gravatar
kpeter (Peter Kovacs)
kpeter@inf.elte.hu
Rework and improve Suurballe (#323) - Improve the implementation: use a specific, faster variant of residual Dijkstra for the first search. - Some reorganizatiopn to make the code simpler. - Small doc improvements.
0 1 0
default
1 file changed with 110 insertions and 74 deletions:
↑ Collapse diff ↑
Ignore white space 6 line context
... ...
@@ -46,7 +46,7 @@
46 46
  /// Note that this problem is a special case of the \ref min_cost_flow
47 47
  /// "minimum cost flow problem". This implementation is actually an
48 48
  /// efficient specialized version of the \ref CapacityScaling
49
  /// "Successive Shortest Path" algorithm directly for this problem.
49
  /// "successive shortest path" algorithm directly for this problem.
50 50
  /// Therefore this class provides query functions for flow values and
51 51
  /// node potentials (the dual solution) just like the minimum cost flow
52 52
  /// algorithms.
... ...
@@ -57,7 +57,7 @@
57 57
  ///
58 58
  /// \warning Length values should be \e non-negative.
59 59
  ///
60
  /// \note For finding node-disjoint paths this algorithm can be used
60
  /// \note For finding \e node-disjoint paths, this algorithm can be used
61 61
  /// along with the \ref SplitNodes adaptor.
62 62
#ifdef DOXYGEN
63 63
  template <typename GR, typename LEN>
... ...
@@ -109,39 +109,36 @@
109 109

	
110 110
    private:
111 111

	
112
      // The digraph the algorithm runs on
113 112
      const Digraph &_graph;
114

	
115
      // The main maps
113
      const LengthMap &_length;
116 114
      const FlowMap &_flow;
117
      const LengthMap &_length;
118
      PotentialMap &_potential;
119

	
120
      // The distance map
121
      PotentialMap _dist;
122
      // The pred arc map
115
      PotentialMap &_pi;
123 116
      PredMap &_pred;
124
      // The processed (i.e. permanently labeled) nodes
125
      std::vector<Node> _proc_nodes;
126

	
127 117
      Node _s;
128 118
      Node _t;
119
      
120
      PotentialMap _dist;
121
      std::vector<Node> _proc_nodes;
129 122

	
130 123
    public:
131 124

	
132
      /// Constructor.
133
      ResidualDijkstra( const Digraph &graph,
134
                        const FlowMap &flow,
135
                        const LengthMap &length,
136
                        PotentialMap &potential,
137
                        PredMap &pred,
138
                        Node s, Node t ) :
139
        _graph(graph), _flow(flow), _length(length), _potential(potential),
140
        _dist(graph), _pred(pred), _s(s), _t(t) {}
125
      // Constructor
126
      ResidualDijkstra(Suurballe &srb) :
127
        _graph(srb._graph), _length(srb._length),
128
        _flow(*srb._flow), _pi(*srb._potential), _pred(srb._pred), 
129
        _s(srb._s), _t(srb._t), _dist(_graph) {}
130
        
131
      // Run the algorithm and return true if a path is found
132
      // from the source node to the target node.
133
      bool run(int cnt) {
134
        return cnt == 0 ? startFirst() : start();
135
      }
141 136

	
142
      /// \brief Run the algorithm. It returns \c true if a path is found
143
      /// from the source node to the target node.
144
      bool run() {
137
    private:
138
    
139
      // Execute the algorithm for the first time (the flow and potential
140
      // functions have to be identically zero).
141
      bool startFirst() {
145 142
        HeapCrossRef heap_cross_ref(_graph, Heap::PRE_HEAP);
146 143
        Heap heap(heap_cross_ref);
147 144
        heap.push(_s, 0);
... ...
@@ -151,29 +148,74 @@
151 148
        // Process nodes
152 149
        while (!heap.empty() && heap.top() != _t) {
153 150
          Node u = heap.top(), v;
154
          Length d = heap.prio() + _potential[u], nd;
151
          Length d = heap.prio(), dn;
155 152
          _dist[u] = heap.prio();
153
          _proc_nodes.push_back(u);
156 154
          heap.pop();
155

	
156
          // Traverse outgoing arcs
157
          for (OutArcIt e(_graph, u); e != INVALID; ++e) {
158
            v = _graph.target(e);
159
            switch(heap.state(v)) {
160
              case Heap::PRE_HEAP:
161
                heap.push(v, d + _length[e]);
162
                _pred[v] = e;
163
                break;
164
              case Heap::IN_HEAP:
165
                dn = d + _length[e];
166
                if (dn < heap[v]) {
167
                  heap.decrease(v, dn);
168
                  _pred[v] = e;
169
                }
170
                break;
171
              case Heap::POST_HEAP:
172
                break;
173
            }
174
          }
175
        }
176
        if (heap.empty()) return false;
177

	
178
        // Update potentials of processed nodes
179
        Length t_dist = heap.prio();
180
        for (int i = 0; i < int(_proc_nodes.size()); ++i)
181
          _pi[_proc_nodes[i]] = _dist[_proc_nodes[i]] - t_dist;
182
        return true;
183
      }
184

	
185
      // Execute the algorithm.
186
      bool start() {
187
        HeapCrossRef heap_cross_ref(_graph, Heap::PRE_HEAP);
188
        Heap heap(heap_cross_ref);
189
        heap.push(_s, 0);
190
        _pred[_s] = INVALID;
191
        _proc_nodes.clear();
192

	
193
        // Process nodes
194
        while (!heap.empty() && heap.top() != _t) {
195
          Node u = heap.top(), v;
196
          Length d = heap.prio() + _pi[u], dn;
197
          _dist[u] = heap.prio();
157 198
          _proc_nodes.push_back(u);
199
          heap.pop();
158 200

	
159 201
          // Traverse outgoing arcs
160 202
          for (OutArcIt e(_graph, u); e != INVALID; ++e) {
161 203
            if (_flow[e] == 0) {
162 204
              v = _graph.target(e);
163 205
              switch(heap.state(v)) {
164
              case Heap::PRE_HEAP:
165
                heap.push(v, d + _length[e] - _potential[v]);
166
                _pred[v] = e;
167
                break;
168
              case Heap::IN_HEAP:
169
                nd = d + _length[e] - _potential[v];
170
                if (nd < heap[v]) {
171
                  heap.decrease(v, nd);
206
                case Heap::PRE_HEAP:
207
                  heap.push(v, d + _length[e] - _pi[v]);
172 208
                  _pred[v] = e;
173
                }
174
                break;
175
              case Heap::POST_HEAP:
176
                break;
209
                  break;
210
                case Heap::IN_HEAP:
211
                  dn = d + _length[e] - _pi[v];
212
                  if (dn < heap[v]) {
213
                    heap.decrease(v, dn);
214
                    _pred[v] = e;
215
                  }
216
                  break;
217
                case Heap::POST_HEAP:
218
                  break;
177 219
              }
178 220
            }
179 221
          }
... ...
@@ -183,19 +225,19 @@
183 225
            if (_flow[e] == 1) {
184 226
              v = _graph.source(e);
185 227
              switch(heap.state(v)) {
186
              case Heap::PRE_HEAP:
187
                heap.push(v, d - _length[e] - _potential[v]);
188
                _pred[v] = e;
189
                break;
190
              case Heap::IN_HEAP:
191
                nd = d - _length[e] - _potential[v];
192
                if (nd < heap[v]) {
193
                  heap.decrease(v, nd);
228
                case Heap::PRE_HEAP:
229
                  heap.push(v, d - _length[e] - _pi[v]);
194 230
                  _pred[v] = e;
195
                }
196
                break;
197
              case Heap::POST_HEAP:
198
                break;
231
                  break;
232
                case Heap::IN_HEAP:
233
                  dn = d - _length[e] - _pi[v];
234
                  if (dn < heap[v]) {
235
                    heap.decrease(v, dn);
236
                    _pred[v] = e;
237
                  }
238
                  break;
239
                case Heap::POST_HEAP:
240
                  break;
199 241
              }
200 242
            }
201 243
          }
... ...
@@ -205,7 +247,7 @@
205 247
        // Update potentials of processed nodes
206 248
        Length t_dist = heap.prio();
207 249
        for (int i = 0; i < int(_proc_nodes.size()); ++i)
208
          _potential[_proc_nodes[i]] += _dist[_proc_nodes[i]] - t_dist;
250
          _pi[_proc_nodes[i]] += _dist[_proc_nodes[i]] - t_dist;
209 251
        return true;
210 252
      }
211 253

	
... ...
@@ -226,19 +268,16 @@
226 268
    bool _local_potential;
227 269

	
228 270
    // The source node
229
    Node _source;
271
    Node _s;
230 272
    // The target node
231
    Node _target;
273
    Node _t;
232 274

	
233 275
    // Container to store the found paths
234
    std::vector< SimplePath<Digraph> > paths;
276
    std::vector<Path> _paths;
235 277
    int _path_num;
236 278

	
237 279
    // The pred arc map
238 280
    PredMap _pred;
239
    // Implementation of the Dijkstra algorithm for finding augmenting
240
    // shortest paths in the residual network
241
    ResidualDijkstra *_dijkstra;
242 281

	
243 282
  public:
244 283

	
... ...
@@ -258,7 +297,6 @@
258 297
    ~Suurballe() {
259 298
      if (_local_flow) delete _flow;
260 299
      if (_local_potential) delete _potential;
261
      delete _dijkstra;
262 300
    }
263 301

	
264 302
    /// \brief Set the flow map.
... ...
@@ -342,7 +380,7 @@
342 380
    ///
343 381
    /// \param s The source node.
344 382
    void init(const Node& s) {
345
      _source = s;
383
      _s = s;
346 384

	
347 385
      // Initialize maps
348 386
      if (!_flow) {
... ...
@@ -372,20 +410,18 @@
372 410
    ///
373 411
    /// \pre \ref init() must be called before using this function.
374 412
    int findFlow(const Node& t, int k = 2) {
375
      _target = t;
376
      _dijkstra =
377
        new ResidualDijkstra( _graph, *_flow, _length, *_potential, _pred,
378
                              _source, _target );
413
      _t = t;
414
      ResidualDijkstra dijkstra(*this);
379 415

	
380 416
      // Find shortest paths
381 417
      _path_num = 0;
382 418
      while (_path_num < k) {
383 419
        // Run Dijkstra
384
        if (!_dijkstra->run()) break;
420
        if (!dijkstra.run(_path_num)) break;
385 421
        ++_path_num;
386 422

	
387 423
        // Set the flow along the found shortest path
388
        Node u = _target;
424
        Node u = _t;
389 425
        Arc e;
390 426
        while ((e = _pred[u]) != INVALID) {
391 427
          if (u == _graph.target(e)) {
... ...
@@ -402,8 +438,8 @@
402 438

	
403 439
    /// \brief Compute the paths from the flow.
404 440
    ///
405
    /// This function computes the paths from the found minimum cost flow,
406
    /// which is the union of some arc-disjoint paths.
441
    /// This function computes arc-disjoint paths from the found minimum
442
    /// cost flow, which is the union of them.
407 443
    ///
408 444
    /// \pre \ref init() and \ref findFlow() must be called before using
409 445
    /// this function.
... ...
@@ -411,15 +447,15 @@
411 447
      FlowMap res_flow(_graph);
412 448
      for(ArcIt a(_graph); a != INVALID; ++a) res_flow[a] = (*_flow)[a];
413 449

	
414
      paths.clear();
415
      paths.resize(_path_num);
450
      _paths.clear();
451
      _paths.resize(_path_num);
416 452
      for (int i = 0; i < _path_num; ++i) {
417
        Node n = _source;
418
        while (n != _target) {
453
        Node n = _s;
454
        while (n != _t) {
419 455
          OutArcIt e(_graph, n);
420 456
          for ( ; res_flow[e] == 0; ++e) ;
421 457
          n = _graph.target(e);
422
          paths[i].addBack(e);
458
          _paths[i].addBack(e);
423 459
          res_flow[e] = 0;
424 460
        }
425 461
      }
... ...
@@ -518,7 +554,7 @@
518 554
    /// \pre \ref run() or \ref findPaths() must be called before using
519 555
    /// this function.
520 556
    const Path& path(int i) const {
521
      return paths[i];
557
      return _paths[i];
522 558
    }
523 559

	
524 560
    /// @}
0 comments (0 inline)