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 costs 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 | #if defined(_MSC_VER) |
---|
80 | // Turn off compiler warning about long names |
---|
81 | # pragma warning(disable:4786) |
---|
82 | #endif |
---|
83 | |
---|
84 | #include <math.h> |
---|
85 | |
---|
86 | #include "CoinHelperFunctions.hpp" |
---|
87 | #include "ClpSimplexPrimal.hpp" |
---|
88 | #include "ClpFactorization.hpp" |
---|
89 | #include "ClpNonLinearCost.hpp" |
---|
90 | #include "CoinPackedMatrix.hpp" |
---|
91 | #include "CoinIndexedVector.hpp" |
---|
92 | #include "CoinWarmStartBasis.hpp" |
---|
93 | #include "ClpPrimalColumnPivot.hpp" |
---|
94 | #include "ClpMessage.hpp" |
---|
95 | #include <cfloat> |
---|
96 | #include <cassert> |
---|
97 | #include <string> |
---|
98 | #include <stdio.h> |
---|
99 | #include <iostream> |
---|
100 | // primal |
---|
101 | int ClpSimplexPrimal::primal (int ifValuesPass ) |
---|
102 | { |
---|
103 | |
---|
104 | /* |
---|
105 | Method |
---|
106 | |
---|
107 | It tries to be a single phase approach with a weight of 1.0 being |
---|
108 | given to getting optimal and a weight of infeasibilityCost_ being |
---|
109 | given to getting primal feasible. In this version I have tried to |
---|
110 | be clever in a stupid way. The idea of fake bounds in dual |
---|
111 | seems to work so the primal analogue would be that of getting |
---|
112 | bounds on reduced costs (by a presolve approach) and using |
---|
113 | these for being above or below feasible region. I decided to waste |
---|
114 | memory and keep these explicitly. This allows for non-linear |
---|
115 | costs! |
---|
116 | |
---|
117 | The code is designed to take advantage of sparsity so arrays are |
---|
118 | seldom zeroed out from scratch or gone over in their entirety. |
---|
119 | The only exception is a full scan to find incoming variable for |
---|
120 | Dantzig row choice. For steepest edge we keep an updated list |
---|
121 | of dual infeasibilities (actually squares). |
---|
122 | On easy problems we don't need full scan - just |
---|
123 | pick first reasonable. |
---|
124 | |
---|
125 | One problem is how to tackle degeneracy and accuracy. At present |
---|
126 | I am using the modification of costs which I put in OSL and which was |
---|
127 | extended by Gill et al. I am still not sure of the exact details. |
---|
128 | |
---|
129 | The flow of primal is three while loops as follows: |
---|
130 | |
---|
131 | while (not finished) { |
---|
132 | |
---|
133 | while (not clean solution) { |
---|
134 | |
---|
135 | Factorize and/or clean up solution by changing bounds so |
---|
136 | primal feasible. If looks finished check fake primal bounds. |
---|
137 | Repeat until status is iterating (-1) or finished (0,1,2) |
---|
138 | |
---|
139 | } |
---|
140 | |
---|
141 | while (status==-1) { |
---|
142 | |
---|
143 | Iterate until no pivot in or out or time to re-factorize. |
---|
144 | |
---|
145 | Flow is: |
---|
146 | |
---|
147 | choose pivot column (incoming variable). if none then |
---|
148 | we are primal feasible so looks as if done but we need to |
---|
149 | break and check bounds etc. |
---|
150 | |
---|
151 | Get pivot column in tableau |
---|
152 | |
---|
153 | Choose outgoing row. If we don't find one then we look |
---|
154 | primal unbounded so break and check bounds etc. (Also the |
---|
155 | pivot tolerance is larger after any iterations so that may be |
---|
156 | reason) |
---|
157 | |
---|
158 | If we do find outgoing row, we may have to adjust costs to |
---|
159 | keep going forwards (anti-degeneracy). Check pivot will be stable |
---|
160 | and if unstable throw away iteration and break to re-factorize. |
---|
161 | If minor error re-factorize after iteration. |
---|
162 | |
---|
163 | Update everything (this may involve changing bounds on |
---|
164 | variables to stay primal feasible. |
---|
165 | |
---|
166 | } |
---|
167 | |
---|
168 | } |
---|
169 | |
---|
170 | At present we never check we are going forwards. I overdid that in |
---|
171 | OSL so will try and make a last resort. |
---|
172 | |
---|
173 | Needs partial scan pivot in option. |
---|
174 | |
---|
175 | May need other anti-degeneracy measures, especially if we try and use |
---|
176 | loose tolerances as a way to solve in fewer iterations. |
---|
177 | |
---|
178 | I like idea of dynamic scaling. This gives opportunity to decouple |
---|
179 | different implications of scaling for accuracy, iteration count and |
---|
180 | feasibility tolerance. |
---|
181 | |
---|
182 | */ |
---|
183 | |
---|
184 | // sanity check |
---|
185 | assert (numberRows_==matrix_->getNumRows()); |
---|
186 | assert (numberColumns_==matrix_->getNumCols()); |
---|
187 | // for moment all arrays must exist |
---|
188 | assert(columnLower_); |
---|
189 | assert(columnUpper_); |
---|
190 | assert(rowLower_); |
---|
191 | assert(rowUpper_); |
---|
192 | |
---|
193 | |
---|
194 | algorithm_ = +1; |
---|
195 | primalTolerance_=dblParam_[ClpPrimalTolerance]; |
---|
196 | dualTolerance_=dblParam_[ClpDualTolerance]; |
---|
197 | |
---|
198 | // put in standard form (and make row copy) |
---|
199 | // create modifiable copies of model rim and do optional scaling |
---|
200 | bool goodMatrix=createRim(7+8+16,true); |
---|
201 | |
---|
202 | // save infeasibility cost |
---|
203 | double saveInfeasibilityCost = infeasibilityCost_; |
---|
204 | // Save perturbation |
---|
205 | int savePerturbation = perturbation_; |
---|
206 | // save if sparse factorization wanted |
---|
207 | int saveSparse = factorization_->sparseThreshold(); |
---|
208 | |
---|
209 | if (goodMatrix) { |
---|
210 | // Model looks okay |
---|
211 | int iRow,iColumn; |
---|
212 | // Do initial factorization |
---|
213 | // and set certain stuff |
---|
214 | // We can either set increasing rows so ...IsBasic gives pivot row |
---|
215 | // or we can just increment iBasic one by one |
---|
216 | // for now let ...iBasic give pivot row |
---|
217 | factorization_->increasingRows(2); |
---|
218 | // row activities have negative sign |
---|
219 | factorization_->slackValue(-1.0); |
---|
220 | factorization_->zeroTolerance(1.0e-13); |
---|
221 | |
---|
222 | |
---|
223 | // If user asked for perturbation - do it |
---|
224 | |
---|
225 | if (perturbation_<100) |
---|
226 | perturb(); |
---|
227 | |
---|
228 | // for primal we will change bounds using infeasibilityCost_ |
---|
229 | if (nonLinearCost_==NULL) { |
---|
230 | // get a valid nonlinear cost function |
---|
231 | delete nonLinearCost_; |
---|
232 | nonLinearCost_= new ClpNonLinearCost(this); |
---|
233 | } |
---|
234 | |
---|
235 | // loop round to clean up solution if values pass |
---|
236 | int numberThrownOut = -1; |
---|
237 | int firstSuperBasic=numberRows_+numberColumns_; |
---|
238 | int totalNumberThrownOut=0; |
---|
239 | while(numberThrownOut) { |
---|
240 | int status = internalFactorize(0+10*ifValuesPass); |
---|
241 | if (status<0) |
---|
242 | return 1; // some error |
---|
243 | else |
---|
244 | totalNumberThrownOut+= status; |
---|
245 | |
---|
246 | // for this we need clean basis so it is after factorize |
---|
247 | numberThrownOut=gutsOfSolution(rowActivityWork_,columnActivityWork_, |
---|
248 | ifValuesPass); |
---|
249 | totalNumberThrownOut+= numberThrownOut; |
---|
250 | |
---|
251 | // find first superbasic - columns, then rows |
---|
252 | if (ifValuesPass) { |
---|
253 | nextSuperBasic(firstSuperBasic); |
---|
254 | if (firstSuperBasic==numberRows_+numberColumns_) |
---|
255 | ifValuesPass=0; // signal no values pass |
---|
256 | } |
---|
257 | } |
---|
258 | |
---|
259 | if (totalNumberThrownOut) |
---|
260 | handler_->message(CLP_SINGULARITIES,messages_) |
---|
261 | <<totalNumberThrownOut |
---|
262 | <<CoinMessageEol; |
---|
263 | |
---|
264 | problemStatus_ = -1; |
---|
265 | numberIterations_=0; |
---|
266 | |
---|
267 | int lastCleaned=0; // last time objective or bounds cleaned up |
---|
268 | |
---|
269 | // number of times we have declared optimality |
---|
270 | numberTimesOptimal_=0; |
---|
271 | |
---|
272 | // Progress indicator |
---|
273 | ClpSimplexProgress progress(this); |
---|
274 | |
---|
275 | // Say no pivot has occurred (for steepest edge and updates) |
---|
276 | pivotRow_=-2; |
---|
277 | |
---|
278 | // This says whether to restore things etc |
---|
279 | int factorType=0; |
---|
280 | // Save iteration number |
---|
281 | int saveNumber = -1; |
---|
282 | /* |
---|
283 | Status of problem: |
---|
284 | 0 - optimal |
---|
285 | 1 - infeasible |
---|
286 | 2 - unbounded |
---|
287 | -1 - iterating |
---|
288 | -2 - factorization wanted |
---|
289 | -3 - redo checking without factorization |
---|
290 | -4 - looks infeasible |
---|
291 | -5 - looks unbounded |
---|
292 | */ |
---|
293 | while (problemStatus_<0) { |
---|
294 | // clear |
---|
295 | for (iRow=0;iRow<4;iRow++) { |
---|
296 | rowArray_[iRow]->clear(); |
---|
297 | } |
---|
298 | |
---|
299 | for (iColumn=0;iColumn<2;iColumn++) { |
---|
300 | columnArray_[iColumn]->clear(); |
---|
301 | } |
---|
302 | |
---|
303 | // give matrix (and model costs and bounds a chance to be |
---|
304 | // refreshed (normally null) |
---|
305 | matrix_->refresh(this); |
---|
306 | // If getting nowhere - why not give it a kick |
---|
307 | #if 0 |
---|
308 | // primal perturbation not coded yet |
---|
309 | if (perturbation_<101&&numberIterations_>2*(numberRows_+numberColumns_)) |
---|
310 | perturb(); |
---|
311 | #endif |
---|
312 | // If we have done no iterations - special |
---|
313 | if (saveNumber==numberIterations_) |
---|
314 | factorType=3; |
---|
315 | // may factorize, checks if problem finished |
---|
316 | statusOfProblemInPrimal(lastCleaned,factorType,progress); |
---|
317 | |
---|
318 | // Say good factorization |
---|
319 | factorType=1; |
---|
320 | if (saveSparse) { |
---|
321 | // use default at present |
---|
322 | factorization_->sparseThreshold(0); |
---|
323 | factorization_->goSparse(); |
---|
324 | } |
---|
325 | |
---|
326 | // Say no pivot has occurred (for steepest edge and updates) |
---|
327 | pivotRow_=-2; |
---|
328 | |
---|
329 | // Save iteration number |
---|
330 | saveNumber = numberIterations_; |
---|
331 | |
---|
332 | // Iterate |
---|
333 | whileIterating(firstSuperBasic); |
---|
334 | } |
---|
335 | } |
---|
336 | // if infeasible get real values |
---|
337 | if (problemStatus_==1) { |
---|
338 | infeasibilityCost_=0.0; |
---|
339 | createRim(7); |
---|
340 | nonLinearCost_->checkInfeasibilities(true); |
---|
341 | sumPrimalInfeasibilities_=nonLinearCost_->sumInfeasibilities(); |
---|
342 | numberPrimalInfeasibilities_= nonLinearCost_->numberInfeasibilities(); |
---|
343 | // and get good feasible duals |
---|
344 | computeDuals(); |
---|
345 | } |
---|
346 | factorization_->sparseThreshold(saveSparse); |
---|
347 | // Get rid of some arrays and empty factorization |
---|
348 | deleteRim(); |
---|
349 | handler_->message(CLP_SIMPLEX_FINISHED+problemStatus_,messages_) |
---|
350 | <<objectiveValue() |
---|
351 | <<CoinMessageEol; |
---|
352 | // Restore any saved stuff |
---|
353 | perturbation_ = savePerturbation; |
---|
354 | infeasibilityCost_ = saveInfeasibilityCost; |
---|
355 | return problemStatus_; |
---|
356 | } |
---|
357 | /* |
---|
358 | Reasons to come out: |
---|
359 | -1 iterations etc |
---|
360 | -2 inaccuracy |
---|
361 | -3 slight inaccuracy (and done iterations) |
---|
362 | -4 end of values pass and done iterations |
---|
363 | +0 looks optimal (might be infeasible - but we will investigate) |
---|
364 | +2 looks unbounded |
---|
365 | +3 max iterations |
---|
366 | */ |
---|
367 | int |
---|
368 | ClpSimplexPrimal::whileIterating(int & firstSuperBasic) |
---|
369 | { |
---|
370 | |
---|
371 | // Say if values pass |
---|
372 | int ifValuesPass=0; |
---|
373 | int returnCode=-1; |
---|
374 | int startIteration = numberIterations_; |
---|
375 | if (firstSuperBasic<numberRows_+numberColumns_) |
---|
376 | ifValuesPass=1; |
---|
377 | int saveNumber = numberIterations_; |
---|
378 | // status stays at -1 while iterating, >=0 finished, -2 to invert |
---|
379 | // status -3 to go to top without an invert |
---|
380 | while (problemStatus_==-1) { |
---|
381 | #ifdef CLP_DEBUG |
---|
382 | { |
---|
383 | int i; |
---|
384 | // not [1] as has information |
---|
385 | for (i=0;i<4;i++) { |
---|
386 | if (i!=1) |
---|
387 | rowArray_[i]->checkClear(); |
---|
388 | } |
---|
389 | for (i=0;i<2;i++) { |
---|
390 | columnArray_[i]->checkClear(); |
---|
391 | } |
---|
392 | } |
---|
393 | #endif |
---|
394 | #if CLP_DEBUG>2 |
---|
395 | // very expensive |
---|
396 | if (numberIterations_>0&&numberIterations_<-2534) { |
---|
397 | handler_->setLogLevel(63); |
---|
398 | double saveValue = objectiveValue_; |
---|
399 | double * saveRow1 = new double[numberRows_]; |
---|
400 | double * saveRow2 = new double[numberRows_]; |
---|
401 | memcpy(saveRow1,rowReducedCost_,numberRows_*sizeof(double)); |
---|
402 | memcpy(saveRow2,rowActivityWork_,numberRows_*sizeof(double)); |
---|
403 | double * saveColumn1 = new double[numberColumns_]; |
---|
404 | double * saveColumn2 = new double[numberColumns_]; |
---|
405 | memcpy(saveColumn1,reducedCostWork_,numberColumns_*sizeof(double)); |
---|
406 | memcpy(saveColumn2,columnActivityWork_,numberColumns_*sizeof(double)); |
---|
407 | createRim(7); |
---|
408 | gutsOfSolution(rowActivityWork_,columnActivityWork_); |
---|
409 | printf("xxx %d old obj %g, recomputed %g, sum primal inf %g\n", |
---|
410 | numberIterations_, |
---|
411 | saveValue,objectiveValue_,sumPrimalInfeasibilities_); |
---|
412 | memcpy(rowReducedCost_,saveRow1,numberRows_*sizeof(double)); |
---|
413 | memcpy(rowActivityWork_,saveRow2,numberRows_*sizeof(double)); |
---|
414 | memcpy(reducedCostWork_,saveColumn1,numberColumns_*sizeof(double)); |
---|
415 | memcpy(columnActivityWork_,saveColumn2,numberColumns_*sizeof(double)); |
---|
416 | delete [] saveRow1; |
---|
417 | delete [] saveRow2; |
---|
418 | delete [] saveColumn1; |
---|
419 | delete [] saveColumn2; |
---|
420 | objectiveValue_=saveValue; |
---|
421 | } |
---|
422 | #endif |
---|
423 | if (!ifValuesPass) { |
---|
424 | // choose column to come in |
---|
425 | // can use pivotRow_ to update weights |
---|
426 | // pass in list of cost changes so can do row updates (rowArray_[1]) |
---|
427 | // NOTE rowArray_[0] is used by computeDuals which is a |
---|
428 | // slow way of getting duals but might be used |
---|
429 | primalColumn(rowArray_[1],rowArray_[2],rowArray_[3], |
---|
430 | columnArray_[0],columnArray_[1]); |
---|
431 | } else { |
---|
432 | // in values pass |
---|
433 | if (ifValuesPass>0) { |
---|
434 | nextSuperBasic(firstSuperBasic); |
---|
435 | if (firstSuperBasic==numberRows_+numberColumns_) |
---|
436 | ifValuesPass=-1; // signal end of values pass after this |
---|
437 | } else { |
---|
438 | // end of values pass - initialize weights etc |
---|
439 | primalColumnPivot_->saveWeights(this,5); |
---|
440 | ifValuesPass=0; |
---|
441 | if(saveNumber != numberIterations_) { |
---|
442 | problemStatus_=-2; // factorize now |
---|
443 | pivotRow_=-1; // say no weights update |
---|
444 | returnCode=-4; |
---|
445 | break; |
---|
446 | } |
---|
447 | |
---|
448 | // and get variable |
---|
449 | primalColumn(rowArray_[1],rowArray_[2],rowArray_[3], |
---|
450 | columnArray_[0],columnArray_[1]); |
---|
451 | } |
---|
452 | } |
---|
453 | pivotRow_=-1; |
---|
454 | sequenceOut_=-1; |
---|
455 | rowArray_[1]->clear(); |
---|
456 | if (sequenceIn_>=0) { |
---|
457 | // we found a pivot column |
---|
458 | #ifdef CLP_DEBUG |
---|
459 | if ((handler_->logLevel()&32)) { |
---|
460 | char x = isColumn(sequenceIn_) ? 'C' :'R'; |
---|
461 | std::cout<<"pivot column "<< |
---|
462 | x<<sequenceWithin(sequenceIn_)<<std::endl; |
---|
463 | } |
---|
464 | #endif |
---|
465 | // update the incoming column |
---|
466 | unpack(rowArray_[1]); |
---|
467 | // save reduced cost |
---|
468 | double saveDj = dualIn_; |
---|
469 | factorization_->updateColumn(rowArray_[2],rowArray_[1],true); |
---|
470 | // do ratio test and re-compute dj |
---|
471 | primalRow(rowArray_[1],rowArray_[3],rowArray_[2],rowArray_[0], |
---|
472 | ifValuesPass); |
---|
473 | if (ifValuesPass) { |
---|
474 | saveDj=dualIn_; |
---|
475 | if (pivotRow_==-1||(pivotRow_>=0&&fabs(alpha_)<1.0e-5)) { |
---|
476 | if(fabs(dualIn_)<1.0e2*dualTolerance_) { |
---|
477 | // try other way |
---|
478 | directionIn_=-directionIn_; |
---|
479 | primalRow(rowArray_[1],rowArray_[3],rowArray_[2],rowArray_[0], |
---|
480 | 0); |
---|
481 | } |
---|
482 | if (pivotRow_==-1||(pivotRow_>=0&&fabs(alpha_)<1.0e-5)) { |
---|
483 | // reject it |
---|
484 | char x = isColumn(sequenceIn_) ? 'C' :'R'; |
---|
485 | handler_->message(CLP_SIMPLEX_FLAG,messages_) |
---|
486 | <<x<<sequenceWithin(sequenceIn_) |
---|
487 | <<CoinMessageEol; |
---|
488 | setFlagged(sequenceIn_); |
---|
489 | lastBadIteration_ = numberIterations_; // say be more cautious |
---|
490 | rowArray_[1]->clear(); |
---|
491 | pivotRow_=-1; |
---|
492 | continue; |
---|
493 | } |
---|
494 | } |
---|
495 | } |
---|
496 | |
---|
497 | #ifdef CLP_DEBUG |
---|
498 | if ((handler_->logLevel()&32)) |
---|
499 | printf("btran dj %g, ftran dj %g\n",saveDj,dualIn_); |
---|
500 | #endif |
---|
501 | if ((saveDj*dualIn_<1.0e-20&&!ifValuesPass)|| |
---|
502 | fabs(saveDj-dualIn_)>1.0e-3*(1.0+fabs(saveDj))) { |
---|
503 | handler_->message(CLP_PRIMAL_DJ,messages_) |
---|
504 | <<saveDj<<dualIn_ |
---|
505 | <<CoinMessageEol; |
---|
506 | if(saveNumber != numberIterations_) { |
---|
507 | problemStatus_=-2; // factorize now |
---|
508 | rowArray_[1]->clear(); |
---|
509 | pivotRow_=-1; // say no weights update |
---|
510 | returnCode=-2; |
---|
511 | if(saveNumber+1 == numberIterations_) { |
---|
512 | // not looking wonderful - try cleaning bounds |
---|
513 | // put non-basics to bounds in case tolerance moved |
---|
514 | nonLinearCost_->checkInfeasibilities(true); |
---|
515 | } |
---|
516 | break; |
---|
517 | } else { |
---|
518 | // take on more relaxed criterion |
---|
519 | if (saveDj*dualIn_<1.0e-20|| |
---|
520 | fabs(saveDj-dualIn_)>1.0e-1*(1.0+fabs(dualIn_))) { |
---|
521 | // need to reject something |
---|
522 | char x = isColumn(sequenceIn_) ? 'C' :'R'; |
---|
523 | handler_->message(CLP_SIMPLEX_FLAG,messages_) |
---|
524 | <<x<<sequenceWithin(sequenceIn_) |
---|
525 | <<CoinMessageEol; |
---|
526 | setFlagged(sequenceIn_); |
---|
527 | lastBadIteration_ = numberIterations_; // say be more cautious |
---|
528 | rowArray_[1]->clear(); |
---|
529 | pivotRow_=-1; |
---|
530 | continue; |
---|
531 | } |
---|
532 | } |
---|
533 | } |
---|
534 | if (pivotRow_>=0) { |
---|
535 | // if stable replace in basis |
---|
536 | int updateStatus = factorization_->replaceColumn(rowArray_[2], |
---|
537 | pivotRow_, |
---|
538 | alpha_); |
---|
539 | if (updateStatus==1) { |
---|
540 | // slight error |
---|
541 | if (factorization_->pivots()>5) { |
---|
542 | problemStatus_=-2; // factorize now |
---|
543 | returnCode=-3; |
---|
544 | } |
---|
545 | } else if (updateStatus==2) { |
---|
546 | // major error |
---|
547 | // later we may need to unwind more e.g. fake bounds |
---|
548 | if(saveNumber != numberIterations_) { |
---|
549 | problemStatus_=-2; // factorize now |
---|
550 | returnCode=-2; |
---|
551 | break; |
---|
552 | } else { |
---|
553 | // need to reject something |
---|
554 | char x = isColumn(sequenceIn_) ? 'C' :'R'; |
---|
555 | handler_->message(CLP_SIMPLEX_FLAG,messages_) |
---|
556 | <<x<<sequenceWithin(sequenceIn_) |
---|
557 | <<CoinMessageEol; |
---|
558 | setFlagged(sequenceIn_); |
---|
559 | lastBadIteration_ = numberIterations_; // say be more cautious |
---|
560 | rowArray_[1]->clear(); |
---|
561 | pivotRow_=-1; |
---|
562 | continue; |
---|
563 | } |
---|
564 | } else if (updateStatus==3) { |
---|
565 | // out of memory |
---|
566 | // increase space if not many iterations |
---|
567 | if (factorization_->pivots()< |
---|
568 | 0.5*factorization_->maximumPivots()&& |
---|
569 | factorization_->pivots()<200) |
---|
570 | factorization_->areaFactor( |
---|
571 | factorization_->areaFactor() * 1.1); |
---|
572 | problemStatus_=-2; // factorize now |
---|
573 | } |
---|
574 | // here do part of steepest - ready for next iteration |
---|
575 | primalColumnPivot_->updateWeights(rowArray_[1]); |
---|
576 | } else { |
---|
577 | if (pivotRow_==-1) { |
---|
578 | // no outgoing row is valid |
---|
579 | #ifdef CLP_DEBUG |
---|
580 | if (handler_->logLevel()&32) |
---|
581 | printf("** no row pivot\n"); |
---|
582 | #endif |
---|
583 | if (!factorization_->pivots()) { |
---|
584 | problemStatus_=-5; //say looks unbounded |
---|
585 | // do ray |
---|
586 | delete [] ray_; |
---|
587 | ray_ = new double [numberColumns_]; |
---|
588 | ClpFillN(ray_,numberColumns_,0.0); |
---|
589 | int number=rowArray_[1]->getNumElements(); |
---|
590 | int * index = rowArray_[1]->getIndices(); |
---|
591 | double * array = rowArray_[1]->denseVector(); |
---|
592 | double way=-directionIn_; |
---|
593 | int i; |
---|
594 | double zeroTolerance=1.0e-12; |
---|
595 | if (sequenceIn_<numberColumns_) |
---|
596 | ray_[sequenceIn_]=directionIn_; |
---|
597 | for (i=0;i<number;i++) { |
---|
598 | int iRow=index[i]; |
---|
599 | int iPivot=pivotVariable_[iRow]; |
---|
600 | double arrayValue = array[iRow]; |
---|
601 | if (iPivot<numberColumns_&&fabs(arrayValue)>=zeroTolerance) |
---|
602 | ray_[iPivot] = way* array[iRow]; |
---|
603 | } |
---|
604 | } |
---|
605 | rowArray_[0]->clear(); |
---|
606 | returnCode=2; |
---|
607 | break; |
---|
608 | } else { |
---|
609 | // flipping from bound to bound |
---|
610 | } |
---|
611 | } |
---|
612 | |
---|
613 | // update primal solution |
---|
614 | |
---|
615 | double objectiveChange=0.0; |
---|
616 | // Cost on pivot row may change - may need to change dualIn |
---|
617 | double oldCost=0.0; |
---|
618 | if (pivotRow_>=0) |
---|
619 | oldCost = cost(pivotVariable_[pivotRow_]); |
---|
620 | // rowArray_[1] is not empty - used to update djs |
---|
621 | updatePrimalsInPrimal(rowArray_[1],theta_, objectiveChange); |
---|
622 | if (pivotRow_>=0) |
---|
623 | dualIn_ += (oldCost-cost(pivotVariable_[pivotRow_])); |
---|
624 | |
---|
625 | int whatNext=housekeeping(objectiveChange); |
---|
626 | |
---|
627 | if (whatNext==1) { |
---|
628 | problemStatus_ =-2; // refactorize |
---|
629 | } else if (whatNext==2) { |
---|
630 | // maximum iterations or equivalent |
---|
631 | problemStatus_= 3; |
---|
632 | returnCode=3; |
---|
633 | break; |
---|
634 | } else if(numberIterations_ == startIteration |
---|
635 | + 2 * factorization_->maximumPivots()) { |
---|
636 | // done a lot of flips - be safe |
---|
637 | problemStatus_ =-2; // refactorize |
---|
638 | } |
---|
639 | } else { |
---|
640 | // no pivot column |
---|
641 | #ifdef CLP_DEBUG |
---|
642 | if (handler_->logLevel()&32) |
---|
643 | printf("** no column pivot\n"); |
---|
644 | #endif |
---|
645 | if (nonLinearCost_->numberInfeasibilities()) |
---|
646 | problemStatus_=-4; // might be infeasible |
---|
647 | returnCode=0; |
---|
648 | break; |
---|
649 | } |
---|
650 | } |
---|
651 | return returnCode; |
---|
652 | } |
---|
653 | /* Checks if finished. Updates status */ |
---|
654 | void |
---|
655 | ClpSimplexPrimal::statusOfProblemInPrimal(int & lastCleaned,int type, |
---|
656 | ClpSimplexProgress &progress) |
---|
657 | { |
---|
658 | if (type==2) { |
---|
659 | // trouble - restore solution |
---|
660 | memcpy(status_ ,saveStatus_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
661 | memcpy(rowActivityWork_,savedSolution_+numberColumns_ , |
---|
662 | numberRows_*sizeof(double)); |
---|
663 | memcpy(columnActivityWork_,savedSolution_ , |
---|
664 | numberColumns_*sizeof(double)); |
---|
665 | forceFactorization_=1; // a bit drastic but .. |
---|
666 | pivotRow_=-1; // say no weights update |
---|
667 | changeMade_++; // say change made |
---|
668 | } |
---|
669 | int tentativeStatus = problemStatus_; |
---|
670 | if (problemStatus_>-3||problemStatus_==-4) { |
---|
671 | // factorize |
---|
672 | // later on we will need to recover from singularities |
---|
673 | // also we could skip if first time |
---|
674 | // do weights |
---|
675 | // This may save pivotRow_ for use |
---|
676 | primalColumnPivot_->saveWeights(this,1); |
---|
677 | |
---|
678 | if (type) { |
---|
679 | // is factorization okay? |
---|
680 | if (internalFactorize(1)) { |
---|
681 | // no - restore previous basis |
---|
682 | assert (type==1); |
---|
683 | memcpy(status_ ,saveStatus_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
684 | memcpy(rowActivityWork_,savedSolution_+numberColumns_ , |
---|
685 | numberRows_*sizeof(double)); |
---|
686 | memcpy(columnActivityWork_,savedSolution_ , |
---|
687 | numberColumns_*sizeof(double)); |
---|
688 | forceFactorization_=1; // a bit drastic but .. |
---|
689 | type = 2; |
---|
690 | assert (internalFactorize(1)==0); |
---|
691 | changeMade_++; // say change made |
---|
692 | } |
---|
693 | } |
---|
694 | if (problemStatus_!=-4) |
---|
695 | problemStatus_=-3; |
---|
696 | } |
---|
697 | // at this stage status is -3 or -5 if looks unbounded |
---|
698 | // get primal and dual solutions |
---|
699 | // put back original bounds and then check |
---|
700 | createRim(7); |
---|
701 | gutsOfSolution(rowActivityWork_, columnActivityWork_); |
---|
702 | // Check if looping |
---|
703 | int loop = progress.looping(); |
---|
704 | if (loop>=0) { |
---|
705 | problemStatus_ = loop; //exit if in loop |
---|
706 | return ; |
---|
707 | } else if (loop<-1) { |
---|
708 | // something may have changed |
---|
709 | gutsOfSolution(rowActivityWork_, columnActivityWork_); |
---|
710 | } |
---|
711 | progressFlag_ = 0; //reset progress flag |
---|
712 | |
---|
713 | handler_->message(CLP_SIMPLEX_STATUS,messages_) |
---|
714 | <<numberIterations_<<objectiveValue(); |
---|
715 | handler_->printing(sumPrimalInfeasibilities_>0.0) |
---|
716 | <<sumPrimalInfeasibilities_<<numberPrimalInfeasibilities_; |
---|
717 | handler_->printing(sumDualInfeasibilities_>0.0) |
---|
718 | <<sumDualInfeasibilities_<<numberDualInfeasibilities_; |
---|
719 | handler_->printing(numberDualInfeasibilitiesWithoutFree_ |
---|
720 | <numberDualInfeasibilities_) |
---|
721 | <<numberDualInfeasibilities_- |
---|
722 | numberDualInfeasibilitiesWithoutFree_; |
---|
723 | handler_->message()<<CoinMessageEol; |
---|
724 | assert (primalFeasible()); |
---|
725 | // we may wish to say it is optimal even if infeasible |
---|
726 | bool alwaysOptimal = (specialOptions_&1)!=0; |
---|
727 | // give code benefit of doubt |
---|
728 | if (sumOfRelaxedDualInfeasibilities_ == 0.0&& |
---|
729 | sumOfRelaxedPrimalInfeasibilities_ == 0.0) { |
---|
730 | // say optimal (with these bounds etc) |
---|
731 | numberDualInfeasibilities_ = 0; |
---|
732 | sumDualInfeasibilities_ = 0.0; |
---|
733 | numberPrimalInfeasibilities_ = 0; |
---|
734 | sumPrimalInfeasibilities_ = 0.0; |
---|
735 | } |
---|
736 | if (dualFeasible()||problemStatus_==-4||(type==3&&problemStatus_!=-5)) { |
---|
737 | if (nonLinearCost_->numberInfeasibilities()&&!alwaysOptimal) { |
---|
738 | //may need infeasiblity cost changed |
---|
739 | // we can see if we can construct a ray |
---|
740 | // make up a new objective |
---|
741 | double saveWeight = infeasibilityCost_; |
---|
742 | // save nonlinear cost as we are going to switch off costs |
---|
743 | ClpNonLinearCost * nonLinear = nonLinearCost_; |
---|
744 | // do twice to make sure Primal solution has settled |
---|
745 | // put non-basics to bounds in case tolerance moved |
---|
746 | // put back original bounds |
---|
747 | createRim(7); |
---|
748 | nonLinearCost_->checkInfeasibilities(true); |
---|
749 | gutsOfSolution(rowActivityWork_, columnActivityWork_); |
---|
750 | |
---|
751 | infeasibilityCost_=1.0e100; |
---|
752 | // put back original bounds |
---|
753 | createRim(7); |
---|
754 | nonLinearCost_->checkInfeasibilities(true); |
---|
755 | nonLinearCost_=NULL; |
---|
756 | // scale |
---|
757 | int i; |
---|
758 | for (i=0;i<numberRows_+numberColumns_;i++) |
---|
759 | cost_[i] *= 1.0e-100; |
---|
760 | gutsOfSolution(rowActivityWork_, columnActivityWork_); |
---|
761 | nonLinearCost_=nonLinear; |
---|
762 | infeasibilityCost_=saveWeight; |
---|
763 | if (infeasibilityCost_>=1.0e20|| |
---|
764 | numberDualInfeasibilities_==0) { |
---|
765 | // we are infeasible - use as ray |
---|
766 | delete [] ray_; |
---|
767 | ray_ = new double [numberRows_]; |
---|
768 | memcpy(ray_,dual_,numberRows_*sizeof(double)); |
---|
769 | // and get feasible duals |
---|
770 | infeasibilityCost_=0.0; |
---|
771 | createRim(7); |
---|
772 | nonLinearCost_->checkInfeasibilities(true); |
---|
773 | gutsOfSolution(rowActivityWork_, columnActivityWork_); |
---|
774 | // so will exit |
---|
775 | infeasibilityCost_=1.0e30; |
---|
776 | } |
---|
777 | |
---|
778 | if (infeasibilityCost_<1.0e20) { |
---|
779 | infeasibilityCost_ *= 5.0; |
---|
780 | changeMade_++; // say change made |
---|
781 | handler_->message(CLP_PRIMAL_WEIGHT,messages_) |
---|
782 | <<infeasibilityCost_ |
---|
783 | <<CoinMessageEol; |
---|
784 | // put back original bounds and then check |
---|
785 | createRim(7); |
---|
786 | nonLinearCost_->checkInfeasibilities(true); |
---|
787 | gutsOfSolution(rowActivityWork_, columnActivityWork_); |
---|
788 | problemStatus_=-1; //continue |
---|
789 | } else { |
---|
790 | // say infeasible |
---|
791 | problemStatus_ = 1; |
---|
792 | } |
---|
793 | } else { |
---|
794 | // may be optimal |
---|
795 | if ( lastCleaned!=numberIterations_) { |
---|
796 | handler_->message(CLP_PRIMAL_OPTIMAL,messages_) |
---|
797 | <<primalTolerance_ |
---|
798 | <<CoinMessageEol; |
---|
799 | if (numberTimesOptimal_<4) { |
---|
800 | numberTimesOptimal_++; |
---|
801 | changeMade_++; // say change made |
---|
802 | if (numberTimesOptimal_==1) { |
---|
803 | // better to have small tolerance even if slower |
---|
804 | factorization_->zeroTolerance(1.0e-15); |
---|
805 | } |
---|
806 | lastCleaned=numberIterations_; |
---|
807 | handler_->message(CLP_PRIMAL_ORIGINAL,messages_) |
---|
808 | <<CoinMessageEol; |
---|
809 | primalTolerance_=dblParam_[ClpPrimalTolerance]; |
---|
810 | |
---|
811 | // put back original bounds and then check |
---|
812 | createRim(7); |
---|
813 | nonLinearCost_->checkInfeasibilities(true); |
---|
814 | gutsOfSolution(rowActivityWork_, columnActivityWork_); |
---|
815 | problemStatus_ = -1; |
---|
816 | } else { |
---|
817 | problemStatus_=0; // optimal |
---|
818 | if (lastCleaned<numberIterations_) { |
---|
819 | handler_->message(CLP_SIMPLEX_GIVINGUP,messages_) |
---|
820 | <<CoinMessageEol; |
---|
821 | } |
---|
822 | } |
---|
823 | } else { |
---|
824 | problemStatus_=0; // optimal |
---|
825 | } |
---|
826 | } |
---|
827 | } else { |
---|
828 | // see if looks unbounded |
---|
829 | if (problemStatus_==-5) { |
---|
830 | if (nonLinearCost_->numberInfeasibilities()) { |
---|
831 | //we need infeasiblity cost changed |
---|
832 | if (infeasibilityCost_<1.0e20) { |
---|
833 | infeasibilityCost_ *= 5.0; |
---|
834 | changeMade_++; // say change made |
---|
835 | handler_->message(CLP_PRIMAL_WEIGHT,messages_) |
---|
836 | <<infeasibilityCost_ |
---|
837 | <<CoinMessageEol; |
---|
838 | // put back original bounds and then check |
---|
839 | createRim(7); |
---|
840 | gutsOfSolution(rowActivityWork_, columnActivityWork_); |
---|
841 | problemStatus_=-1; //continue |
---|
842 | } else { |
---|
843 | // say unbounded |
---|
844 | problemStatus_ = 2; |
---|
845 | } |
---|
846 | } else { |
---|
847 | // say unbounded |
---|
848 | problemStatus_ = 2; |
---|
849 | } |
---|
850 | } else { |
---|
851 | // carry on |
---|
852 | problemStatus_ = -1; |
---|
853 | } |
---|
854 | } |
---|
855 | if (type==0||type==1) { |
---|
856 | if (!type) { |
---|
857 | // create save arrays |
---|
858 | delete [] saveStatus_; |
---|
859 | delete [] savedSolution_; |
---|
860 | saveStatus_ = new unsigned char [numberRows_+numberColumns_]; |
---|
861 | savedSolution_ = new double [numberRows_+numberColumns_]; |
---|
862 | } |
---|
863 | // save arrays |
---|
864 | memcpy(saveStatus_,status_,(numberColumns_+numberRows_)*sizeof(char)); |
---|
865 | memcpy(savedSolution_+numberColumns_ ,rowActivityWork_, |
---|
866 | numberRows_*sizeof(double)); |
---|
867 | memcpy(savedSolution_ ,columnActivityWork_,numberColumns_*sizeof(double)); |
---|
868 | } |
---|
869 | // restore weights (if saved) - also recompute infeasibility list |
---|
870 | if (tentativeStatus>-3) |
---|
871 | primalColumnPivot_->saveWeights(this,(type <2) ? 2 : 4); |
---|
872 | else |
---|
873 | primalColumnPivot_->saveWeights(this,3); |
---|
874 | if (problemStatus_<0&&!changeMade_) { |
---|
875 | problemStatus_=4; // unknown |
---|
876 | } |
---|
877 | } |
---|
878 | /* |
---|
879 | Row array has pivot column |
---|
880 | This chooses pivot row. |
---|
881 | For speed, we may need to go to a bucket approach when many |
---|
882 | variables go through bounds |
---|
883 | On exit rhsArray will have changes in costs of basic variables |
---|
884 | */ |
---|
885 | void |
---|
886 | ClpSimplexPrimal::primalRow(CoinIndexedVector * rowArray, |
---|
887 | CoinIndexedVector * rhsArray, |
---|
888 | CoinIndexedVector * spareArray, |
---|
889 | CoinIndexedVector * spareArray2, |
---|
890 | int valuesPass) |
---|
891 | { |
---|
892 | if (valuesPass) { |
---|
893 | dualIn_ = cost_[sequenceIn_]; |
---|
894 | |
---|
895 | double * work=rowArray->denseVector(); |
---|
896 | int number=rowArray->getNumElements(); |
---|
897 | int * which=rowArray->getIndices(); |
---|
898 | |
---|
899 | int iIndex; |
---|
900 | |
---|
901 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
902 | |
---|
903 | int iRow = which[iIndex]; |
---|
904 | double alpha = work[iRow]; |
---|
905 | int iPivot=pivotVariable_[iRow]; |
---|
906 | dualIn_ -= alpha*cost(iPivot); |
---|
907 | } |
---|
908 | // determine direction here |
---|
909 | if (dualIn_<-dualTolerance_) { |
---|
910 | directionIn_=1; |
---|
911 | } else if (dualIn_>dualTolerance_) { |
---|
912 | directionIn_=-1; |
---|
913 | } else { |
---|
914 | // towards nearest bound |
---|
915 | if (valueIn_-lowerIn_<upperIn_-valueIn_) { |
---|
916 | directionIn_=-1; |
---|
917 | dualIn_=dualTolerance_; |
---|
918 | } else { |
---|
919 | directionIn_=1; |
---|
920 | dualIn_=-dualTolerance_; |
---|
921 | } |
---|
922 | } |
---|
923 | } |
---|
924 | |
---|
925 | // sequence stays as row number until end |
---|
926 | pivotRow_=-1; |
---|
927 | int numberSwapped=0; |
---|
928 | int numberRemaining=0; |
---|
929 | |
---|
930 | int numberThru =0; // number gone thru a barrier |
---|
931 | int lastThru =0; // number gone thru a barrier on last time |
---|
932 | |
---|
933 | double totalThru=0.0; // for when variables flip |
---|
934 | double acceptablePivot=1.0e-7; |
---|
935 | if (factorization_->pivots()) |
---|
936 | acceptablePivot=1.0e-5; // if we have iterated be more strict |
---|
937 | double bestEverPivot=acceptablePivot; |
---|
938 | int lastPivotRow = -1; |
---|
939 | double lastPivot=0.0; |
---|
940 | double lastTheta=1.0e50; |
---|
941 | int lastNumberSwapped=0; |
---|
942 | |
---|
943 | // use spareArrays to put ones looked at in |
---|
944 | // First one is list of candidates |
---|
945 | // We could compress if we really know we won't need any more |
---|
946 | // Second array has current set of pivot candidates |
---|
947 | // with a backup list saved in double * part of indexed vector |
---|
948 | |
---|
949 | // for zeroing out arrays after |
---|
950 | int maximumSwapped=0; |
---|
951 | // pivot elements |
---|
952 | double * spare; |
---|
953 | // indices |
---|
954 | int * index, * indexSwapped; |
---|
955 | int * saveSwapped; |
---|
956 | spareArray->clear(); |
---|
957 | spareArray2->clear(); |
---|
958 | spare = spareArray->denseVector(); |
---|
959 | index = spareArray->getIndices(); |
---|
960 | saveSwapped = (int *) spareArray2->denseVector(); |
---|
961 | indexSwapped = spareArray2->getIndices(); |
---|
962 | |
---|
963 | // we also need somewhere for effective rhs |
---|
964 | double * rhs=rhsArray->denseVector(); |
---|
965 | |
---|
966 | /* |
---|
967 | First we get a list of possible pivots. We can also see if the |
---|
968 | problem looks unbounded. |
---|
969 | |
---|
970 | At first we increase theta and see what happens. We start |
---|
971 | theta at a reasonable guess. If in right area then we do bit by bit. |
---|
972 | We save possible pivot candidates |
---|
973 | |
---|
974 | */ |
---|
975 | |
---|
976 | // do first pass to get possibles |
---|
977 | // We can also see if unbounded |
---|
978 | // We also re-compute reduced cost |
---|
979 | |
---|
980 | dualIn_ = cost_[sequenceIn_]; |
---|
981 | |
---|
982 | double * work=rowArray->denseVector(); |
---|
983 | int number=rowArray->getNumElements(); |
---|
984 | int * which=rowArray->getIndices(); |
---|
985 | |
---|
986 | // we need to swap sign if coming in from ub |
---|
987 | double way = directionIn_; |
---|
988 | double maximumMovement; |
---|
989 | if (way>0.0) |
---|
990 | maximumMovement = min(1.0e30,upperIn_-valueIn_); |
---|
991 | else |
---|
992 | maximumMovement = min(1.0e30,valueIn_-lowerIn_); |
---|
993 | |
---|
994 | double tentativeTheta = maximumMovement; |
---|
995 | double upperTheta = maximumMovement; |
---|
996 | |
---|
997 | int iIndex; |
---|
998 | |
---|
999 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
1000 | |
---|
1001 | int iRow = which[iIndex]; |
---|
1002 | double alpha = work[iRow]; |
---|
1003 | int iPivot=pivotVariable_[iRow]; |
---|
1004 | dualIn_ -= alpha*cost(iPivot); |
---|
1005 | alpha *= way; |
---|
1006 | double oldValue = solution(iPivot); |
---|
1007 | // get where in bound sequence |
---|
1008 | if (alpha>0.0) { |
---|
1009 | // basic variable going towards lower bound |
---|
1010 | double bound = lower(iPivot); |
---|
1011 | oldValue -= bound; |
---|
1012 | } else if (alpha<0.0) { |
---|
1013 | // basic variable going towards upper bound |
---|
1014 | double bound = upper(iPivot); |
---|
1015 | oldValue = bound-oldValue; |
---|
1016 | } |
---|
1017 | double value = oldValue-tentativeTheta*fabs(alpha); |
---|
1018 | assert (oldValue>=-primalTolerance_*1.0001); |
---|
1019 | if (value<-primalTolerance_) { |
---|
1020 | // add to list |
---|
1021 | spare[numberRemaining]=alpha; |
---|
1022 | rhs[iRow]=oldValue; |
---|
1023 | index[numberRemaining++]=iRow; |
---|
1024 | double value=oldValue-upperTheta*fabs(alpha); |
---|
1025 | if (value<-primalTolerance_&&fabs(alpha)>=acceptablePivot) |
---|
1026 | upperTheta = (oldValue+primalTolerance_)/fabs(alpha); |
---|
1027 | } |
---|
1028 | } |
---|
1029 | |
---|
1030 | // we need to keep where rhs non-zeros are |
---|
1031 | int numberInRhs=numberRemaining; |
---|
1032 | memcpy(rhsArray->getIndices(),index,numberInRhs*sizeof(int)); |
---|
1033 | rhsArray->setNumElements(numberInRhs); |
---|
1034 | |
---|
1035 | theta_=maximumMovement; |
---|
1036 | |
---|
1037 | double dualCheck = fabs(dualIn_); |
---|
1038 | // but make a bit more pessimistic |
---|
1039 | dualCheck=max(dualCheck-100.0*dualTolerance_,0.99*dualCheck); |
---|
1040 | |
---|
1041 | bool goBackOne = false; |
---|
1042 | |
---|
1043 | if (numberRemaining) { |
---|
1044 | |
---|
1045 | // looks like pivoting |
---|
1046 | // now try until reasonable theta |
---|
1047 | tentativeTheta = max(10.0*upperTheta,1.0e-7); |
---|
1048 | tentativeTheta = min(tentativeTheta,maximumMovement); |
---|
1049 | |
---|
1050 | // loops increasing tentative theta until can't go through |
---|
1051 | |
---|
1052 | while (tentativeTheta <= maximumMovement) { |
---|
1053 | double thruThis = 0.0; |
---|
1054 | |
---|
1055 | double bestPivot=acceptablePivot; |
---|
1056 | pivotRow_ = -1; |
---|
1057 | |
---|
1058 | numberSwapped = 0; |
---|
1059 | |
---|
1060 | upperTheta = maximumMovement; |
---|
1061 | |
---|
1062 | for (iIndex=0;iIndex<numberRemaining;iIndex++) { |
---|
1063 | |
---|
1064 | int iRow = index[iIndex]; |
---|
1065 | double alpha = spare[iIndex]; |
---|
1066 | double oldValue = rhs[iRow]; |
---|
1067 | double value = oldValue-tentativeTheta*fabs(alpha); |
---|
1068 | |
---|
1069 | if (value<-primalTolerance_) { |
---|
1070 | // how much would it cost to go thru |
---|
1071 | thruThis += alpha* |
---|
1072 | nonLinearCost_->changeInCost(pivotVariable_[iRow],alpha); |
---|
1073 | // goes on swapped list (also means candidates if too many) |
---|
1074 | indexSwapped[numberSwapped++]=iRow; |
---|
1075 | if (fabs(alpha)>bestPivot) { |
---|
1076 | bestPivot=fabs(alpha); |
---|
1077 | pivotRow_ = iRow; |
---|
1078 | theta_ = oldValue/bestPivot; |
---|
1079 | } |
---|
1080 | } else { |
---|
1081 | value = oldValue-upperTheta*fabs(alpha); |
---|
1082 | if (value<-primalTolerance_ && fabs(alpha)>=acceptablePivot) |
---|
1083 | upperTheta = (oldValue+primalTolerance_)/fabs(alpha); |
---|
1084 | } |
---|
1085 | } |
---|
1086 | |
---|
1087 | maximumSwapped = max(maximumSwapped,numberSwapped); |
---|
1088 | |
---|
1089 | if (totalThru+thruThis>=dualCheck) { |
---|
1090 | // We should be pivoting in this batch |
---|
1091 | // so compress down to this lot |
---|
1092 | |
---|
1093 | int saveNumber = numberRemaining; |
---|
1094 | numberRemaining=0; |
---|
1095 | for (iIndex=0;iIndex<numberSwapped;iIndex++) { |
---|
1096 | int iRow = indexSwapped[iIndex]; |
---|
1097 | spare[numberRemaining]=way*work[iRow]; |
---|
1098 | index[numberRemaining++]=iRow; |
---|
1099 | } |
---|
1100 | memset(spare+numberRemaining,0, |
---|
1101 | (saveNumber-numberRemaining)*sizeof(double)); |
---|
1102 | int iTry; |
---|
1103 | #define MAXTRY 100 |
---|
1104 | // first get ratio with tolerance |
---|
1105 | for (iTry=0;iTry<MAXTRY;iTry++) { |
---|
1106 | |
---|
1107 | upperTheta=maximumMovement; |
---|
1108 | numberSwapped = 0; |
---|
1109 | |
---|
1110 | for (iIndex=0;iIndex<numberRemaining;iIndex++) { |
---|
1111 | |
---|
1112 | int iRow = index[iIndex]; |
---|
1113 | double alpha = fabs(spare[iIndex]); |
---|
1114 | double oldValue = rhs[iRow]; |
---|
1115 | double value = oldValue-upperTheta*alpha; |
---|
1116 | |
---|
1117 | if (value<-primalTolerance_ && alpha>=acceptablePivot) |
---|
1118 | upperTheta = (oldValue+primalTolerance_)/alpha; |
---|
1119 | |
---|
1120 | } |
---|
1121 | |
---|
1122 | // now look at best in this lot |
---|
1123 | bestPivot=acceptablePivot; |
---|
1124 | pivotRow_=-1; |
---|
1125 | for (iIndex=0;iIndex<numberRemaining;iIndex++) { |
---|
1126 | |
---|
1127 | int iRow = index[iIndex]; |
---|
1128 | double alpha = spare[iIndex]; |
---|
1129 | double oldValue = rhs[iRow]; |
---|
1130 | double value = oldValue-upperTheta*fabs(alpha); |
---|
1131 | |
---|
1132 | if (value<=0) { |
---|
1133 | // how much would it cost to go thru |
---|
1134 | totalThru += alpha* |
---|
1135 | nonLinearCost_->changeInCost(pivotVariable_[iRow],alpha); |
---|
1136 | // goes on swapped list (also means candidates if too many) |
---|
1137 | indexSwapped[numberSwapped++]=iRow; |
---|
1138 | if (fabs(alpha)>bestPivot) { |
---|
1139 | bestPivot=fabs(alpha); |
---|
1140 | theta_ = oldValue/bestPivot; |
---|
1141 | pivotRow_=iRow; |
---|
1142 | } |
---|
1143 | } else { |
---|
1144 | value = oldValue-upperTheta*fabs(alpha); |
---|
1145 | if (value<-primalTolerance_ && fabs(alpha)>=acceptablePivot) |
---|
1146 | upperTheta = (oldValue+primalTolerance_)/fabs(alpha); |
---|
1147 | } |
---|
1148 | } |
---|
1149 | |
---|
1150 | maximumSwapped = max(maximumSwapped,numberSwapped); |
---|
1151 | if (bestPivot<0.1*bestEverPivot&& |
---|
1152 | bestEverPivot>1.0e-6&&bestPivot<1.0e-3) { |
---|
1153 | // back to previous one |
---|
1154 | goBackOne = true; |
---|
1155 | break; |
---|
1156 | } else if (pivotRow_==-1&&upperTheta>largeValue_) { |
---|
1157 | if (lastPivot>acceptablePivot) { |
---|
1158 | // back to previous one |
---|
1159 | goBackOne = true; |
---|
1160 | //break; |
---|
1161 | } else { |
---|
1162 | // can only get here if all pivots so far too small |
---|
1163 | } |
---|
1164 | break; |
---|
1165 | } else if (totalThru>=dualCheck) { |
---|
1166 | break; // no point trying another loop |
---|
1167 | } else { |
---|
1168 | // skip this lot |
---|
1169 | nonLinearCost_->goThru(numberSwapped,way,indexSwapped, work,rhs); |
---|
1170 | lastPivotRow=pivotRow_; |
---|
1171 | lastTheta = theta_; |
---|
1172 | lastThru = numberThru; |
---|
1173 | numberThru += numberSwapped; |
---|
1174 | lastNumberSwapped = numberSwapped; |
---|
1175 | memcpy(saveSwapped,indexSwapped,lastNumberSwapped*sizeof(int)); |
---|
1176 | if (bestPivot>bestEverPivot) |
---|
1177 | bestEverPivot=bestPivot; |
---|
1178 | } |
---|
1179 | } |
---|
1180 | break; |
---|
1181 | } else { |
---|
1182 | // skip this lot |
---|
1183 | nonLinearCost_->goThru(numberSwapped,way,indexSwapped, work,rhs); |
---|
1184 | lastPivotRow=pivotRow_; |
---|
1185 | lastTheta = theta_; |
---|
1186 | lastThru = numberThru; |
---|
1187 | numberThru += numberSwapped; |
---|
1188 | lastNumberSwapped = numberSwapped; |
---|
1189 | memcpy(saveSwapped,indexSwapped,lastNumberSwapped*sizeof(int)); |
---|
1190 | if (bestPivot>bestEverPivot) |
---|
1191 | bestEverPivot=bestPivot; |
---|
1192 | totalThru += thruThis; |
---|
1193 | tentativeTheta = 2.0*upperTheta; |
---|
1194 | } |
---|
1195 | } |
---|
1196 | // can get here without pivotRow_ set but with lastPivotRow |
---|
1197 | if (goBackOne||(pivotRow_<0&&lastPivotRow>=0)) { |
---|
1198 | // back to previous one |
---|
1199 | pivotRow_=lastPivotRow; |
---|
1200 | theta_ = lastTheta; |
---|
1201 | // undo this lot |
---|
1202 | nonLinearCost_->goBack(lastNumberSwapped,saveSwapped,rhs); |
---|
1203 | memcpy(indexSwapped,saveSwapped,lastNumberSwapped*sizeof(int)); |
---|
1204 | numberSwapped = lastNumberSwapped; |
---|
1205 | } |
---|
1206 | } |
---|
1207 | |
---|
1208 | if (pivotRow_>=0) { |
---|
1209 | |
---|
1210 | #define MINIMUMTHETA 1.0e-12 |
---|
1211 | // will we need to increase tolerance |
---|
1212 | #ifdef CLP_DEBUG |
---|
1213 | bool found=false; |
---|
1214 | #endif |
---|
1215 | double largestInfeasibility = primalTolerance_; |
---|
1216 | if (theta_<MINIMUMTHETA) { |
---|
1217 | theta_=MINIMUMTHETA; |
---|
1218 | for (iIndex=0;iIndex<numberSwapped;iIndex++) { |
---|
1219 | int iRow = indexSwapped[iIndex]; |
---|
1220 | #ifdef CLP_DEBUG |
---|
1221 | if (iRow==pivotRow_) |
---|
1222 | found=true; |
---|
1223 | #endif |
---|
1224 | largestInfeasibility = max (largestInfeasibility, |
---|
1225 | -(rhs[iRow]-fabs(work[iRow])*theta_)); |
---|
1226 | } |
---|
1227 | #ifdef CLP_DEBUG |
---|
1228 | assert(found); |
---|
1229 | if (largestInfeasibility>primalTolerance_&&(handler_->logLevel()&32)) |
---|
1230 | printf("Primal tolerance increased from %g to %g\n", |
---|
1231 | primalTolerance_,largestInfeasibility); |
---|
1232 | #endif |
---|
1233 | primalTolerance_ = max(primalTolerance_,largestInfeasibility); |
---|
1234 | } |
---|
1235 | alpha_ = work[pivotRow_]; |
---|
1236 | // translate to sequence |
---|
1237 | sequenceOut_ = pivotVariable_[pivotRow_]; |
---|
1238 | valueOut_ = solution(sequenceOut_); |
---|
1239 | lowerOut_=lower_[sequenceOut_]; |
---|
1240 | upperOut_=upper_[sequenceOut_]; |
---|
1241 | |
---|
1242 | if (way<0.0) |
---|
1243 | theta_ = - theta_; |
---|
1244 | double newValue = valueOut_ - theta_*alpha_; |
---|
1245 | if (alpha_*way<0.0) { |
---|
1246 | directionOut_=-1; // to upper bound |
---|
1247 | if (fabs(theta_)>1.0e-6) |
---|
1248 | upperOut_ = nonLinearCost_->nearest(sequenceOut_,newValue); |
---|
1249 | else |
---|
1250 | upperOut_ = newValue; |
---|
1251 | } else { |
---|
1252 | directionOut_=1; // to lower bound |
---|
1253 | if (fabs(theta_)>1.0e-6) |
---|
1254 | lowerOut_ = nonLinearCost_->nearest(sequenceOut_,newValue); |
---|
1255 | else |
---|
1256 | lowerOut_ = newValue; |
---|
1257 | } |
---|
1258 | dualOut_ = reducedCost(sequenceOut_); |
---|
1259 | } else if (maximumMovement<1.0e20) { |
---|
1260 | // flip |
---|
1261 | pivotRow_ = -2; // so we can tell its a flip |
---|
1262 | sequenceOut_ = sequenceIn_; |
---|
1263 | valueOut_ = valueIn_; |
---|
1264 | dualOut_ = dualIn_; |
---|
1265 | lowerOut_ = lowerIn_; |
---|
1266 | upperOut_ = upperIn_; |
---|
1267 | alpha_ = 0.0; |
---|
1268 | if (way<0.0) { |
---|
1269 | directionOut_=1; // to lower bound |
---|
1270 | theta_ = lowerOut_ - valueOut_; |
---|
1271 | } else { |
---|
1272 | directionOut_=-1; // to upper bound |
---|
1273 | theta_ = upperOut_ - valueOut_; |
---|
1274 | } |
---|
1275 | } |
---|
1276 | |
---|
1277 | // clear arrays |
---|
1278 | |
---|
1279 | memset(spare,0,numberRemaining*sizeof(double)); |
---|
1280 | memset(saveSwapped,0,maximumSwapped*sizeof(int)); |
---|
1281 | |
---|
1282 | // put back original bounds etc |
---|
1283 | nonLinearCost_->goBackAll(rhsArray); |
---|
1284 | |
---|
1285 | rhsArray->clear(); |
---|
1286 | |
---|
1287 | } |
---|
1288 | /* |
---|
1289 | Chooses primal pivot column |
---|
1290 | updateArray has cost updates (also use pivotRow_ from last iteration) |
---|
1291 | Would be faster with separate region to scan |
---|
1292 | and will have this (with square of infeasibility) when steepest |
---|
1293 | For easy problems we can just choose one of the first columns we look at |
---|
1294 | */ |
---|
1295 | void |
---|
1296 | ClpSimplexPrimal::primalColumn(CoinIndexedVector * updates, |
---|
1297 | CoinIndexedVector * spareRow1, |
---|
1298 | CoinIndexedVector * spareRow2, |
---|
1299 | CoinIndexedVector * spareColumn1, |
---|
1300 | CoinIndexedVector * spareColumn2) |
---|
1301 | { |
---|
1302 | sequenceIn_ = primalColumnPivot_->pivotColumn(updates,spareRow1, |
---|
1303 | spareRow2,spareColumn1, |
---|
1304 | spareColumn2); |
---|
1305 | if (sequenceIn_>=0) { |
---|
1306 | valueIn_=solution_[sequenceIn_]; |
---|
1307 | lowerIn_=lower_[sequenceIn_]; |
---|
1308 | upperIn_=upper_[sequenceIn_]; |
---|
1309 | dualIn_=dj_[sequenceIn_]; |
---|
1310 | if (dualIn_>0.0) |
---|
1311 | directionIn_ = -1; |
---|
1312 | else |
---|
1313 | directionIn_ = 1; |
---|
1314 | } else { |
---|
1315 | sequenceIn_ = -1; |
---|
1316 | } |
---|
1317 | } |
---|
1318 | /* The primals are updated by the given array. |
---|
1319 | Returns number of infeasibilities. |
---|
1320 | After rowArray will have list of cost changes |
---|
1321 | */ |
---|
1322 | int |
---|
1323 | ClpSimplexPrimal::updatePrimalsInPrimal(CoinIndexedVector * rowArray, |
---|
1324 | double theta, |
---|
1325 | double & objectiveChange) |
---|
1326 | { |
---|
1327 | double * work=rowArray->denseVector(); |
---|
1328 | int number=rowArray->getNumElements(); |
---|
1329 | int * which=rowArray->getIndices(); |
---|
1330 | |
---|
1331 | int newNumber = 0; |
---|
1332 | |
---|
1333 | nonLinearCost_->setChangeInCost(0.0); |
---|
1334 | int iIndex; |
---|
1335 | |
---|
1336 | for (iIndex=0;iIndex<number;iIndex++) { |
---|
1337 | |
---|
1338 | int iRow = which[iIndex]; |
---|
1339 | double alpha = work[iRow]; |
---|
1340 | int iPivot=pivotVariable_[iRow]; |
---|
1341 | double & value = solutionAddress(iPivot); |
---|
1342 | double change = theta*alpha; |
---|
1343 | value -= change; |
---|
1344 | |
---|
1345 | if (change>0.0) { |
---|
1346 | // going down |
---|
1347 | if (value<=lower(iPivot)+primalTolerance_) { |
---|
1348 | double difference = nonLinearCost_->setOne(iPivot,value); |
---|
1349 | work[iRow] = difference; |
---|
1350 | if (difference) { |
---|
1351 | //change reduced cost on this |
---|
1352 | reducedCostAddress(iPivot) = -difference; |
---|
1353 | which[newNumber++]=iRow; |
---|
1354 | } |
---|
1355 | } else { |
---|
1356 | work[iRow]=0.0; |
---|
1357 | } |
---|
1358 | } else { |
---|
1359 | // going up |
---|
1360 | if (value>=upper(iPivot)-primalTolerance_) { |
---|
1361 | double difference = nonLinearCost_->setOne(iPivot,value); |
---|
1362 | work[iRow] = difference; |
---|
1363 | if (difference) { |
---|
1364 | //change reduced cost on this |
---|
1365 | reducedCostAddress(iPivot) = -difference; |
---|
1366 | which[newNumber++]=iRow; |
---|
1367 | } |
---|
1368 | } else { |
---|
1369 | work[iRow]=0.0; |
---|
1370 | } |
---|
1371 | } |
---|
1372 | } |
---|
1373 | objectiveChange += nonLinearCost_->changeInCost(); |
---|
1374 | rowArray->setNumElements(newNumber); |
---|
1375 | return 0; |
---|
1376 | } |
---|
1377 | void |
---|
1378 | ClpSimplexPrimal::nextSuperBasic(int & firstSuperBasic) |
---|
1379 | { |
---|
1380 | int iColumn; |
---|
1381 | if (firstSuperBasic==numberRows_+numberColumns_) { |
---|
1382 | // initialization |
---|
1383 | iColumn=0; |
---|
1384 | } else { |
---|
1385 | // normal |
---|
1386 | sequenceIn_=firstSuperBasic; |
---|
1387 | valueIn_=solution_[sequenceIn_]; |
---|
1388 | lowerIn_=lower_[sequenceIn_]; |
---|
1389 | upperIn_=upper_[sequenceIn_]; |
---|
1390 | dualIn_=dj_[sequenceIn_]; |
---|
1391 | iColumn=firstSuperBasic+1; |
---|
1392 | } |
---|
1393 | for (;iColumn<numberRows_+numberColumns_;iColumn++) { |
---|
1394 | if (getStatus(iColumn)==superBasic) { |
---|
1395 | // is it really super basic |
---|
1396 | if (fabs(solution_[iColumn]-lower_[iColumn])<=primalTolerance_) { |
---|
1397 | solution_[iColumn]=lower_[iColumn]; |
---|
1398 | setStatus(iColumn,atLowerBound); |
---|
1399 | } else if (fabs(solution_[iColumn]-upper_[iColumn]) |
---|
1400 | <=primalTolerance_) { |
---|
1401 | solution_[iColumn]=upper_[iColumn]; |
---|
1402 | setStatus(iColumn,atUpperBound); |
---|
1403 | } else if (lower_[iColumn]<-1.0e20&&upper_[iColumn]>1.0e20) { |
---|
1404 | setStatus(iColumn,isFree); |
---|
1405 | } else { |
---|
1406 | break; |
---|
1407 | } |
---|
1408 | } |
---|
1409 | } |
---|
1410 | firstSuperBasic = iColumn; |
---|
1411 | } |
---|
1412 | // Perturbs problem |
---|
1413 | void |
---|
1414 | ClpSimplexPrimal::perturb() |
---|
1415 | { |
---|
1416 | if (perturbation_>100) |
---|
1417 | return; //perturbed already |
---|
1418 | abort(); |
---|
1419 | } |
---|
1420 | // Do not change infeasibility cost and always say optimal |
---|
1421 | void |
---|
1422 | ClpSimplexPrimal::alwaysOptimal(bool onOff) |
---|
1423 | { |
---|
1424 | if (onOff) |
---|
1425 | specialOptions_ |= 1; |
---|
1426 | else |
---|
1427 | specialOptions_ &= ~1; |
---|
1428 | } |
---|
1429 | |
---|