lemon-project-template-glpk
diff deps/glpk/src/glpios11.c @ 9:33de93886c88
Import GLPK 4.47
author | Alpar Juttner <alpar@cs.elte.hu> |
---|---|
date | Sun, 06 Nov 2011 20:59:10 +0100 |
parents | |
children |
line diff
1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 1.2 +++ b/deps/glpk/src/glpios11.c Sun Nov 06 20:59:10 2011 +0100 1.3 @@ -0,0 +1,280 @@ 1.4 +/* glpios11.c (process cuts stored in the local cut pool) */ 1.5 + 1.6 +/*********************************************************************** 1.7 +* This code is part of GLPK (GNU Linear Programming Kit). 1.8 +* 1.9 +* Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 1.10 +* 2009, 2010, 2011 Andrew Makhorin, Department for Applied Informatics, 1.11 +* Moscow Aviation Institute, Moscow, Russia. All rights reserved. 1.12 +* E-mail: <mao@gnu.org>. 1.13 +* 1.14 +* GLPK is free software: you can redistribute it and/or modify it 1.15 +* under the terms of the GNU General Public License as published by 1.16 +* the Free Software Foundation, either version 3 of the License, or 1.17 +* (at your option) any later version. 1.18 +* 1.19 +* GLPK is distributed in the hope that it will be useful, but WITHOUT 1.20 +* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 1.21 +* or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public 1.22 +* License for more details. 1.23 +* 1.24 +* You should have received a copy of the GNU General Public License 1.25 +* along with GLPK. If not, see <http://www.gnu.org/licenses/>. 1.26 +***********************************************************************/ 1.27 + 1.28 +#include "glpios.h" 1.29 + 1.30 +/*********************************************************************** 1.31 +* NAME 1.32 +* 1.33 +* ios_process_cuts - process cuts stored in the local cut pool 1.34 +* 1.35 +* SYNOPSIS 1.36 +* 1.37 +* #include "glpios.h" 1.38 +* void ios_process_cuts(glp_tree *T); 1.39 +* 1.40 +* DESCRIPTION 1.41 +* 1.42 +* The routine ios_process_cuts analyzes each cut currently stored in 1.43 +* the local cut pool, which must be non-empty, and either adds the cut 1.44 +* to the current subproblem or just discards it. All cuts are assumed 1.45 +* to be locally valid. On exit the local cut pool remains unchanged. 1.46 +* 1.47 +* REFERENCES 1.48 +* 1.49 +* 1. E.Balas, S.Ceria, G.Cornuejols, "Mixed 0-1 Programming by 1.50 +* Lift-and-Project in a Branch-and-Cut Framework", Management Sc., 1.51 +* 42 (1996) 1229-1246. 1.52 +* 1.53 +* 2. G.Andreello, A.Caprara, and M.Fischetti, "Embedding Cuts in 1.54 +* a Branch&Cut Framework: a Computational Study with {0,1/2}-Cuts", 1.55 +* Preliminary Draft, October 28, 2003, pp.6-8. */ 1.56 + 1.57 +struct info 1.58 +{ /* estimated cut efficiency */ 1.59 + IOSCUT *cut; 1.60 + /* pointer to cut in the cut pool */ 1.61 + char flag; 1.62 + /* if this flag is set, the cut is included into the current 1.63 + subproblem */ 1.64 + double eff; 1.65 + /* cut efficacy (normalized residual) */ 1.66 + double deg; 1.67 + /* lower bound to objective degradation */ 1.68 +}; 1.69 + 1.70 +static int fcmp(const void *arg1, const void *arg2) 1.71 +{ const struct info *info1 = arg1, *info2 = arg2; 1.72 + if (info1->deg == 0.0 && info2->deg == 0.0) 1.73 + { if (info1->eff > info2->eff) return -1; 1.74 + if (info1->eff < info2->eff) return +1; 1.75 + } 1.76 + else 1.77 + { if (info1->deg > info2->deg) return -1; 1.78 + if (info1->deg < info2->deg) return +1; 1.79 + } 1.80 + return 0; 1.81 +} 1.82 + 1.83 +static double parallel(IOSCUT *a, IOSCUT *b, double work[]); 1.84 + 1.85 +void ios_process_cuts(glp_tree *T) 1.86 +{ IOSPOOL *pool; 1.87 + IOSCUT *cut; 1.88 + IOSAIJ *aij; 1.89 + struct info *info; 1.90 + int k, kk, max_cuts, len, ret, *ind; 1.91 + double *val, *work; 1.92 + /* the current subproblem must exist */ 1.93 + xassert(T->curr != NULL); 1.94 + /* the pool must exist and be non-empty */ 1.95 + pool = T->local; 1.96 + xassert(pool != NULL); 1.97 + xassert(pool->size > 0); 1.98 + /* allocate working arrays */ 1.99 + info = xcalloc(1+pool->size, sizeof(struct info)); 1.100 + ind = xcalloc(1+T->n, sizeof(int)); 1.101 + val = xcalloc(1+T->n, sizeof(double)); 1.102 + work = xcalloc(1+T->n, sizeof(double)); 1.103 + for (k = 1; k <= T->n; k++) work[k] = 0.0; 1.104 + /* build the list of cuts stored in the cut pool */ 1.105 + for (k = 0, cut = pool->head; cut != NULL; cut = cut->next) 1.106 + k++, info[k].cut = cut, info[k].flag = 0; 1.107 + xassert(k == pool->size); 1.108 + /* estimate efficiency of all cuts in the cut pool */ 1.109 + for (k = 1; k <= pool->size; k++) 1.110 + { double temp, dy, dz; 1.111 + cut = info[k].cut; 1.112 + /* build the vector of cut coefficients and compute its 1.113 + Euclidean norm */ 1.114 + len = 0; temp = 0.0; 1.115 + for (aij = cut->ptr; aij != NULL; aij = aij->next) 1.116 + { xassert(1 <= aij->j && aij->j <= T->n); 1.117 + len++, ind[len] = aij->j, val[len] = aij->val; 1.118 + temp += aij->val * aij->val; 1.119 + } 1.120 + if (temp < DBL_EPSILON * DBL_EPSILON) temp = DBL_EPSILON; 1.121 + /* transform the cut to express it only through non-basic 1.122 + (auxiliary and structural) variables */ 1.123 + len = glp_transform_row(T->mip, len, ind, val); 1.124 + /* determine change in the cut value and in the objective 1.125 + value for the adjacent basis by simulating one step of the 1.126 + dual simplex */ 1.127 + ret = _glp_analyze_row(T->mip, len, ind, val, cut->type, 1.128 + cut->rhs, 1e-9, NULL, NULL, NULL, NULL, &dy, &dz); 1.129 + /* determine normalized residual and lower bound to objective 1.130 + degradation */ 1.131 + if (ret == 0) 1.132 + { info[k].eff = fabs(dy) / sqrt(temp); 1.133 + /* if some reduced costs violates (slightly) their zero 1.134 + bounds (i.e. have wrong signs) due to round-off errors, 1.135 + dz also may have wrong sign being close to zero */ 1.136 + if (T->mip->dir == GLP_MIN) 1.137 + { if (dz < 0.0) dz = 0.0; 1.138 + info[k].deg = + dz; 1.139 + } 1.140 + else /* GLP_MAX */ 1.141 + { if (dz > 0.0) dz = 0.0; 1.142 + info[k].deg = - dz; 1.143 + } 1.144 + } 1.145 + else if (ret == 1) 1.146 + { /* the constraint is not violated at the current point */ 1.147 + info[k].eff = info[k].deg = 0.0; 1.148 + } 1.149 + else if (ret == 2) 1.150 + { /* no dual feasible adjacent basis exists */ 1.151 + info[k].eff = 1.0; 1.152 + info[k].deg = DBL_MAX; 1.153 + } 1.154 + else 1.155 + xassert(ret != ret); 1.156 + /* if the degradation is too small, just ignore it */ 1.157 + if (info[k].deg < 0.01) info[k].deg = 0.0; 1.158 + } 1.159 + /* sort the list of cuts by decreasing objective degradation and 1.160 + then by decreasing efficacy */ 1.161 + qsort(&info[1], pool->size, sizeof(struct info), fcmp); 1.162 + /* only first (most efficient) max_cuts in the list are qualified 1.163 + as candidates to be added to the current subproblem */ 1.164 + max_cuts = (T->curr->level == 0 ? 90 : 10); 1.165 + if (max_cuts > pool->size) max_cuts = pool->size; 1.166 + /* add cuts to the current subproblem */ 1.167 +#if 0 1.168 + xprintf("*** adding cuts ***\n"); 1.169 +#endif 1.170 + for (k = 1; k <= max_cuts; k++) 1.171 + { int i, len; 1.172 + /* if this cut seems to be inefficient, skip it */ 1.173 + if (info[k].deg < 0.01 && info[k].eff < 0.01) continue; 1.174 + /* if the angle between this cut and every other cut included 1.175 + in the current subproblem is small, skip this cut */ 1.176 + for (kk = 1; kk < k; kk++) 1.177 + { if (info[kk].flag) 1.178 + { if (parallel(info[k].cut, info[kk].cut, work) > 0.90) 1.179 + break; 1.180 + } 1.181 + } 1.182 + if (kk < k) continue; 1.183 + /* add this cut to the current subproblem */ 1.184 +#if 0 1.185 + xprintf("eff = %g; deg = %g\n", info[k].eff, info[k].deg); 1.186 +#endif 1.187 + cut = info[k].cut, info[k].flag = 1; 1.188 + i = glp_add_rows(T->mip, 1); 1.189 + if (cut->name != NULL) 1.190 + glp_set_row_name(T->mip, i, cut->name); 1.191 + xassert(T->mip->row[i]->origin == GLP_RF_CUT); 1.192 + T->mip->row[i]->klass = cut->klass; 1.193 + len = 0; 1.194 + for (aij = cut->ptr; aij != NULL; aij = aij->next) 1.195 + len++, ind[len] = aij->j, val[len] = aij->val; 1.196 + glp_set_mat_row(T->mip, i, len, ind, val); 1.197 + xassert(cut->type == GLP_LO || cut->type == GLP_UP); 1.198 + glp_set_row_bnds(T->mip, i, cut->type, cut->rhs, cut->rhs); 1.199 + } 1.200 + /* free working arrays */ 1.201 + xfree(info); 1.202 + xfree(ind); 1.203 + xfree(val); 1.204 + xfree(work); 1.205 + return; 1.206 +} 1.207 + 1.208 +#if 0 1.209 +/*********************************************************************** 1.210 +* Given a cut a * x >= b (<= b) the routine efficacy computes the cut 1.211 +* efficacy as follows: 1.212 +* 1.213 +* eff = d * (a * x~ - b) / ||a||, 1.214 +* 1.215 +* where d is -1 (in case of '>= b') or +1 (in case of '<= b'), x~ is 1.216 +* the vector of values of structural variables in optimal solution to 1.217 +* LP relaxation of the current subproblem, ||a|| is the Euclidean norm 1.218 +* of the vector of cut coefficients. 1.219 +* 1.220 +* If the cut is violated at point x~, the efficacy eff is positive, 1.221 +* and its value is the Euclidean distance between x~ and the cut plane 1.222 +* a * x = b in the space of structural variables. 1.223 +* 1.224 +* Following geometrical intuition, it is quite natural to consider 1.225 +* this distance as a first-order measure of the expected efficacy of 1.226 +* the cut: the larger the distance the better the cut [1]. */ 1.227 + 1.228 +static double efficacy(glp_tree *T, IOSCUT *cut) 1.229 +{ glp_prob *mip = T->mip; 1.230 + IOSAIJ *aij; 1.231 + double s = 0.0, t = 0.0, temp; 1.232 + for (aij = cut->ptr; aij != NULL; aij = aij->next) 1.233 + { xassert(1 <= aij->j && aij->j <= mip->n); 1.234 + s += aij->val * mip->col[aij->j]->prim; 1.235 + t += aij->val * aij->val; 1.236 + } 1.237 + temp = sqrt(t); 1.238 + if (temp < DBL_EPSILON) temp = DBL_EPSILON; 1.239 + if (cut->type == GLP_LO) 1.240 + temp = (s >= cut->rhs ? 0.0 : (cut->rhs - s) / temp); 1.241 + else if (cut->type == GLP_UP) 1.242 + temp = (s <= cut->rhs ? 0.0 : (s - cut->rhs) / temp); 1.243 + else 1.244 + xassert(cut != cut); 1.245 + return temp; 1.246 +} 1.247 +#endif 1.248 + 1.249 +/*********************************************************************** 1.250 +* Given two cuts a1 * x >= b1 (<= b1) and a2 * x >= b2 (<= b2) the 1.251 +* routine parallel computes the cosine of angle between the cut planes 1.252 +* a1 * x = b1 and a2 * x = b2 (which is the acute angle between two 1.253 +* normals to these planes) in the space of structural variables as 1.254 +* follows: 1.255 +* 1.256 +* cos phi = (a1' * a2) / (||a1|| * ||a2||), 1.257 +* 1.258 +* where (a1' * a2) is a dot product of vectors of cut coefficients, 1.259 +* ||a1|| and ||a2|| are Euclidean norms of vectors a1 and a2. 1.260 +* 1.261 +* Note that requirement cos phi = 0 forces the cuts to be orthogonal, 1.262 +* i.e. with disjoint support, while requirement cos phi <= 0.999 means 1.263 +* only avoiding duplicate (parallel) cuts [1]. */ 1.264 + 1.265 +static double parallel(IOSCUT *a, IOSCUT *b, double work[]) 1.266 +{ IOSAIJ *aij; 1.267 + double s = 0.0, sa = 0.0, sb = 0.0, temp; 1.268 + for (aij = a->ptr; aij != NULL; aij = aij->next) 1.269 + { work[aij->j] = aij->val; 1.270 + sa += aij->val * aij->val; 1.271 + } 1.272 + for (aij = b->ptr; aij != NULL; aij = aij->next) 1.273 + { s += work[aij->j] * aij->val; 1.274 + sb += aij->val * aij->val; 1.275 + } 1.276 + for (aij = a->ptr; aij != NULL; aij = aij->next) 1.277 + work[aij->j] = 0.0; 1.278 + temp = sqrt(sa) * sqrt(sb); 1.279 + if (temp < DBL_EPSILON * DBL_EPSILON) temp = DBL_EPSILON; 1.280 + return s / temp; 1.281 +} 1.282 + 1.283 +/* eof */