gravatar
kpeter (Peter Kovacs)
kpeter@inf.elte.hu
Implement the scaling Price Refinement heuristic in CostScaling (#417) instead of Early Termination. These two heuristics are similar, but the newer one is faster and not only makes it possible to skip some epsilon phases, but it can improve the performance of the other phases, as well.
0 1 0
default
1 file changed with 232 insertions and 27 deletions:
↑ Collapse diff ↑
Show white space 192 line context
... ...
@@ -887,453 +887,658 @@
887 887
          Value fa = flow[a];
888 888
          _res_cap[_arc_idf[a]] = cap[a] - fa;
889 889
          _res_cap[_arc_idb[a]] = fa;
890 890
        }
891 891
        for (int a = _first_out[_root]; a != _res_arc_num; ++a) {
892 892
          int ra = _reverse[a];
893 893
          _res_cap[a] = 0;
894 894
          _res_cap[ra] = 0;
895 895
          _cost[a] = 0;
896 896
          _cost[ra] = 0;
897 897
        }
898 898
      }
899 899

	
900 900
      // Initialize data structures for buckets
901 901
      _max_rank = _alpha * _res_node_num;
902 902
      _buckets.resize(_max_rank);
903 903
      _bucket_next.resize(_res_node_num + 1);
904 904
      _bucket_prev.resize(_res_node_num + 1);
905 905
      _rank.resize(_res_node_num + 1);
906 906

	
907 907
      return OPTIMAL;
908 908
    }
909 909

	
910 910
    // Execute the algorithm and transform the results
911 911
    void start(Method method) {
912 912
      const int MAX_PARTIAL_PATH_LENGTH = 4;
913 913

	
914 914
      switch (method) {
915 915
        case PUSH:
916 916
          startPush();
917 917
          break;
918 918
        case AUGMENT:
919 919
          startAugment(_res_node_num - 1);
920 920
          break;
921 921
        case PARTIAL_AUGMENT:
922 922
          startAugment(MAX_PARTIAL_PATH_LENGTH);
923 923
          break;
924 924
      }
925 925

	
926 926
      // Compute node potentials for the original costs
927 927
      _arc_vec.clear();
928 928
      _cost_vec.clear();
929 929
      for (int j = 0; j != _res_arc_num; ++j) {
930 930
        if (_res_cap[j] > 0) {
931 931
          _arc_vec.push_back(IntPair(_source[j], _target[j]));
932 932
          _cost_vec.push_back(_scost[j]);
933 933
        }
934 934
      }
935 935
      _sgr.build(_res_node_num, _arc_vec.begin(), _arc_vec.end());
936 936

	
937 937
      typename BellmanFord<StaticDigraph, LargeCostArcMap>
938 938
        ::template SetDistMap<LargeCostNodeMap>::Create bf(_sgr, _cost_map);
939 939
      bf.distMap(_pi_map);
940 940
      bf.init(0);
941 941
      bf.start();
942 942

	
943 943
      // Handle non-zero lower bounds
944 944
      if (_have_lower) {
945 945
        int limit = _first_out[_root];
946 946
        for (int j = 0; j != limit; ++j) {
947 947
          if (!_forward[j]) _res_cap[j] += _lower[j];
948 948
        }
949 949
      }
950 950
    }
951 951

	
952 952
    // Initialize a cost scaling phase
953 953
    void initPhase() {
954 954
      // Saturate arcs not satisfying the optimality condition
955 955
      for (int u = 0; u != _res_node_num; ++u) {
956 956
        int last_out = _first_out[u+1];
957 957
        LargeCost pi_u = _pi[u];
958 958
        for (int a = _first_out[u]; a != last_out; ++a) {
959 959
          Value delta = _res_cap[a];
960 960
          if (delta > 0) {
961 961
            int v = _target[a];
962 962
            if (_cost[a] + pi_u - _pi[v] < 0) {
963 963
              _excess[u] -= delta;
964 964
              _excess[v] += delta;
965 965
              _res_cap[a] = 0;
966 966
              _res_cap[_reverse[a]] += delta;
967 967
            }
968 968
          }
969 969
        }
970 970
      }
971 971

	
972 972
      // Find active nodes (i.e. nodes with positive excess)
973 973
      for (int u = 0; u != _res_node_num; ++u) {
974 974
        if (_excess[u] > 0) _active_nodes.push_back(u);
975 975
      }
976 976

	
977 977
      // Initialize the next arcs
978 978
      for (int u = 0; u != _res_node_num; ++u) {
979 979
        _next_out[u] = _first_out[u];
980 980
      }
981 981
    }
982 982

	
983
    // Early termination heuristic
984
    bool earlyTermination() {
985
      const double EARLY_TERM_FACTOR = 3.0;
983
    // Price (potential) refinement heuristic
984
    bool priceRefinement() {
986 985

	
987
      // Build a static residual graph
988
      _arc_vec.clear();
989
      _cost_vec.clear();
990
      for (int j = 0; j != _res_arc_num; ++j) {
991
        if (_res_cap[j] > 0) {
992
          _arc_vec.push_back(IntPair(_source[j], _target[j]));
993
          _cost_vec.push_back(_cost[j] + 1);
986
      // Stack for stroing the topological order
987
      IntVector stack(_res_node_num);
988
      int stack_top;
989

	
990
      // Perform phases
991
      while (topologicalSort(stack, stack_top)) {
992

	
993
        // Compute node ranks in the acyclic admissible network and
994
        // store the nodes in buckets
995
        for (int i = 0; i != _res_node_num; ++i) {
996
          _rank[i] = 0;
994 997
        }
998
        const int bucket_end = _root + 1;
999
        for (int r = 0; r != _max_rank; ++r) {
1000
          _buckets[r] = bucket_end;
995 1001
      }
996
      _sgr.build(_res_node_num, _arc_vec.begin(), _arc_vec.end());
1002
        int top_rank = 0;
1003
        for ( ; stack_top >= 0; --stack_top) {
1004
          int u = stack[stack_top], v;
1005
          int rank_u = _rank[u];
997 1006

	
998
      // Run Bellman-Ford algorithm to check if the current flow is optimal
999
      BellmanFord<StaticDigraph, LargeCostArcMap> bf(_sgr, _cost_map);
1000
      bf.init(0);
1001
      bool done = false;
1002
      int K = int(EARLY_TERM_FACTOR * std::sqrt(double(_res_node_num)));
1003
      for (int i = 0; i < K && !done; ++i) {
1004
        done = bf.processNextWeakRound();
1007
          LargeCost rc, pi_u = _pi[u];
1008
          int last_out = _first_out[u+1];
1009
          for (int a = _first_out[u]; a != last_out; ++a) {
1010
            if (_res_cap[a] > 0) {
1011
              v = _target[a];
1012
              rc = _cost[a] + pi_u - _pi[v];
1013
              if (rc < 0) {
1014
                LargeCost nrc = static_cast<LargeCost>((-rc - 0.5) / _epsilon);
1015
                if (nrc < LargeCost(_max_rank)) {
1016
                  int new_rank_v = rank_u + static_cast<int>(nrc);
1017
                  if (new_rank_v > _rank[v]) {
1018
                    _rank[v] = new_rank_v;
1005 1019
      }
1006
      return done;
1020
                }
1021
              }
1022
            }
1023
          }
1024

	
1025
          if (rank_u > 0) {
1026
            top_rank = std::max(top_rank, rank_u);
1027
            int bfirst = _buckets[rank_u];
1028
            _bucket_next[u] = bfirst;
1029
            _bucket_prev[bfirst] = u;
1030
            _buckets[rank_u] = u;
1031
          }
1032
        }
1033

	
1034
        // Check if the current flow is epsilon-optimal
1035
        if (top_rank == 0) {
1036
          return true;
1037
        }
1038

	
1039
        // Process buckets in top-down order
1040
        for (int rank = top_rank; rank > 0; --rank) {
1041
          while (_buckets[rank] != bucket_end) {
1042
            // Remove the first node from the current bucket
1043
            int u = _buckets[rank];
1044
            _buckets[rank] = _bucket_next[u];
1045

	
1046
            // Search the outgoing arcs of u
1047
            LargeCost rc, pi_u = _pi[u];
1048
            int last_out = _first_out[u+1];
1049
            int v, old_rank_v, new_rank_v;
1050
            for (int a = _first_out[u]; a != last_out; ++a) {
1051
              if (_res_cap[a] > 0) {
1052
                v = _target[a];
1053
                old_rank_v = _rank[v];
1054

	
1055
                if (old_rank_v < rank) {
1056

	
1057
                  // Compute the new rank of node v
1058
                  rc = _cost[a] + pi_u - _pi[v];
1059
                  if (rc < 0) {
1060
                    new_rank_v = rank;
1061
                  } else {
1062
                    LargeCost nrc = rc / _epsilon;
1063
                    new_rank_v = 0;
1064
                    if (nrc < LargeCost(_max_rank)) {
1065
                      new_rank_v = rank - 1 - static_cast<int>(nrc);
1066
                    }
1067
                  }
1068

	
1069
                  // Change the rank of node v
1070
                  if (new_rank_v > old_rank_v) {
1071
                    _rank[v] = new_rank_v;
1072

	
1073
                    // Remove v from its old bucket
1074
                    if (old_rank_v > 0) {
1075
                      if (_buckets[old_rank_v] == v) {
1076
                        _buckets[old_rank_v] = _bucket_next[v];
1077
                      } else {
1078
                        int pv = _bucket_prev[v], nv = _bucket_next[v];
1079
                        _bucket_next[pv] = nv;
1080
                        _bucket_prev[nv] = pv;
1081
                      }
1082
                    }
1083

	
1084
                    // Insert v into its new bucket
1085
                    int nv = _buckets[new_rank_v];
1086
                    _bucket_next[v] = nv;
1087
                    _bucket_prev[nv] = v;
1088
                    _buckets[new_rank_v] = v;
1089
                  }
1090
                }
1091
              }
1092
            }
1093

	
1094
            // Refine potential of node u
1095
            _pi[u] -= rank * _epsilon;
1096
          }
1097
        }
1098

	
1099
      }
1100

	
1101
      return false;
1102
    }
1103

	
1104
    // Find and cancel cycles in the admissible network and
1105
    // determine topological order using DFS
1106
    bool topologicalSort(IntVector &stack, int &stack_top) {
1107
      const int MAX_CYCLE_CANCEL = 1;
1108

	
1109
      BoolVector reached(_res_node_num, false);
1110
      BoolVector processed(_res_node_num, false);
1111
      IntVector pred(_res_node_num);
1112
      for (int i = 0; i != _res_node_num; ++i) {
1113
        _next_out[i] = _first_out[i];
1114
      }
1115
      stack_top = -1;
1116

	
1117
      int cycle_cnt = 0;
1118
      for (int start = 0; start != _res_node_num; ++start) {
1119
        if (reached[start]) continue;
1120

	
1121
        // Start DFS search from this start node
1122
        pred[start] = -1;
1123
        int tip = start, v;
1124
        while (true) {
1125
          // Check the outgoing arcs of the current tip node
1126
          reached[tip] = true;
1127
          LargeCost pi_tip = _pi[tip];
1128
          int a, last_out = _first_out[tip+1];
1129
          for (a = _next_out[tip]; a != last_out; ++a) {
1130
            if (_res_cap[a] > 0) {
1131
              v = _target[a];
1132
              if (_cost[a] + pi_tip - _pi[v] < 0) {
1133
                if (!reached[v]) {
1134
                  // A new node is reached
1135
                  reached[v] = true;
1136
                  pred[v] = tip;
1137
                  _next_out[tip] = a;
1138
                  tip = v;
1139
                  a = _next_out[tip];
1140
                  last_out = _first_out[tip+1];
1141
                  break;
1142
                }
1143
                else if (!processed[v]) {
1144
                  // A cycle is found
1145
                  ++cycle_cnt;
1146
                  _next_out[tip] = a;
1147

	
1148
                  // Find the minimum residual capacity along the cycle
1149
                  Value d, delta = _res_cap[a];
1150
                  int u, delta_node = tip;
1151
                  for (u = tip; u != v; ) {
1152
                    u = pred[u];
1153
                    d = _res_cap[_next_out[u]];
1154
                    if (d <= delta) {
1155
                      delta = d;
1156
                      delta_node = u;
1157
                    }
1158
                  }
1159

	
1160
                  // Augment along the cycle
1161
                  _res_cap[a] -= delta;
1162
                  _res_cap[_reverse[a]] += delta;
1163
                  for (u = tip; u != v; ) {
1164
                    u = pred[u];
1165
                    int ca = _next_out[u];
1166
                    _res_cap[ca] -= delta;
1167
                    _res_cap[_reverse[ca]] += delta;
1168
                  }
1169

	
1170
                  // Check the maximum number of cycle canceling
1171
                  if (cycle_cnt >= MAX_CYCLE_CANCEL) {
1172
                    return false;
1173
                  }
1174

	
1175
                  // Roll back search to delta_node
1176
                  if (delta_node != tip) {
1177
                    for (u = tip; u != delta_node; u = pred[u]) {
1178
                      reached[u] = false;
1179
                    }
1180
                    tip = delta_node;
1181
                    a = _next_out[tip] + 1;
1182
                    last_out = _first_out[tip+1];
1183
                    break;
1184
                  }
1185
                }
1186
              }
1187
            }
1188
          }
1189

	
1190
          // Step back to the previous node
1191
          if (a == last_out) {
1192
            processed[tip] = true;
1193
            stack[++stack_top] = tip;
1194
            tip = pred[tip];
1195
            if (tip < 0) {
1196
              // Finish DFS from the current start node
1197
              break;
1198
            }
1199
            ++_next_out[tip];
1200
          }
1201
        }
1202

	
1203
      }
1204

	
1205
      return (cycle_cnt == 0);
1007 1206
    }
1008 1207

	
1009 1208
    // Global potential update heuristic
1010 1209
    void globalUpdate() {
1011 1210
      const int bucket_end = _root + 1;
1012 1211

	
1013 1212
      // Initialize buckets
1014 1213
      for (int r = 0; r != _max_rank; ++r) {
1015 1214
        _buckets[r] = bucket_end;
1016 1215
      }
1017 1216
      Value total_excess = 0;
1018 1217
      int b0 = bucket_end;
1019 1218
      for (int i = 0; i != _res_node_num; ++i) {
1020 1219
        if (_excess[i] < 0) {
1021 1220
          _rank[i] = 0;
1022 1221
          _bucket_next[i] = b0;
1023 1222
          _bucket_prev[b0] = i;
1024 1223
          b0 = i;
1025 1224
        } else {
1026 1225
          total_excess += _excess[i];
1027 1226
          _rank[i] = _max_rank;
1028 1227
        }
1029 1228
      }
1030 1229
      if (total_excess == 0) return;
1031 1230
      _buckets[0] = b0;
1032 1231

	
1033 1232
      // Search the buckets
1034 1233
      int r = 0;
1035 1234
      for ( ; r != _max_rank; ++r) {
1036 1235
        while (_buckets[r] != bucket_end) {
1037 1236
          // Remove the first node from the current bucket
1038 1237
          int u = _buckets[r];
1039 1238
          _buckets[r] = _bucket_next[u];
1040 1239

	
1041 1240
          // Search the incomming arcs of u
1042 1241
          LargeCost pi_u = _pi[u];
1043 1242
          int last_out = _first_out[u+1];
1044 1243
          for (int a = _first_out[u]; a != last_out; ++a) {
1045 1244
            int ra = _reverse[a];
1046 1245
            if (_res_cap[ra] > 0) {
1047 1246
              int v = _source[ra];
1048 1247
              int old_rank_v = _rank[v];
1049 1248
              if (r < old_rank_v) {
1050 1249
                // Compute the new rank of v
1051 1250
                LargeCost nrc = (_cost[ra] + _pi[v] - pi_u) / _epsilon;
1052 1251
                int new_rank_v = old_rank_v;
1053 1252
                if (nrc < LargeCost(_max_rank)) {
1054 1253
                  new_rank_v = r + 1 + static_cast<int>(nrc);
1055 1254
                }
1056 1255

	
1057 1256
                // Change the rank of v
1058 1257
                if (new_rank_v < old_rank_v) {
1059 1258
                  _rank[v] = new_rank_v;
1060 1259
                  _next_out[v] = _first_out[v];
1061 1260

	
1062 1261
                  // Remove v from its old bucket
1063 1262
                  if (old_rank_v < _max_rank) {
1064 1263
                    if (_buckets[old_rank_v] == v) {
1065 1264
                      _buckets[old_rank_v] = _bucket_next[v];
1066 1265
                    } else {
1067 1266
                      int pv = _bucket_prev[v], nv = _bucket_next[v];
1068 1267
                      _bucket_next[pv] = nv;
1069 1268
                      _bucket_prev[nv] = pv;
1070 1269
                    }
1071 1270
                  }
1072 1271

	
1073 1272
                  // Insert v into its new bucket
1074 1273
                  int nv = _buckets[new_rank_v];
1075 1274
                  _bucket_next[v] = nv;
1076 1275
                  _bucket_prev[nv] = v;
1077 1276
                  _buckets[new_rank_v] = v;
1078 1277
                }
1079 1278
              }
1080 1279
            }
1081 1280
          }
1082 1281

	
1083 1282
          // Finish search if there are no more active nodes
1084 1283
          if (_excess[u] > 0) {
1085 1284
            total_excess -= _excess[u];
1086 1285
            if (total_excess <= 0) break;
1087 1286
          }
1088 1287
        }
1089 1288
        if (total_excess <= 0) break;
1090 1289
      }
1091 1290

	
1092 1291
      // Relabel nodes
1093 1292
      for (int u = 0; u != _res_node_num; ++u) {
1094 1293
        int k = std::min(_rank[u], r);
1095 1294
        if (k > 0) {
1096 1295
          _pi[u] -= _epsilon * k;
1097 1296
          _next_out[u] = _first_out[u];
1098 1297
        }
1099 1298
      }
1100 1299
    }
1101 1300

	
1102 1301
    /// Execute the algorithm performing augment and relabel operations
1103 1302
    void startAugment(int max_length) {
1104 1303
      // Paramters for heuristics
1105
      const int EARLY_TERM_EPSILON_LIMIT = 1000;
1304
      const int PRICE_REFINEMENT_LIMIT = 2;
1106 1305
      const double GLOBAL_UPDATE_FACTOR = 1.0;
1107 1306
      const int global_update_skip = static_cast<int>(GLOBAL_UPDATE_FACTOR *
1108 1307
        (_res_node_num + _sup_node_num * _sup_node_num));
1109 1308
      int next_global_update_limit = global_update_skip;
1110 1309

	
1111 1310
      // Perform cost scaling phases
1112 1311
      IntVector path;
1113 1312
      BoolVector path_arc(_res_arc_num, false);
1114 1313
      int relabel_cnt = 0;
1314
      int eps_phase_cnt = 0;
1115 1315
      for ( ; _epsilon >= 1; _epsilon = _epsilon < _alpha && _epsilon > 1 ?
1116 1316
                                        1 : _epsilon / _alpha )
1117 1317
      {
1118
        // Early termination heuristic
1119
        if (_epsilon <= EARLY_TERM_EPSILON_LIMIT) {
1120
          if (earlyTermination()) break;
1318
        ++eps_phase_cnt;
1319

	
1320
        // Price refinement heuristic
1321
        if (eps_phase_cnt >= PRICE_REFINEMENT_LIMIT) {
1322
          if (priceRefinement()) continue;
1121 1323
        }
1122 1324

	
1123 1325
        // Initialize current phase
1124 1326
        initPhase();
1125 1327

	
1126 1328
        // Perform partial augment and relabel operations
1127 1329
        while (true) {
1128 1330
          // Select an active node (FIFO selection)
1129 1331
          while (_active_nodes.size() > 0 &&
1130 1332
                 _excess[_active_nodes.front()] <= 0) {
1131 1333
            _active_nodes.pop_front();
1132 1334
          }
1133 1335
          if (_active_nodes.size() == 0) break;
1134 1336
          int start = _active_nodes.front();
1135 1337

	
1136 1338
          // Find an augmenting path from the start node
1137 1339
          int tip = start;
1138 1340
          while (int(path.size()) < max_length && _excess[tip] >= 0) {
1139 1341
            int u;
1140 1342
            LargeCost rc, min_red_cost = std::numeric_limits<LargeCost>::max();
1141 1343
            LargeCost pi_tip = _pi[tip];
1142 1344
            int last_out = _first_out[tip+1];
1143 1345
            for (int a = _next_out[tip]; a != last_out; ++a) {
1144 1346
              if (_res_cap[a] > 0) {
1145 1347
                u = _target[a];
1146 1348
                rc = _cost[a] + pi_tip - _pi[u];
1147 1349
                if (rc < 0) {
1148 1350
                  path.push_back(a);
1149 1351
                  _next_out[tip] = a;
1150 1352
                  if (path_arc[a]) {
1151 1353
                    goto augment;   // a cycle is found, stop path search
1152 1354
                  }
1153 1355
                  tip = u;
1154 1356
                  path_arc[a] = true;
1155 1357
                  goto next_step;
1156 1358
                }
1157 1359
                else if (rc < min_red_cost) {
1158 1360
                  min_red_cost = rc;
1159 1361
                }
1160 1362
              }
1161 1363
            }
1162 1364

	
1163 1365
            // Relabel tip node
1164 1366
            if (tip != start) {
1165 1367
              int ra = _reverse[path.back()];
1166 1368
              min_red_cost =
1167 1369
                std::min(min_red_cost, _cost[ra] + pi_tip - _pi[_target[ra]]);
1168 1370
            }
1169 1371
            last_out = _next_out[tip];
1170 1372
            for (int a = _first_out[tip]; a != last_out; ++a) {
1171 1373
              if (_res_cap[a] > 0) {
1172 1374
                rc = _cost[a] + pi_tip - _pi[_target[a]];
1173 1375
                if (rc < min_red_cost) {
1174 1376
                  min_red_cost = rc;
1175 1377
                }
1176 1378
              }
1177 1379
            }
1178 1380
            _pi[tip] -= min_red_cost + _epsilon;
1179 1381
            _next_out[tip] = _first_out[tip];
1180 1382
            ++relabel_cnt;
1181 1383

	
1182 1384
            // Step back
1183 1385
            if (tip != start) {
1184 1386
              int pa = path.back();
1185 1387
              path_arc[pa] = false;
1186 1388
              tip = _source[pa];
1187 1389
              path.pop_back();
1188 1390
            }
1189 1391

	
1190 1392
          next_step: ;
1191 1393
          }
1192 1394

	
1193 1395
          // Augment along the found path (as much flow as possible)
1194 1396
        augment:
1195 1397
          Value delta;
1196 1398
          int pa, u, v = start;
1197 1399
          for (int i = 0; i != int(path.size()); ++i) {
1198 1400
            pa = path[i];
1199 1401
            u = v;
1200 1402
            v = _target[pa];
1201 1403
            path_arc[pa] = false;
1202 1404
            delta = std::min(_res_cap[pa], _excess[u]);
1203 1405
            _res_cap[pa] -= delta;
1204 1406
            _res_cap[_reverse[pa]] += delta;
1205 1407
            _excess[u] -= delta;
1206 1408
            _excess[v] += delta;
1207 1409
            if (_excess[v] > 0 && _excess[v] <= delta) {
1208 1410
              _active_nodes.push_back(v);
1209 1411
            }
1210 1412
          }
1211 1413
          path.clear();
1212 1414

	
1213 1415
          // Global update heuristic
1214 1416
          if (relabel_cnt >= next_global_update_limit) {
1215 1417
            globalUpdate();
1216 1418
            next_global_update_limit += global_update_skip;
1217 1419
          }
1218 1420
        }
1219 1421

	
1220 1422
      }
1221 1423

	
1222 1424
    }
1223 1425

	
1224 1426
    /// Execute the algorithm performing push and relabel operations
1225 1427
    void startPush() {
1226 1428
      // Paramters for heuristics
1227
      const int EARLY_TERM_EPSILON_LIMIT = 1000;
1429
      const int PRICE_REFINEMENT_LIMIT = 2;
1228 1430
      const double GLOBAL_UPDATE_FACTOR = 2.0;
1229 1431

	
1230 1432
      const int global_update_skip = static_cast<int>(GLOBAL_UPDATE_FACTOR *
1231 1433
        (_res_node_num + _sup_node_num * _sup_node_num));
1232 1434
      int next_global_update_limit = global_update_skip;
1233 1435

	
1234 1436
      // Perform cost scaling phases
1235 1437
      BoolVector hyper(_res_node_num, false);
1236 1438
      LargeCostVector hyper_cost(_res_node_num);
1237 1439
      int relabel_cnt = 0;
1440
      int eps_phase_cnt = 0;
1238 1441
      for ( ; _epsilon >= 1; _epsilon = _epsilon < _alpha && _epsilon > 1 ?
1239 1442
                                        1 : _epsilon / _alpha )
1240 1443
      {
1241
        // Early termination heuristic
1242
        if (_epsilon <= EARLY_TERM_EPSILON_LIMIT) {
1243
          if (earlyTermination()) break;
1444
        ++eps_phase_cnt;
1445

	
1446
        // Price refinement heuristic
1447
        if (eps_phase_cnt >= PRICE_REFINEMENT_LIMIT) {
1448
          if (priceRefinement()) continue;
1244 1449
        }
1245 1450

	
1246 1451
        // Initialize current phase
1247 1452
        initPhase();
1248 1453

	
1249 1454
        // Perform push and relabel operations
1250 1455
        while (_active_nodes.size() > 0) {
1251 1456
          LargeCost min_red_cost, rc, pi_n;
1252 1457
          Value delta;
1253 1458
          int n, t, a, last_out = _res_arc_num;
1254 1459

	
1255 1460
        next_node:
1256 1461
          // Select an active node (FIFO selection)
1257 1462
          n = _active_nodes.front();
1258 1463
          last_out = _first_out[n+1];
1259 1464
          pi_n = _pi[n];
1260 1465

	
1261 1466
          // Perform push operations if there are admissible arcs
1262 1467
          if (_excess[n] > 0) {
1263 1468
            for (a = _next_out[n]; a != last_out; ++a) {
1264 1469
              if (_res_cap[a] > 0 &&
1265 1470
                  _cost[a] + pi_n - _pi[_target[a]] < 0) {
1266 1471
                delta = std::min(_res_cap[a], _excess[n]);
1267 1472
                t = _target[a];
1268 1473

	
1269 1474
                // Push-look-ahead heuristic
1270 1475
                Value ahead = -_excess[t];
1271 1476
                int last_out_t = _first_out[t+1];
1272 1477
                LargeCost pi_t = _pi[t];
1273 1478
                for (int ta = _next_out[t]; ta != last_out_t; ++ta) {
1274 1479
                  if (_res_cap[ta] > 0 &&
1275 1480
                      _cost[ta] + pi_t - _pi[_target[ta]] < 0)
1276 1481
                    ahead += _res_cap[ta];
1277 1482
                  if (ahead >= delta) break;
1278 1483
                }
1279 1484
                if (ahead < 0) ahead = 0;
1280 1485

	
1281 1486
                // Push flow along the arc
1282 1487
                if (ahead < delta && !hyper[t]) {
1283 1488
                  _res_cap[a] -= ahead;
1284 1489
                  _res_cap[_reverse[a]] += ahead;
1285 1490
                  _excess[n] -= ahead;
1286 1491
                  _excess[t] += ahead;
1287 1492
                  _active_nodes.push_front(t);
1288 1493
                  hyper[t] = true;
1289 1494
                  hyper_cost[t] = _cost[a] + pi_n - pi_t;
1290 1495
                  _next_out[n] = a;
1291 1496
                  goto next_node;
1292 1497
                } else {
1293 1498
                  _res_cap[a] -= delta;
1294 1499
                  _res_cap[_reverse[a]] += delta;
1295 1500
                  _excess[n] -= delta;
1296 1501
                  _excess[t] += delta;
1297 1502
                  if (_excess[t] > 0 && _excess[t] <= delta)
1298 1503
                    _active_nodes.push_back(t);
1299 1504
                }
1300 1505

	
1301 1506
                if (_excess[n] == 0) {
1302 1507
                  _next_out[n] = a;
1303 1508
                  goto remove_nodes;
1304 1509
                }
1305 1510
              }
1306 1511
            }
1307 1512
            _next_out[n] = a;
1308 1513
          }
1309 1514

	
1310 1515
          // Relabel the node if it is still active (or hyper)
1311 1516
          if (_excess[n] > 0 || hyper[n]) {
1312 1517
             min_red_cost = hyper[n] ? -hyper_cost[n] :
1313 1518
               std::numeric_limits<LargeCost>::max();
1314 1519
            for (int a = _first_out[n]; a != last_out; ++a) {
1315 1520
              if (_res_cap[a] > 0) {
1316 1521
                rc = _cost[a] + pi_n - _pi[_target[a]];
1317 1522
                if (rc < min_red_cost) {
1318 1523
                  min_red_cost = rc;
1319 1524
                }
1320 1525
              }
1321 1526
            }
1322 1527
            _pi[n] -= min_red_cost + _epsilon;
1323 1528
            _next_out[n] = _first_out[n];
1324 1529
            hyper[n] = false;
1325 1530
            ++relabel_cnt;
1326 1531
          }
1327 1532

	
1328 1533
          // Remove nodes that are not active nor hyper
1329 1534
        remove_nodes:
1330 1535
          while ( _active_nodes.size() > 0 &&
1331 1536
                  _excess[_active_nodes.front()] <= 0 &&
1332 1537
                  !hyper[_active_nodes.front()] ) {
1333 1538
            _active_nodes.pop_front();
1334 1539
          }
1335 1540

	
1336 1541
          // Global update heuristic
1337 1542
          if (relabel_cnt >= next_global_update_limit) {
1338 1543
            globalUpdate();
1339 1544
            for (int u = 0; u != _res_node_num; ++u)
0 comments (0 inline)