difference); |
---|
1423 | } |
---|
1424 | } else { |
---|
1425 | // going up |
---|
1426 | if (value>upper_[iPivot]-primalTolerance_) { |
---|
1427 | if (iPivot==sequenceOut_&&value<upper_[iPivot]+1.001*primalTolerance_) |
---|
1428 | value=upper_[iPivot]; |
---|
1429 | double difference = nonLinearCost_->setOne(iPivot,value); |
---|
1430 | assert (!difference); |
---|
1431 | } |
---|
1432 | } |
---|
1433 | } |
---|
1434 | #endif |
---|
1435 | if (active(iRow)) { |
---|
1436 | clearActive(iRow); |
---|
1437 | // But make sure one going out is feasible |
---|
1438 | if (change>0.0) { |
---|
1439 | // going down |
---|
1440 | if (value<lower_[iPivot]+primalTolerance_) { |
---|
1441 | if (iPivot==sequenceOut_&&value>lower_[iPivot]-1.001*primalTolerance_) |
---|
1442 | value=lower_[iPivot]; |
---|
1443 | double difference = nonLinearCost_->setOne(iPivot,value); |
---|
1444 | if (difference) { |
---|
1445 | if (iRow==pivotRow_) |
---|
1446 | pivotPosition=newNumber; |
---|
1447 | work[newNumber] = difference; |
---|
1448 | //change reduced cost on this |
---|
1449 | dj_[iPivot] = -difference; |
---|
1450 | which[newNumber++]=iRow; |
---|
1451 | } |
---|
1452 | } |
---|
1453 | } else { |
---|
1454 | // going up |
---|
1455 | if (value>upper_[iPivot]-primalTolerance_) { |
---|
1456 | if (iPivot==sequenceOut_&&value<upper_[iPivot]+1.001*primalTolerance_) |
---|
1457 | value=upper_[iPivot]; |
---|
1458 | double difference = nonLinearCost_->setOne(iPivot,value); |
---|
1459 | if (difference) { |
---|
1460 | if (iRow==pivotRow_) |
---|
1461 | pivotPosition=newNumber; |
---|
1462 | work[newNumber] = difference; |
---|
1463 | //change reduced cost on this |
---|
1464 | dj_[iPivot] = -difference; |
---|
1465 | which[newNumber++]=iRow; |
---|
1466 | } |
---|
1467 | } |
---|
1468 | } |
---|
1469 | } |
---|
1470 | } |
---|
1471 | } else { |
---|
1472 | // values pass so look at all |
---|
1473 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
1474 | |
---|
1475 | int iRow = which[iIndex]; |
---|
1476 | double alpha = work[iIndex]; |
---|
1477 | work[iIndex]=0.0; |
---|
1478 | int iPivot=pivotVariable_[iRow]; |
---|
1479 | double change = theta*alpha; |
---|
1480 | double value = solution_[iPivot] - change; |
---|
1481 | solution_[iPivot]=value; |
---|
1482 | clearActive(iRow); |
---|
1483 | // But make sure one going out is feasible |
---|
1484 | if (change>0.0) { |
---|
1485 | // going down |
---|
1486 | if (value<lower_[iPivot]+primalTolerance_) { |
---|
1487 | if (iPivot==sequenceOut_&&value>lower_[iPivot]-1.001*primalTolerance_) |
---|
1488 | value=lower_[iPivot]; |
---|
1489 | double difference = nonLinearCost_->setOne(iPivot,value); |
---|
1490 | if (difference) { |
---|
1491 | if (iRow==pivotRow_) |
---|
1492 | pivotPosition=newNumber; |
---|
1493 | work[newNumber] = difference; |
---|
1494 | //change reduced cost on this |
---|
1495 | dj_[iPivot] = -difference; |
---|
1496 | which[newNumber++]=iRow; |
---|
1497 | } |
---|
1498 | } |
---|
1499 | } else { |
---|
1500 | // going up |
---|
1501 | if (value>upper_[iPivot]-primalTolerance_) { |
---|
1502 | if (iPivot==sequenceOut_&&value<upper_[iPivot]+1.001*primalTolerance_) |
---|
1503 | value=upper_[iPivot]; |
---|
1504 | double difference = nonLinearCost_->setOne(iPivot,value); |
---|
1505 | if (difference) { |
---|
1506 | if (iRow==pivotRow_) |
---|
1507 | pivotPosition=newNumber; |
---|
1508 | work[newNumber] = difference; |
---|
1509 | //change reduced cost on this |
---|
1510 | dj_[iPivot] = -difference; |
---|
1511 | which[newNumber++]=iRow; |
---|
1512 | } |
---|
1513 | } |
---|
1514 | } |
---|
1515 | } |
---|
1516 | } |
---|
1517 | objectiveChange += nonLinearCost_->changeInCost(); |
---|
1518 | rowArray->setPacked(); |
---|
1519 | #if 0 |
---|
1520 | rowArray->setNumElements(newNumber); |
---|
1521 | rowArray->expand(); |
---|
1522 | if (pivotRow_>=0) { |
---|
1523 | dualIn_ += (oldCost-cost_[sequenceOut_]); |
---|
1524 | // update change vector to include pivot |
---|
1525 | rowArray->add(pivotRow_,-dualIn_); |
---|
1526 | // and convert to packed |
---|
1527 | rowArray->scanAndPack(); |
---|
1528 | } else { |
---|
1529 | // and convert to packed |
---|
1530 | rowArray->scanAndPack(); |
---|
1531 | } |
---|
1532 | #else |
---|
1533 | if (pivotRow_>=0) { |
---|
1534 | dualIn_ += (oldCost-cost_[sequenceOut_]); |
---|
1535 | // update change vector to include pivot |
---|
1536 | if (pivotPosition>=0) { |
---|
1537 | work[pivotPosition] -= dualIn_; |
---|
1538 | } else { |
---|
1539 | work[newNumber]=-dualIn_; |
---|
1540 | which[newNumber++]=pivotRow_; |
---|
1541 | } |
---|
1542 | } |
---|
1543 | rowArray->setNumElements(newNumber); |
---|
1544 | #endif |
---|
1545 | return 0; |
---|
1546 | } |
---|
1547 | // Perturbs problem |
---|
1548 | void |
---|
1549 | ClpSimplexPrimal::perturb(int type) |
---|
1550 | { |
---|
1551 | if (perturbation_>100) |
---|
1552 | return; //perturbed already |
---|
1553 | if (perturbation_==100) |
---|
1554 | perturbation_=50; // treat as normal |
---|
1555 | int savePerturbation = perturbation_; |
---|
1556 | int i; |
---|
1557 | if (!numberIterations_) |
---|
1558 | cleanStatus(); // make sure status okay |
---|
1559 | if (!numberIterations_&&perturbation_==50) { |
---|
1560 | // See if we need to perturb |
---|
1561 | double * sort = new double[numberRows_]; |
---|
1562 | for (i=0;i<numberRows_;i++) { |
---|
1563 | double lo = fabs(lower_[i]); |
---|
1564 | double up = fabs(upper_[i]); |
---|
1565 | double value=0.0; |
---|
1566 | if (lo&&lo<1.0e20) { |
---|
1567 | if (up&&up<1.0e20) |
---|
1568 | value = 0.5*(lo+up); |
---|
1569 | else |
---|
1570 | value=lo; |
---|
1571 | } else { |
---|
1572 | if (up&&up<1.0e20) |
---|
1573 | value = up; |
---|
1574 | } |
---|
1575 | sort[i]=value; |
---|
1576 | } |
---|
1577 | std::sort(sort,sort+numberRows_); |
---|
1578 | int number=1; |
---|
1579 | double last = sort[0]; |
---|
1580 | for (i=1;i<numberRows_;i++) { |
---|
1581 | if (last!=sort[i]) |
---|
1582 | number++; |
---|
1583 | last=sort[i]; |
---|
1584 | } |
---|
1585 | delete [] sort; |
---|
1586 | if (number*4>numberRows_) { |
---|
1587 | perturbation_=100; |
---|
1588 | return; // good enough |
---|
1589 | } |
---|
1590 | } |
---|
1591 | // primal perturbation |
---|
1592 | double perturbation=1.0e-20; |
---|
1593 | int numberNonZero=0; |
---|
1594 | // maximum fraction of rhs/bounds to perturb |
---|
1595 | double maximumFraction = 1.0e-5; |
---|
1596 | if (perturbation_>=50) { |
---|
1597 | perturbation = 1.0e-4; |
---|
1598 | for (i=0;i<numberColumns_+numberRows_;i++) { |
---|
1599 | if (upper_[i]>lower_[i]+primalTolerance_) { |
---|
1600 | double lowerValue, upperValue; |
---|
1601 | if (lower_[i]>-1.0e20) |
---|
1602 | lowerValue = fabs(lower_[i]); |
---|
1603 | else |
---|
1604 | lowerValue=0.0; |
---|
1605 | if (upper_[i]<1.0e20) |
---|
1606 | upperValue = fabs(upper_[i]); |
---|
1607 | else |
---|
1608 | upperValue=0.0; |
---|
1609 | double value = max(fabs(lowerValue),fabs(upperValue)); |
---|
1610 | value = min(value,upper_[i]-lower_[i]); |
---|
1611 | #if 1 |
---|
1612 | if (value) { |
---|
1613 | perturbation += value; |
---|
1614 | numberNonZero++; |
---|
1615 | } |
---|
1616 | #else |
---|
1617 | perturbation = max(perturbation,value); |
---|
1618 | #endif |
---|
1619 | } |
---|
1620 | } |
---|
1621 | if (numberNonZero) |
---|
1622 | perturbation /= (double) numberNonZero; |
---|
1623 | else |
---|
1624 | perturbation = 1.0e-1; |
---|
1625 | } else if (perturbation_<100) { |
---|
1626 | perturbation = pow(10.0,perturbation_); |
---|
1627 | // user is in charge |
---|
1628 | maximumFraction = 1.0; |
---|
1629 | } |
---|
1630 | double largestZero=0.0; |
---|
1631 | double largest=0.0; |
---|
1632 | double largestPerCent=0.0; |
---|
1633 | bool printOut=(handler_->logLevel()==63); |
---|
1634 | printOut=false; //off |
---|
1635 | // Check if all slack |
---|
1636 | int number=0; |
---|
1637 | int iSequence; |
---|
1638 | for (iSequence=0;iSequence<numberRows_;iSequence++) { |
---|
1639 | if (getRowStatus(iSequence)==basic) |
---|
1640 | number++; |
---|
1641 | } |
---|
1642 | if (number!=numberRows_) |
---|
1643 | type=1; |
---|
1644 | // modify bounds |
---|
1645 | // Change so at least 1.0e-5 and no more than 0.1 |
---|
1646 | // For now just no more than 0.1 |
---|
1647 | // printf("Pert type %d perturbation %g, maxF %g\n",type,perturbation,maximumFraction); |
---|
1648 | if (type==1) { |
---|
1649 | //double multiplier = perturbation*maximumFraction; |
---|
1650 | for (iSequence=0;iSequence<numberRows_+numberColumns_;iSequence++) { |
---|
1651 | if (getStatus(iSequence)==basic) { |
---|
1652 | double solutionValue = solution_[iSequence]; |
---|
1653 | double lowerValue = lower_[iSequence]; |
---|
1654 | double upperValue = upper_[iSequence]; |
---|
1655 | double difference = upperValue-lowerValue; |
---|
1656 | difference = min(difference,perturbation); |
---|
1657 | difference = min(difference,fabs(solutionValue)+1.0); |
---|
1658 | double value = maximumFraction*(difference+1.0); |
---|
1659 | value = min(value,0.1); |
---|
1660 | value *= CoinDrand48(); |
---|
1661 | if (solutionValue-lowerValue<=primalTolerance_) { |
---|
1662 | lower_[iSequence] -= value; |
---|
1663 | } else if (upperValue-solutionValue<=primalTolerance_) { |
---|
1664 | upper_[iSequence] += value; |
---|
1665 | } else { |
---|
1666 | #if 0 |
---|
1667 | if (iSequence>=numberColumns_) { |
---|
1668 | // may not be at bound - but still perturb (unless free) |
---|
1669 | if (upperValue>1.0e30&&lowerValue<-1.0e30) |
---|
1670 | value=0.0; |
---|
1671 | else |
---|
1672 | value = - value; // as -1.0 in matrix |
---|
1673 | } else { |
---|
1674 | value = 0.0; |
---|
1675 | } |
---|
1676 | #else |
---|
1677 | value=0.0; |
---|
1678 | #endif |
---|
1679 | } |
---|
1680 | if (value) { |
---|
1681 | if (printOut) |
---|
1682 | printf("col %d lower from %g to %g, upper from %g to %g\n", |
---|
1683 | iSequence,lower_[iSequence],lowerValue,upper_[iSequence],upperValue); |
---|
1684 | if (solutionValue) { |
---|
1685 | largest = max(largest,value); |
---|
1686 | if (value>(fabs(solutionValue)+1.0)*largestPerCent) |
---|
1687 | largestPerCent=value/(fabs(solutionValue)+1.0); |
---|
1688 | } else { |
---|
1689 | largestZero = max(largestZero,value); |
---|
1690 | } |
---|
1691 | } |
---|
1692 | } |
---|
1693 | } |
---|
1694 | } else { |
---|
1695 | double tolerance = 100.0*primalTolerance_; |
---|
1696 | for (i=0;i<numberColumns_;i++) { |
---|
1697 | double lowerValue=lower_[i], upperValue=upper_[i]; |
---|
1698 | if (upperValue>lowerValue+primalTolerance_) { |
---|
1699 | double value = perturbation*maximumFraction; |
---|
1700 | value = min(value,0.1); |
---|
1701 | value *= CoinDrand48(); |
---|
1702 | if (savePerturbation!=50) { |
---|
1703 | if (fabs(value)<=primalTolerance_) |
---|
1704 | value=0.0; |
---|
1705 | if (lowerValue>-1.0e20&&lowerValue) |
---|
1706 | lowerValue -= value * (max(1.0e-2,1.0e-5*fabs(lowerValue))); |
---|
1707 | if (upperValue<1.0e20&&upperValue) |
---|
1708 | upperValue += value * (max(1.0e-2,1.0e-5*fabs(upperValue))); |
---|
1709 | } else if (value) { |
---|
1710 | double valueL =value *(max(1.0e-2,1.0e-5*fabs(lowerValue))); |
---|
1711 | // get in range |
---|
1712 | if (valueL<=tolerance) { |
---|
1713 | valueL *= 10.0; |
---|
1714 | while (valueL<=tolerance) |
---|
1715 | valueL *= 10.0; |
---|
1716 | } else if (valueL>1.0) { |
---|
1717 | valueL *= 0.1; |
---|
1718 | while (valueL>1.0) |
---|
1719 | valueL *= 0.1; |
---|
1720 | } |
---|
1721 | if (lowerValue>-1.0e20&&lowerValue) |
---|
1722 | lowerValue -= valueL; |
---|
1723 | double valueU =value *(max(1.0e-2,1.0e-5*fabs(upperValue))); |
---|
1724 | // get in range |
---|
1725 | if (valueU<=tolerance) { |
---|
1726 | valueU *= 10.0; |
---|
1727 | while (valueU<=tolerance) |
---|
1728 | valueU *= 10.0; |
---|
1729 | } else if (valueU>1.0) { |
---|
1730 | valueU *= 0.1; |
---|
1731 | while (valueU>1.0) |
---|
1732 | valueU *= 0.1; |
---|
1733 | } |
---|
1734 | if (upperValue<1.0e20&&upperValue) |
---|
1735 | upperValue += valueU; |
---|
1736 | } |
---|
1737 | if (lowerValue!=lower_[i]) { |
---|
1738 | double difference = fabs(lowerValue-lower_[i]); |
---|
1739 | largest = max(largest,difference); |
---|
1740 | if (difference>fabs(lower_[i])*largestPerCent) |
---|
1741 | largestPerCent=fabs(difference/lower_[i]); |
---|
1742 | } |
---|
1743 | if (upperValue!=upper_[i]) { |
---|
1744 | double difference = fabs(upperValue-upper_[i]); |
---|
1745 | largest = max(largest,difference); |
---|
1746 | if (difference>fabs(upper_[i])*largestPerCent) |
---|
1747 | largestPerCent=fabs(difference/upper_[i]); |
---|
1748 | } |
---|
1749 | if (printOut) |
---|
1750 | printf("col %d lower from %g to %g, upper from %g to %g\n", |
---|
1751 | i,lower_[i],lowerValue,upper_[i],upperValue); |
---|
1752 | } |
---|
1753 | lower_[i]=lowerValue; |
---|
1754 | upper_[i]=upperValue; |
---|
1755 | } |
---|
1756 | for (;i<numberColumns_+numberRows_;i++) { |
---|
1757 | double lowerValue=lower_[i], upperValue=upper_[i]; |
---|
1758 | double value = perturbation*maximumFraction; |
---|
1759 | value = min(value,0.1); |
---|
1760 | value *= CoinDrand48(); |
---|
1761 | if (upperValue>lowerValue+tolerance) { |
---|
1762 | if (savePerturbation!=50) { |
---|
1763 | if (fabs(value)<=primalTolerance_) |
---|
1764 | value=0.0; |
---|
1765 | if (lowerValue>-1.0e20&&lowerValue) |
---|
1766 | lowerValue -= value * (max(1.0e-2,1.0e-5*fabs(lowerValue))); |
---|
1767 | if (upperValue<1.0e20&&upperValue) |
---|
1768 | upperValue += value * (max(1.0e-2,1.0e-5*fabs(upperValue))); |
---|
1769 | } else if (value) { |
---|
1770 | double valueL =value *(max(1.0e-2,1.0e-5*fabs(lowerValue))); |
---|
1771 | // get in range |
---|
1772 | if (valueL<=tolerance) { |
---|
1773 | valueL *= 10.0; |
---|
1774 | while (valueL<=tolerance) |
---|
1775 | valueL *= 10.0; |
---|
1776 | } else if (valueL>1.0) { |
---|
1777 | valueL *= 0.1; |
---|
1778 | while (valueL>1.0) |
---|
1779 | valueL *= 0.1; |
---|
1780 | } |
---|
1781 | if (lowerValue>-1.0e20&&lowerValue) |
---|
1782 | lowerValue -= valueL; |
---|
1783 | double valueU =value *(max(1.0e-2,1.0e-5*fabs(upperValue))); |
---|
1784 | // get in range |
---|
1785 | if (valueU<=tolerance) { |
---|
1786 | valueU *= 10.0; |
---|
1787 | while (valueU<=tolerance) |
---|
1788 | valueU *= 10.0; |
---|
1789 | } else if (valueU>1.0) { |
---|
1790 | valueU *= 0.1; |
---|
1791 | while (valueU>1.0) |
---|
1792 | valueU *= 0.1; |
---|
1793 | } |
---|
1794 | if (upperValue<1.0e20&&upperValue) |
---|
1795 | upperValue += valueU; |
---|
1796 | } |
---|
1797 | } else if (upperValue>0.0) { |
---|
1798 | upperValue -= value * (max(1.0e-2,1.0e-5*fabs(lowerValue))); |
---|
1799 | lowerValue -= value * (max(1.0e-2,1.0e-5*fabs(lowerValue))); |
---|
1800 | } else if (upperValue<0.0) { |
---|
1801 | upperValue += value * (max(1.0e-2,1.0e-5*fabs(lowerValue))); |
---|
1802 | lowerValue += value * (max(1.0e-2,1.0e-5*fabs(lowerValue))); |
---|
1803 | } else { |
---|
1804 | } |
---|
1805 | if (lowerValue!=lower_[i]) { |
---|
1806 | double difference = fabs(lowerValue-lower_[i]); |
---|
1807 | largest = max(largest,difference); |
---|
1808 | if (difference>fabs(lower_[i])*largestPerCent) |
---|
1809 | largestPerCent=fabs(difference/lower_[i]); |
---|
1810 | } |
---|
1811 | if (upperValue!=upper_[i]) { |
---|
1812 | double difference = fabs(upperValue-upper_[i]); |
---|
1813 | largest = max(largest,difference); |
---|
1814 | if (difference>fabs(upper_[i])*largestPerCent) |
---|
1815 | largestPerCent=fabs(difference/upper_[i]); |
---|
1816 | } |
---|
1817 | if (printOut) |
---|
1818 | printf("row %d lower from %g to %g, upper from %g to %g\n", |
---|
1819 | i-numberColumns_,lower_[i],lowerValue,upper_[i],upperValue); |
---|
1820 | lower_[i]=lowerValue; |
---|
1821 | upper_[i]=upperValue; |
---|
1822 | } |
---|
1823 | } |
---|
1824 | // Clean up |
---|
1825 | for (i=0;i<numberColumns_+numberRows_;i++) { |
---|
1826 | switch(getStatus(i)) { |
---|
1827 | |
---|
1828 | case basic: |
---|
1829 | break; |
---|
1830 | case atUpperBound: |
---|
1831 | solution_[i]=upper_[i]; |
---|
1832 | break; |
---|
1833 | case isFixed: |
---|
1834 | case atLowerBound: |
---|
1835 | solution_[i]=lower_[i]; |
---|
1836 | break; |
---|
1837 | case isFree: |
---|
1838 | break; |
---|
1839 | case superBasic: |
---|
1840 | break; |
---|
1841 | } |
---|
1842 | } |
---|
1843 | handler_->message(CLP_SIMPLEX_PERTURB,messages_) |
---|
1844 | <<100.0*maximumFraction<<perturbation<<largest<<100.0*largestPerCent<<largestZero |
---|
1845 | <<CoinMessageEol; |
---|
1846 | // redo nonlinear costs |
---|
1847 | // say perturbed |
---|
1848 | perturbation_=101; |
---|
1849 | } |
---|
1850 | // un perturb |
---|
1851 | bool |
---|
1852 | ClpSimplexPrimal::unPerturb() |
---|
1853 | { |
---|
1854 | if (perturbation_!=101) |
---|
1855 | return false; |
---|
1856 | // put back original bounds and costs |
---|
1857 | createRim(7); |
---|
1858 | sanityCheck(); |
---|
1859 | // unflag |
---|
1860 | unflag(); |
---|
1861 | // get a valid nonlinear cost function |
---|
1862 | delete nonLinearCost_; |
---|
1863 | nonLinearCost_= new ClpNonLinearCost(this); |
---|
1864 | perturbation_ = 102; // stop any further perturbation |
---|
1865 | // move non basic variables to new bounds |
---|
1866 | nonLinearCost_->checkInfeasibilities(0.0); |
---|
1867 | #if 1 |
---|
1868 | // Try using dual |
---|
1869 | return true; |
---|
1870 | #else |
---|
1871 | gutsOfSolution(NULL,NULL); |
---|
1872 | return false; |
---|
1873 | #endif |
---|
1874 | |
---|
1875 | } |
---|
1876 | // Unflag all variables and return number unflagged |
---|
1877 | int |
---|
1878 | ClpSimplexPrimal::unflag() |
---|
1879 | { |
---|
1880 | int i; |
---|
1881 | int number = numberRows_+numberColumns_; |
---|
1882 | int numberFlagged=0; |
---|
1883 | for (i=0;i<number;i++) { |
---|
1884 | if (flagged(i)) { |
---|
1885 | clearFlagged(i); |
---|
1886 | numberFlagged++; |
---|
1887 | } |
---|
1888 | } |
---|
1889 | return numberFlagged; |
---|
1890 | } |
---|
1891 | // Do not change infeasibility cost and always say optimal |
---|
1892 | void |
---|
1893 | ClpSimplexPrimal::alwaysOptimal(bool onOff) |
---|
1894 | { |
---|
1895 | if (onOff) |
---|
1896 | specialOptions_ |= 1; |
---|
1897 | else |
---|
1898 | specialOptions_ &= ~1; |
---|
1899 | } |
---|
1900 | bool |
---|
1901 | ClpSimplexPrimal::alwaysOptimal() const |
---|
1902 | { |
---|
1903 | return (specialOptions_&1)!=0; |
---|
1904 | } |
---|
1905 | // Flatten outgoing variables i.e. - always to exact bound |
---|
1906 | void |
---|
1907 | ClpSimplexPrimal::exactOutgoing(bool onOff) |
---|
1908 | { |
---|
1909 | if (onOff) |
---|
1910 | specialOptions_ |= 4; |
---|
1911 | else |
---|
1912 | specialOptions_ &= ~4; |
---|
1913 | } |
---|
1914 | bool |
---|
1915 | ClpSimplexPrimal::exactOutgoing() const |
---|
1916 | { |
---|
1917 | return (specialOptions_&4)!=0; |
---|
1918 | } |
---|
1919 | /* |
---|
1920 | Reasons to come out (normal mode/user mode): |
---|
1921 | -1 normal |
---|
1922 | -2 factorize now - good iteration/ NA |
---|
1923 | -3 slight inaccuracy - refactorize - iteration done/ same but factor done |
---|
1924 | -4 inaccuracy - refactorize - no iteration/ NA |
---|
1925 | -5 something flagged - go round again/ pivot not possible |
---|
1926 | +2 looks unbounded |
---|
1927 | +3 max iterations (iteration done) |
---|
1928 | */ |
---|
1929 | int |
---|
1930 | ClpSimplexPrimal::pivotResult(int ifValuesPass) |
---|
1931 | { |
---|
1932 | |
---|
1933 | bool roundAgain=true; |
---|
1934 | int returnCode=-1; |
---|
1935 | |
---|
1936 | // loop round if user setting and doing refactorization |
---|
1937 | while (roundAgain) { |
---|
1938 | roundAgain=false; |
---|
1939 | returnCode=-1; |
---|
1940 | pivotRow_=-1; |
---|
1941 | sequenceOut_=-1; |
---|
1942 | rowArray_[1]->clear(); |
---|
1943 | #if 0 |
---|
1944 | { |
---|
1945 | int seq[]={612,643}; |
---|
1946 | int k; |
---|
1947 | for (k=0;k<sizeof(seq)/sizeof(int);k++) { |
---|
1948 | int iSeq=seq[k]; |
---|
1949 | if (getColumnStatus(iSeq)!=basic) { |
---|
1950 | double djval; |
---|
1951 | double * work; |
---|
1952 | int number; |
---|
1953 | int * which; |
---|
1954 | |
---|
1955 | int iIndex; |
---|
1956 | unpack(rowArray_[1],iSeq); |
---|
1957 | factorization_->updateColumn(rowArray_[2],rowArray_[1]); |
---|
1958 | djval = cost_[iSeq]; |
---|
1959 | work=rowArray_[1]->denseVector(); |
---|
1960 | number=rowArray_[1]->getNumElements(); |
---|
1961 | which=rowArray_[1]->getIndices(); |
---|
1962 | |
---|
1963 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
1964 | |
---|
1965 | int iRow = which[iIndex]; |
---|
1966 | double alpha = work[iRow]; |
---|
1967 | int iPivot=pivotVariable_[iRow]; |
---|
1968 | djval -= alpha*cost_[iPivot]; |
---|
1969 | } |
---|
1970 | double comp = 1.0e-8 + 1.0e-7*(max(fabs(dj_[iSeq]),fabs(djval))); |
---|
1971 | if (fabs(djval-dj_[iSeq])>comp) |
---|
1972 | printf("Bad dj %g for %d - true is %g\n", |
---|
1973 | dj_[iSeq],iSeq,djval); |
---|
1974 | assert (fabs(djval)<1.0e-3||djval*dj_[iSeq]>0.0); |
---|
1975 | rowArray_[1]->clear(); |
---|
1976 | } |
---|
1977 | } |
---|
1978 | } |
---|
1979 | #endif |
---|
1980 | |
---|
1981 | // we found a pivot column |
---|
1982 | // update the incoming column |
---|
1983 | unpackPacked(rowArray_[1]); |
---|
1984 | // save reduced cost |
---|
1985 | double saveDj = dualIn_; |
---|
1986 | factorization_->updateColumnFT(rowArray_[2],rowArray_[1]); |
---|
1987 | // do ratio test and re-compute dj |
---|
1988 | primalRow(rowArray_[1],rowArray_[3],rowArray_[2],rowArray_[0], |
---|
1989 | ifValuesPass); |
---|
1990 | if (ifValuesPass) { |
---|
1991 | saveDj=dualIn_; |
---|
1992 | if (pivotRow_==-1||(pivotRow_>=0&&fabs(alpha_)<1.0e-5)) { |
---|
1993 | if(fabs(dualIn_)<1.0e2*dualTolerance_) { |
---|
1994 | // try other way |
---|
1995 | directionIn_=-directionIn_; |
---|
1996 | primalRow(rowArray_[1],rowArray_[3],rowArray_[2],rowArray_[0], |
---|
1997 | 0); |
---|
1998 | } |
---|
1999 | if (pivotRow_==-1||(pivotRow_>=0&&fabs(alpha_)<1.0e-5)) { |
---|
2000 | if (solveType_==1) { |
---|
2001 | // reject it |
---|
2002 | char x = isColumn(sequenceIn_) ? 'C' :'R'; |
---|
2003 | handler_->message(CLP_SIMPLEX_FLAG,messages_) |
---|
2004 | <<x<<sequenceWithin(sequenceIn_) |
---|
2005 | <<CoinMessageEol; |
---|
2006 | setFlagged(sequenceIn_); |
---|
2007 | progress_->clearBadTimes(); |
---|
2008 | lastBadIteration_ = numberIterations_; // say be more cautious |
---|
2009 | clearAll(); |
---|
2010 | pivotRow_=-1; |
---|
2011 | } |
---|
2012 | returnCode=-5; |
---|
2013 | break; |
---|
2014 | } |
---|
2015 | } |
---|
2016 | } |
---|
2017 | double checkValue=1.0e-2; |
---|
2018 | if (largestDualError_>1.0e-5) |
---|
2019 | checkValue=1.0e-1; |
---|
2020 | if (solveType_==1&&((saveDj*dualIn_<1.0e-20&&!ifValuesPass)|| |
---|
2021 | fabs(saveDj-dualIn_)>checkValue*(1.0+fabs(saveDj)))) { |
---|
2022 | handler_->message(CLP_PRIMAL_DJ,messages_) |
---|
2023 | <<saveDj<<dualIn_ |
---|
2024 | <<CoinMessageEol; |
---|
2025 | if(lastGoodIteration_ != numberIterations_) { |
---|
2026 | clearAll(); |
---|
2027 | pivotRow_=-1; // say no weights update |
---|
2028 | returnCode=-4; |
---|
2029 | if(lastGoodIteration_+1 == numberIterations_) { |
---|
2030 | // not looking wonderful - try cleaning bounds |
---|
2031 | // put non-basics to bounds in case tolerance moved |
---|
2032 | nonLinearCost_->checkInfeasibilities(0.0); |
---|
2033 | } |
---|
2034 | sequenceOut_=-1; |
---|
2035 | break; |
---|
2036 | } else { |
---|
2037 | // take on more relaxed criterion |
---|
2038 | if (saveDj*dualIn_<1.0e-20|| |
---|
2039 | fabs(saveDj-dualIn_)>2.0e-1*(1.0+fabs(dualIn_))) { |
---|
2040 | // need to reject something |
---|
2041 | char x = isColumn(sequenceIn_) ? 'C' :'R'; |
---|
2042 | handler_->message(CLP_SIMPLEX_FLAG,messages_) |
---|
2043 | <<x<<sequenceWithin(sequenceIn_) |
---|
2044 | <<CoinMessageEol; |
---|
2045 | setFlagged(sequenceIn_); |
---|
2046 | progress_->clearBadTimes(); |
---|
2047 | lastBadIteration_ = numberIterations_; // say be more cautious |
---|
2048 | clearAll(); |
---|
2049 | pivotRow_=-1; |
---|
2050 | returnCode=-5; |
---|
2051 | sequenceOut_=-1; |
---|
2052 | break; |
---|
2053 | } |
---|
2054 | } |
---|
2055 | } |
---|
2056 | if (pivotRow_>=0) { |
---|
2057 | if (solveType_==2) { |
---|
2058 | // **** Coding for user interface |
---|
2059 | // do ray |
---|
2060 | primalRay(rowArray_[1]); |
---|
2061 | // update duals |
---|
2062 | if (pivotRow_>=0) { |
---|
2063 | // as packed need to find pivot row |
---|
2064 | //assert (rowArray_[1]->packedMode()); |
---|
2065 | //int i; |
---|
2066 | |
---|
2067 | //alpha_ = rowArray_[1]->denseVector()[pivotRow_]; |
---|
2068 | assert (fabs(alpha_)>1.0e-8); |
---|
2069 | double multiplier = dualIn_/alpha_; |
---|
2070 | rowArray_[0]->insert(pivotRow_,multiplier); |
---|
2071 | factorization_->updateColumnTranspose(rowArray_[2],rowArray_[0]); |
---|
2072 | // put row of tableau in rowArray[0] and columnArray[0] |
---|
2073 | matrix_->transposeTimes(this,-1.0, |
---|
2074 | rowArray_[0],columnArray_[1],columnArray_[0]); |
---|
2075 | // update column djs |
---|
2076 | int i; |
---|
2077 | int * index = columnArray_[0]->getIndices(); |
---|
2078 | int number = columnArray_[0]->getNumElements(); |
---|
2079 | double * element = columnArray_[0]->denseVector(); |
---|
2080 | for (i=0;i<number;i++) { |
---|
2081 | int ii = index[i]; |
---|
2082 | dj_[ii] += element[ii]; |
---|
2083 | element[ii]=0.0; |
---|
2084 | } |
---|
2085 | columnArray_[0]->setNumElements(0); |
---|
2086 | // and row djs |
---|
2087 | index = rowArray_[0]->getIndices(); |
---|
2088 | number = rowArray_[0]->getNumElements(); |
---|
2089 | element = rowArray_[0]->denseVector(); |
---|
2090 | for (i=0;i<number;i++) { |
---|
2091 | int ii = index[i]; |
---|
2092 | dj_[ii+numberColumns_] += element[ii]; |
---|
2093 | dual_[ii] = dj_[ii+numberColumns_]; |
---|
2094 | element[ii]=0.0; |
---|
2095 | } |
---|
2096 | rowArray_[0]->setNumElements(0); |
---|
2097 | // check incoming |
---|
2098 | assert (fabs(dj_[sequenceIn_])<1.0e-6); |
---|
2099 | } |
---|
2100 | } |
---|
2101 | // if stable replace in basis |
---|
2102 | int updateStatus = factorization_->replaceColumn(rowArray_[2], |
---|
2103 | pivotRow_, |
---|
2104 | alpha_); |
---|
2105 | // if no pivots, bad update but reasonable alpha - take and invert |
---|
2106 | if (updateStatus==2&& |
---|
2107 | lastGoodIteration_==numberIterations_&&fabs(alpha_)>1.0e-5) |
---|
2108 | updateStatus=4; |
---|
2109 | if (updateStatus==1||updateStatus==4) { |
---|
2110 | // slight error |
---|
2111 | if (factorization_->pivots()>5||updateStatus==4) { |
---|
2112 | returnCode=-3; |
---|
2113 | } |
---|
2114 | } else if (updateStatus==2) { |
---|
2115 | // major error |
---|
2116 | // better to have small tolerance even if slower |
---|
2117 | factorization_->zeroTolerance(1.0e-15); |
---|
2118 | int maxFactor = factorization_->maximumPivots(); |
---|
2119 | if (maxFactor>10) { |
---|
2120 | if (forceFactorization_<0) |
---|
2121 | forceFactorization_= maxFactor; |
---|
2122 | forceFactorization_ = max (1,(forceFactorization_>>1)); |
---|
2123 | } |
---|
2124 | // later we may need to unwind more e.g. fake bounds |
---|
2125 | if(lastGoodIteration_ != numberIterations_) { |
---|
2126 | clearAll(); |
---|
2127 | pivotRow_=-1; |
---|
2128 | if (solveType_==1) { |
---|
2129 | returnCode=-4; |
---|
2130 | break; |
---|
2131 | } else { |
---|
2132 | // user in charge - re-factorize |
---|
2133 | int lastCleaned; |
---|
2134 | ClpSimplexProgress dummyProgress; |
---|
2135 | if (saveStatus_) |
---|
2136 | statusOfProblemInPrimal(lastCleaned,1,&dummyProgress,true); |
---|
2137 | else |
---|
2138 | statusOfProblemInPrimal(lastCleaned,0,&dummyProgress,true); |
---|
2139 | roundAgain=true; |
---|
2140 | continue; |
---|
2141 | } |
---|
2142 | } else { |
---|
2143 | // need to reject something |
---|
2144 | if (solveType_==1) { |
---|
2145 | char x = isColumn(sequenceIn_) ? 'C' :'R'; |
---|
2146 | handler_->message(CLP_SIMPLEX_FLAG,messages_) |
---|
2147 | <<x<<sequenceWithin(sequenceIn_) |
---|
2148 | <<CoinMessageEol; |
---|
2149 | setFlagged(sequenceIn_); |
---|
2150 | progress_->clearBadTimes(); |
---|
2151 | } |
---|
2152 | lastBadIteration_ = numberIterations_; // say be more cautious |
---|
2153 | clearAll(); |
---|
2154 | pivotRow_=-1; |
---|
2155 | sequenceOut_=-1; |
---|
2156 | returnCode = -5; |
---|
2157 | break; |
---|
2158 | |
---|
2159 | } |
---|
2160 | } else if (updateStatus==3) { |
---|
2161 | // out of memory |
---|
2162 | // increase space if not many iterations |
---|
2163 | if (factorization_->pivots()< |
---|
2164 | 0.5*factorization_->maximumPivots()&& |
---|
2165 | factorization_->pivots()<200) |
---|
2166 | factorization_->areaFactor( |
---|
2167 | factorization_->areaFactor() * 1.1); |
---|
2168 | returnCode =-2; // factorize now |
---|
2169 | } else if (updateStatus==5) { |
---|
2170 | problemStatus_=-2; // factorize now |
---|
2171 | } |
---|
2172 | // here do part of steepest - ready for next iteration |
---|
2173 | if (!ifValuesPass) |
---|
2174 | primalColumnPivot_->updateWeights(rowArray_[1]); |
---|
2175 | } else { |
---|
2176 | if (pivotRow_==-1) { |
---|
2177 | // no outgoing row is valid |
---|
2178 | rowArray_[0]->clear(); |
---|
2179 | if (!factorization_->pivots()) { |
---|
2180 | returnCode = 2; //say looks unbounded |
---|
2181 | // do ray |
---|
2182 | primalRay(rowArray_[1]); |
---|
2183 | } else if (solveType_==2) { |
---|
2184 | // refactorize |
---|
2185 | int lastCleaned; |
---|
2186 | ClpSimplexProgress dummyProgress; |
---|
2187 | if (saveStatus_) |
---|
2188 | statusOfProblemInPrimal(lastCleaned,1,&dummyProgress,true); |
---|
2189 | else |
---|
2190 | statusOfProblemInPrimal(lastCleaned,0,&dummyProgress,true); |
---|
2191 | roundAgain=true; |
---|
2192 | continue; |
---|
2193 | } else { |
---|
2194 | returnCode = 4; //say looks unbounded but has iterated |
---|
2195 | } |
---|
2196 | break; |
---|
2197 | } else { |
---|
2198 | // flipping from bound to bound |
---|
2199 | } |
---|
2200 | } |
---|
2201 | |
---|
2202 | |
---|
2203 | // update primal solution |
---|
2204 | |
---|
2205 | double objectiveChange=0.0; |
---|
2206 | // after this rowArray_[1] is not empty - used to update djs |
---|
2207 | updatePrimalsInPrimal(rowArray_[1],theta_, objectiveChange,ifValuesPass); |
---|
2208 | |
---|
2209 | double oldValue = valueIn_; |
---|
2210 | if (directionIn_==-1) { |
---|
2211 | // as if from upper bound |
---|
2212 | if (sequenceIn_!=sequenceOut_) { |
---|
2213 | // variable becoming basic |
---|
2214 | valueIn_ -= fabs(theta_); |
---|
2215 | } else { |
---|
2216 | valueIn_=lowerIn_; |
---|
2217 | } |
---|
2218 | } else { |
---|
2219 | // as if from lower bound |
---|
2220 | if (sequenceIn_!=sequenceOut_) { |
---|
2221 | // variable becoming basic |
---|
2222 | valueIn_ += fabs(theta_); |
---|
2223 | } else { |
---|
2224 | valueIn_=upperIn_; |
---|
2225 | } |
---|
2226 | } |
---|
2227 | objectiveChange += dualIn_*(valueIn_-oldValue); |
---|
2228 | // outgoing |
---|
2229 | if (sequenceIn_!=sequenceOut_) { |
---|
2230 | if (directionOut_>0) { |
---|
2231 | valueOut_ = lowerOut_; |
---|
2232 | } else { |
---|
2233 | valueOut_ = upperOut_; |
---|
2234 | } |
---|
2235 | if(valueOut_<lower_[sequenceOut_]-primalTolerance_) |
---|
2236 | valueOut_=lower_[sequenceOut_]-0.9*primalTolerance_; |
---|
2237 | else if (valueOut_>upper_[sequenceOut_]+primalTolerance_) |
---|
2238 | valueOut_=upper_[sequenceOut_]+0.9*primalTolerance_; |
---|
2239 | // may not be exactly at bound and bounds may have changed |
---|
2240 | // Make sure outgoing looks feasible |
---|
2241 | directionOut_=nonLinearCost_->setOneOutgoing(sequenceOut_,valueOut_); |
---|
2242 | solution_[sequenceOut_]=valueOut_; |
---|
2243 | } |
---|
2244 | // change cost and bounds on incoming if primal |
---|
2245 | nonLinearCost_->setOne(sequenceIn_,valueIn_); |
---|
2246 | int whatNext=housekeeping(objectiveChange); |
---|
2247 | |
---|
2248 | #if 0 |
---|
2249 | if (numberIterations_==1148) |
---|
2250 | whatNext=1; |
---|
2251 | if (numberIterations_>1200) |
---|
2252 | exit(0); |
---|
2253 | #endif |
---|
2254 | if (whatNext==1) { |
---|
2255 | returnCode =-2; // refactorize |
---|
2256 | } else if (whatNext==2) { |
---|
2257 | // maximum iterations or equivalent |
---|
2258 | returnCode=3; |
---|
2259 | } else if(numberIterations_ == lastGoodIteration_ |
---|
2260 | + 2 * factorization_->maximumPivots()) { |
---|
2261 | // done a lot of flips - be safe |
---|
2262 | returnCode =-2; // refactorize |
---|
2263 | } |
---|
2264 | } |
---|
2265 | if (solveType_==2&&(returnCode == -2||returnCode==-3)) { |
---|
2266 | // refactorize here |
---|
2267 | int lastCleaned; |
---|
2268 | ClpSimplexProgress dummyProgress; |
---|
2269 | if (saveStatus_) |
---|
2270 | statusOfProblemInPrimal(lastCleaned,1,&dummyProgress,true); |
---|
2271 | else |
---|
2272 | statusOfProblemInPrimal(lastCleaned,0,&dummyProgress,true); |
---|
2273 | if (problemStatus_==5) { |
---|
2274 | printf("Singular basis\n"); |
---|
2275 | problemStatus_=-1; |
---|
2276 | returnCode=5; |
---|
2277 | } |
---|
2278 | } |
---|
2279 | #ifdef CLP_DEBUG |
---|
2280 | { |
---|
2281 | int i; |
---|
2282 | // not [1] as may have information |
---|
2283 | for (i=0;i<4;i++) { |
---|
2284 | if (i!=1) |
---|
2285 | rowArray_[i]->checkClear(); |
---|
2286 | } |
---|
2287 | for (i=0;i<2;i++) { |
---|
2288 | columnArray_[i]->checkClear(); |
---|
2289 | } |
---|
2290 | } |
---|
2291 | #endif |
---|
2292 | return returnCode; |
---|
2293 | } |
---|
2294 | // Create primal ray |
---|
2295 | void |
---|
2296 | ClpSimplexPrimal::primalRay(CoinIndexedVector * rowArray) |
---|
2297 | { |
---|
2298 | delete [] ray_; |
---|
2299 | ray_ = new double [numberColumns_]; |
---|
2300 | ClpFillN(ray_,numberColumns_,0.0); |
---|
2301 | int number=rowArray->getNumElements(); |
---|
2302 | int * index = rowArray->getIndices(); |
---|
2303 | double * array = rowArray->denseVector(); |
---|
2304 | double way=-directionIn_; |
---|
2305 | int i; |
---|
2306 | double zeroTolerance=1.0e-12; |
---|
2307 | if (sequenceIn_<numberColumns_) |
---|
2308 | ray_[sequenceIn_]=directionIn_; |
---|
2309 | if (!rowArray->packedMode()) { |
---|
2310 | for (i=0;i<number;i++) { |
---|
2311 | int iRow=index[i]; |
---|
2312 | int iPivot=pivotVariable_[iRow]; |
---|
2313 | double arrayValue = array[iRow]; |
---|
2314 | if (iPivot<numberColumns_&&fabs(arrayValue)>=zeroTolerance) |
---|
2315 | ray_[iPivot] = way* arrayValue; |
---|
2316 | } |
---|
2317 | } else { |
---|
2318 | for (i=0;i<number;i++) { |
---|
2319 | int iRow=index[i]; |
---|
2320 | int iPivot=pivotVariable_[iRow]; |
---|
2321 | double arrayValue = array[i]; |
---|
2322 | if (iPivot<numberColumns_&&fabs(arrayValue)>=zeroTolerance) |
---|
2323 | ray_[iPivot] = way* arrayValue; |
---|
2324 | } |
---|
2325 | } |
---|
2326 | } |
---|
2327 | /* Get next superbasic -1 if none, |
---|
2328 | Normal type is 1 |
---|
2329 | If type is 3 then initializes sorted list |
---|
2330 | if 2 uses list. |
---|
2331 | */ |
---|
2332 | int |
---|
2333 | ClpSimplexPrimal::nextSuperBasic(int superBasicType,CoinIndexedVector * columnArray) |
---|
2334 | { |
---|
2335 | if (firstFree_>=0&&superBasicType) { |
---|
2336 | int returnValue=firstFree_; |
---|
2337 | int iColumn=firstFree_+1; |
---|
2338 | if (superBasicType>1) { |
---|
2339 | if (superBasicType>2) { |
---|
2340 | // Initialize list |
---|
2341 | // Wild guess that lower bound more natural than upper |
---|
2342 | int number=0; |
---|
2343 | double * work=columnArray->denseVector(); |
---|
2344 | int * which=columnArray->getIndices(); |
---|
2345 | for (iColumn=0;iColumn<numberRows_+numberColumns_;iColumn++) { |
---|
2346 | if (getStatus(iColumn)==superBasic) { |
---|
2347 | if (fabs(solution_[iColumn]-lower_[iColumn])<=primalTolerance_) { |
---|
2348 | solution_[iColumn]=lower_[iColumn]; |
---|
2349 | setStatus(iColumn,atLowerBound); |
---|
2350 | } else if (fabs(solution_[iColumn]-upper_[iColumn]) |
---|
2351 | <=primalTolerance_) { |
---|
2352 | solution_[iColumn]=upper_[iColumn]; |
---|
2353 | setStatus(iColumn,atUpperBound); |
---|
2354 | } else if (lower_[iColumn]<-1.0e20&&upper_[iColumn]>1.0e20) { |
---|
2355 | setStatus(iColumn,isFree); |
---|
2356 | break; |
---|
2357 | } else if (!flagged(iColumn)) { |
---|
2358 | // put ones near bounds at end after sorting |
---|
2359 | work[number]= - min(0.1*(solution_[iColumn]-lower_[iColumn]), |
---|
2360 | upper_[iColumn]-solution_[iColumn]); |
---|
2361 | which[number++] = iColumn; |
---|
2362 | } |
---|
2363 | } |
---|
2364 | } |
---|
2365 | CoinSort_2(work,work+number,which); |
---|
2366 | columnArray->setNumElements(number); |
---|
2367 | memset(work,0,number*sizeof(double)); |
---|
2368 | } |
---|
2369 | int * which=columnArray->getIndices(); |
---|
2370 | int number = columnArray->getNumElements(); |
---|
2371 | if (!number) { |
---|
2372 | // finished |
---|
2373 | iColumn = numberRows_+numberColumns_; |
---|
2374 | returnValue=-1; |
---|
2375 | } else { |
---|
2376 | number--; |
---|
2377 | returnValue=which[number]; |
---|
2378 | iColumn=returnValue; |
---|
2379 | columnArray->setNumElements(number); |
---|
2380 | } |
---|
2381 | } else { |
---|
2382 | for (;iColumn<numberRows_+numberColumns_;iColumn++) { |
---|
2383 | if (getStatus(iColumn)==superBasic) { |
---|
2384 | if (fabs(solution_[iColumn]-lower_[iColumn])<=primalTolerance_) { |
---|
2385 | solution_[iColumn]=lower_[iColumn]; |
---|
2386 | setStatus(iColumn,atLowerBound); |
---|
2387 | } else if (fabs(solution_[iColumn]-upper_[iColumn]) |
---|
2388 | <=primalTolerance_) { |
---|
2389 | solution_[iColumn]=upper_[iColumn]; |
---|
2390 | setStatus(iColumn,atUpperBound); |
---|
2391 | } else if (lower_[iColumn]<-1.0e20&&upper_[iColumn]>1.0e20) { |
---|
2392 | setStatus(iColumn,isFree); |
---|
2393 | break; |
---|
2394 | } else { |
---|
2395 | break; |
---|
2396 | } |
---|
2397 | } |
---|
2398 | } |
---|
2399 | } |
---|
2400 | firstFree_ = iColumn; |
---|
2401 | if (firstFree_==numberRows_+numberColumns_) |
---|
2402 | firstFree_=-1; |
---|
2403 | return returnValue; |
---|
2404 | } else { |
---|
2405 | return -1; |
---|
2406 | } |
---|
2407 | } |
---|
2408 | void |
---|
2409 | ClpSimplexPrimal::clearAll() |
---|
2410 | { |
---|
2411 | int number=rowArray_[1]->getNumElements(); |
---|
2412 | int * which=rowArray_[1]->getIndices(); |
---|
2413 | |
---|
2414 | int iIndex; |
---|
2415 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
2416 | |
---|
2417 | int iRow = which[iIndex]; |
---|
2418 | clearActive(iRow); |
---|
2419 | } |
---|
2420 | rowArray_[1]->clear(); |
---|
2421 | } |
---|
2422 | |
---|