rev |
line source |
alpar@9
|
1 /* glpios12.c (node selection heuristics) */
|
alpar@9
|
2
|
alpar@9
|
3 /***********************************************************************
|
alpar@9
|
4 * This code is part of GLPK (GNU Linear Programming Kit).
|
alpar@9
|
5 *
|
alpar@9
|
6 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008,
|
alpar@9
|
7 * 2009, 2010, 2011 Andrew Makhorin, Department for Applied Informatics,
|
alpar@9
|
8 * Moscow Aviation Institute, Moscow, Russia. All rights reserved.
|
alpar@9
|
9 * E-mail: <mao@gnu.org>.
|
alpar@9
|
10 *
|
alpar@9
|
11 * GLPK is free software: you can redistribute it and/or modify it
|
alpar@9
|
12 * under the terms of the GNU General Public License as published by
|
alpar@9
|
13 * the Free Software Foundation, either version 3 of the License, or
|
alpar@9
|
14 * (at your option) any later version.
|
alpar@9
|
15 *
|
alpar@9
|
16 * GLPK is distributed in the hope that it will be useful, but WITHOUT
|
alpar@9
|
17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
alpar@9
|
18 * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
alpar@9
|
19 * License for more details.
|
alpar@9
|
20 *
|
alpar@9
|
21 * You should have received a copy of the GNU General Public License
|
alpar@9
|
22 * along with GLPK. If not, see <http://www.gnu.org/licenses/>.
|
alpar@9
|
23 ***********************************************************************/
|
alpar@9
|
24
|
alpar@9
|
25 #include "glpios.h"
|
alpar@9
|
26
|
alpar@9
|
27 /***********************************************************************
|
alpar@9
|
28 * NAME
|
alpar@9
|
29 *
|
alpar@9
|
30 * ios_choose_node - select subproblem to continue the search
|
alpar@9
|
31 *
|
alpar@9
|
32 * SYNOPSIS
|
alpar@9
|
33 *
|
alpar@9
|
34 * #include "glpios.h"
|
alpar@9
|
35 * int ios_choose_node(glp_tree *T);
|
alpar@9
|
36 *
|
alpar@9
|
37 * DESCRIPTION
|
alpar@9
|
38 *
|
alpar@9
|
39 * The routine ios_choose_node selects a subproblem from the active
|
alpar@9
|
40 * list to continue the search. The choice depends on the backtracking
|
alpar@9
|
41 * technique option.
|
alpar@9
|
42 *
|
alpar@9
|
43 * RETURNS
|
alpar@9
|
44 *
|
alpar@9
|
45 * The routine ios_choose_node return the reference number of the
|
alpar@9
|
46 * subproblem selected. */
|
alpar@9
|
47
|
alpar@9
|
48 static int most_feas(glp_tree *T);
|
alpar@9
|
49 static int best_proj(glp_tree *T);
|
alpar@9
|
50 static int best_node(glp_tree *T);
|
alpar@9
|
51
|
alpar@9
|
52 int ios_choose_node(glp_tree *T)
|
alpar@9
|
53 { int p;
|
alpar@9
|
54 if (T->parm->bt_tech == GLP_BT_DFS)
|
alpar@9
|
55 { /* depth first search */
|
alpar@9
|
56 xassert(T->tail != NULL);
|
alpar@9
|
57 p = T->tail->p;
|
alpar@9
|
58 }
|
alpar@9
|
59 else if (T->parm->bt_tech == GLP_BT_BFS)
|
alpar@9
|
60 { /* breadth first search */
|
alpar@9
|
61 xassert(T->head != NULL);
|
alpar@9
|
62 p = T->head->p;
|
alpar@9
|
63 }
|
alpar@9
|
64 else if (T->parm->bt_tech == GLP_BT_BLB)
|
alpar@9
|
65 { /* select node with best local bound */
|
alpar@9
|
66 p = best_node(T);
|
alpar@9
|
67 }
|
alpar@9
|
68 else if (T->parm->bt_tech == GLP_BT_BPH)
|
alpar@9
|
69 { if (T->mip->mip_stat == GLP_UNDEF)
|
alpar@9
|
70 { /* "most integer feasible" subproblem */
|
alpar@9
|
71 p = most_feas(T);
|
alpar@9
|
72 }
|
alpar@9
|
73 else
|
alpar@9
|
74 { /* best projection heuristic */
|
alpar@9
|
75 p = best_proj(T);
|
alpar@9
|
76 }
|
alpar@9
|
77 }
|
alpar@9
|
78 else
|
alpar@9
|
79 xassert(T != T);
|
alpar@9
|
80 return p;
|
alpar@9
|
81 }
|
alpar@9
|
82
|
alpar@9
|
83 static int most_feas(glp_tree *T)
|
alpar@9
|
84 { /* select subproblem whose parent has minimal sum of integer
|
alpar@9
|
85 infeasibilities */
|
alpar@9
|
86 IOSNPD *node;
|
alpar@9
|
87 int p;
|
alpar@9
|
88 double best;
|
alpar@9
|
89 p = 0, best = DBL_MAX;
|
alpar@9
|
90 for (node = T->head; node != NULL; node = node->next)
|
alpar@9
|
91 { xassert(node->up != NULL);
|
alpar@9
|
92 if (best > node->up->ii_sum)
|
alpar@9
|
93 p = node->p, best = node->up->ii_sum;
|
alpar@9
|
94 }
|
alpar@9
|
95 return p;
|
alpar@9
|
96 }
|
alpar@9
|
97
|
alpar@9
|
98 static int best_proj(glp_tree *T)
|
alpar@9
|
99 { /* select subproblem using the best projection heuristic */
|
alpar@9
|
100 IOSNPD *root, *node;
|
alpar@9
|
101 int p;
|
alpar@9
|
102 double best, deg, obj;
|
alpar@9
|
103 /* the global bound must exist */
|
alpar@9
|
104 xassert(T->mip->mip_stat == GLP_FEAS);
|
alpar@9
|
105 /* obtain pointer to the root node, which must exist */
|
alpar@9
|
106 root = T->slot[1].node;
|
alpar@9
|
107 xassert(root != NULL);
|
alpar@9
|
108 /* deg estimates degradation of the objective function per unit
|
alpar@9
|
109 of the sum of integer infeasibilities */
|
alpar@9
|
110 xassert(root->ii_sum > 0.0);
|
alpar@9
|
111 deg = (T->mip->mip_obj - root->bound) / root->ii_sum;
|
alpar@9
|
112 /* nothing has been selected so far */
|
alpar@9
|
113 p = 0, best = DBL_MAX;
|
alpar@9
|
114 /* walk through the list of active subproblems */
|
alpar@9
|
115 for (node = T->head; node != NULL; node = node->next)
|
alpar@9
|
116 { xassert(node->up != NULL);
|
alpar@9
|
117 /* obj estimates optimal objective value if the sum of integer
|
alpar@9
|
118 infeasibilities were zero */
|
alpar@9
|
119 obj = node->up->bound + deg * node->up->ii_sum;
|
alpar@9
|
120 if (T->mip->dir == GLP_MAX) obj = - obj;
|
alpar@9
|
121 /* select the subproblem which has the best estimated optimal
|
alpar@9
|
122 objective value */
|
alpar@9
|
123 if (best > obj) p = node->p, best = obj;
|
alpar@9
|
124 }
|
alpar@9
|
125 return p;
|
alpar@9
|
126 }
|
alpar@9
|
127
|
alpar@9
|
128 static int best_node(glp_tree *T)
|
alpar@9
|
129 { /* select subproblem with best local bound */
|
alpar@9
|
130 IOSNPD *node, *best = NULL;
|
alpar@9
|
131 double bound, eps;
|
alpar@9
|
132 switch (T->mip->dir)
|
alpar@9
|
133 { case GLP_MIN:
|
alpar@9
|
134 bound = +DBL_MAX;
|
alpar@9
|
135 for (node = T->head; node != NULL; node = node->next)
|
alpar@9
|
136 if (bound > node->bound) bound = node->bound;
|
alpar@9
|
137 xassert(bound != +DBL_MAX);
|
alpar@9
|
138 eps = 0.001 * (1.0 + fabs(bound));
|
alpar@9
|
139 for (node = T->head; node != NULL; node = node->next)
|
alpar@9
|
140 { if (node->bound <= bound + eps)
|
alpar@9
|
141 { xassert(node->up != NULL);
|
alpar@9
|
142 if (best == NULL ||
|
alpar@9
|
143 #if 1
|
alpar@9
|
144 best->up->ii_sum > node->up->ii_sum) best = node;
|
alpar@9
|
145 #else
|
alpar@9
|
146 best->lp_obj > node->lp_obj) best = node;
|
alpar@9
|
147 #endif
|
alpar@9
|
148 }
|
alpar@9
|
149 }
|
alpar@9
|
150 break;
|
alpar@9
|
151 case GLP_MAX:
|
alpar@9
|
152 bound = -DBL_MAX;
|
alpar@9
|
153 for (node = T->head; node != NULL; node = node->next)
|
alpar@9
|
154 if (bound < node->bound) bound = node->bound;
|
alpar@9
|
155 xassert(bound != -DBL_MAX);
|
alpar@9
|
156 eps = 0.001 * (1.0 + fabs(bound));
|
alpar@9
|
157 for (node = T->head; node != NULL; node = node->next)
|
alpar@9
|
158 { if (node->bound >= bound - eps)
|
alpar@9
|
159 { xassert(node->up != NULL);
|
alpar@9
|
160 if (best == NULL ||
|
alpar@9
|
161 #if 1
|
alpar@9
|
162 best->up->ii_sum > node->up->ii_sum) best = node;
|
alpar@9
|
163 #else
|
alpar@9
|
164 best->lp_obj < node->lp_obj) best = node;
|
alpar@9
|
165 #endif
|
alpar@9
|
166 }
|
alpar@9
|
167 }
|
alpar@9
|
168 break;
|
alpar@9
|
169 default:
|
alpar@9
|
170 xassert(T != T);
|
alpar@9
|
171 }
|
alpar@9
|
172 xassert(best != NULL);
|
alpar@9
|
173 return best->p;
|
alpar@9
|
174 }
|
alpar@9
|
175
|
alpar@9
|
176 /* eof */
|