author Peter Kovacs Fri, 16 Oct 2009 01:06:16 +0200 changeset 853 ec0b1b423b8b parent 852 30c77d1c0cba child 854 9a7e4e606f83
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.
 lemon/suurballe.h file | annotate | diff | comparison | revisions
     1.1 --- a/lemon/suurballe.h	Thu Oct 15 21:04:50 2009 +0200
1.2 +++ b/lemon/suurballe.h	Fri Oct 16 01:06:16 2009 +0200
1.3 @@ -46,7 +46,7 @@
1.4    /// Note that this problem is a special case of the \ref min_cost_flow
1.5    /// "minimum cost flow problem". This implementation is actually an
1.6    /// efficient specialized version of the \ref CapacityScaling
1.7 -  /// "Successive Shortest Path" algorithm directly for this problem.
1.8 +  /// "successive shortest path" algorithm directly for this problem.
1.9    /// Therefore this class provides query functions for flow values and
1.10    /// node potentials (the dual solution) just like the minimum cost flow
1.11    /// algorithms.
1.12 @@ -57,7 +57,7 @@
1.13    ///
1.14    /// \warning Length values should be \e non-negative.
1.15    ///
1.16 -  /// \note For finding node-disjoint paths this algorithm can be used
1.17 +  /// \note For finding \e node-disjoint paths, this algorithm can be used
1.18    /// along with the \ref SplitNodes adaptor.
1.19  #ifdef DOXYGEN
1.20    template <typename GR, typename LEN>
1.21 @@ -109,39 +109,36 @@
1.22
1.23      private:
1.24
1.25 -      // The digraph the algorithm runs on
1.26        const Digraph &_graph;
1.27 -
1.28 -      // The main maps
1.29 +      const LengthMap &_length;
1.30        const FlowMap &_flow;
1.31 -      const LengthMap &_length;
1.32 -      PotentialMap &_potential;
1.33 -
1.34 -      // The distance map
1.35 -      PotentialMap _dist;
1.36 -      // The pred arc map
1.37 +      PotentialMap &_pi;
1.38        PredMap &_pred;
1.39 -      // The processed (i.e. permanently labeled) nodes
1.40 -      std::vector<Node> _proc_nodes;
1.41 -
1.42        Node _s;
1.43        Node _t;
1.44 +
1.45 +      PotentialMap _dist;
1.46 +      std::vector<Node> _proc_nodes;
1.47
1.48      public:
1.49
1.50 -      /// Constructor.
1.51 -      ResidualDijkstra( const Digraph &graph,
1.52 -                        const FlowMap &flow,
1.53 -                        const LengthMap &length,
1.54 -                        PotentialMap &potential,
1.55 -                        PredMap &pred,
1.56 -                        Node s, Node t ) :
1.57 -        _graph(graph), _flow(flow), _length(length), _potential(potential),
1.58 -        _dist(graph), _pred(pred), _s(s), _t(t) {}
1.59 +      // Constructor
1.60 +      ResidualDijkstra(Suurballe &srb) :
1.61 +        _graph(srb._graph), _length(srb._length),
1.62 +        _flow(*srb._flow), _pi(*srb._potential), _pred(srb._pred),
1.63 +        _s(srb._s), _t(srb._t), _dist(_graph) {}
1.64 +
1.65 +      // Run the algorithm and return true if a path is found
1.66 +      // from the source node to the target node.
1.67 +      bool run(int cnt) {
1.68 +        return cnt == 0 ? startFirst() : start();
1.69 +      }
1.70
1.71 -      /// \brief Run the algorithm. It returns \c true if a path is found
1.72 -      /// from the source node to the target node.
1.73 -      bool run() {
1.74 +    private:
1.75 +
1.76 +      // Execute the algorithm for the first time (the flow and potential
1.77 +      // functions have to be identically zero).
1.78 +      bool startFirst() {
1.79          HeapCrossRef heap_cross_ref(_graph, Heap::PRE_HEAP);
1.80          Heap heap(heap_cross_ref);
1.81          heap.push(_s, 0);
1.82 @@ -151,29 +148,74 @@
1.83          // Process nodes
1.84          while (!heap.empty() && heap.top() != _t) {
1.85            Node u = heap.top(), v;
1.86 -          Length d = heap.prio() + _potential[u], nd;
1.87 +          Length d = heap.prio(), dn;
1.88            _dist[u] = heap.prio();
1.89 +          _proc_nodes.push_back(u);
1.90            heap.pop();
1.91 +
1.92 +          // Traverse outgoing arcs
1.93 +          for (OutArcIt e(_graph, u); e != INVALID; ++e) {
1.94 +            v = _graph.target(e);
1.95 +            switch(heap.state(v)) {
1.96 +              case Heap::PRE_HEAP:
1.97 +                heap.push(v, d + _length[e]);
1.98 +                _pred[v] = e;
1.99 +                break;
1.100 +              case Heap::IN_HEAP:
1.101 +                dn = d + _length[e];
1.102 +                if (dn < heap[v]) {
1.103 +                  heap.decrease(v, dn);
1.104 +                  _pred[v] = e;
1.105 +                }
1.106 +                break;
1.107 +              case Heap::POST_HEAP:
1.108 +                break;
1.109 +            }
1.110 +          }
1.111 +        }
1.112 +        if (heap.empty()) return false;
1.113 +
1.114 +        // Update potentials of processed nodes
1.115 +        Length t_dist = heap.prio();
1.116 +        for (int i = 0; i < int(_proc_nodes.size()); ++i)
1.117 +          _pi[_proc_nodes[i]] = _dist[_proc_nodes[i]] - t_dist;
1.118 +        return true;
1.119 +      }
1.120 +
1.121 +      // Execute the algorithm.
1.122 +      bool start() {
1.123 +        HeapCrossRef heap_cross_ref(_graph, Heap::PRE_HEAP);
1.124 +        Heap heap(heap_cross_ref);
1.125 +        heap.push(_s, 0);
1.126 +        _pred[_s] = INVALID;
1.127 +        _proc_nodes.clear();
1.128 +
1.129 +        // Process nodes
1.130 +        while (!heap.empty() && heap.top() != _t) {
1.131 +          Node u = heap.top(), v;
1.132 +          Length d = heap.prio() + _pi[u], dn;
1.133 +          _dist[u] = heap.prio();
1.134            _proc_nodes.push_back(u);
1.135 +          heap.pop();
1.136
1.137            // Traverse outgoing arcs
1.138            for (OutArcIt e(_graph, u); e != INVALID; ++e) {
1.139              if (_flow[e] == 0) {
1.140                v = _graph.target(e);
1.141                switch(heap.state(v)) {
1.142 -              case Heap::PRE_HEAP:
1.143 -                heap.push(v, d + _length[e] - _potential[v]);
1.144 -                _pred[v] = e;
1.145 -                break;
1.146 -              case Heap::IN_HEAP:
1.147 -                nd = d + _length[e] - _potential[v];
1.148 -                if (nd < heap[v]) {
1.149 -                  heap.decrease(v, nd);
1.150 +                case Heap::PRE_HEAP:
1.151 +                  heap.push(v, d + _length[e] - _pi[v]);
1.152                    _pred[v] = e;
1.153 -                }
1.154 -                break;
1.155 -              case Heap::POST_HEAP:
1.156 -                break;
1.157 +                  break;
1.158 +                case Heap::IN_HEAP:
1.159 +                  dn = d + _length[e] - _pi[v];
1.160 +                  if (dn < heap[v]) {
1.161 +                    heap.decrease(v, dn);
1.162 +                    _pred[v] = e;
1.163 +                  }
1.164 +                  break;
1.165 +                case Heap::POST_HEAP:
1.166 +                  break;
1.167                }
1.168              }
1.169            }
1.170 @@ -183,19 +225,19 @@
1.171              if (_flow[e] == 1) {
1.172                v = _graph.source(e);
1.173                switch(heap.state(v)) {
1.174 -              case Heap::PRE_HEAP:
1.175 -                heap.push(v, d - _length[e] - _potential[v]);
1.176 -                _pred[v] = e;
1.177 -                break;
1.178 -              case Heap::IN_HEAP:
1.179 -                nd = d - _length[e] - _potential[v];
1.180 -                if (nd < heap[v]) {
1.181 -                  heap.decrease(v, nd);
1.182 +                case Heap::PRE_HEAP:
1.183 +                  heap.push(v, d - _length[e] - _pi[v]);
1.184                    _pred[v] = e;
1.185 -                }
1.186 -                break;
1.187 -              case Heap::POST_HEAP:
1.188 -                break;
1.189 +                  break;
1.190 +                case Heap::IN_HEAP:
1.191 +                  dn = d - _length[e] - _pi[v];
1.192 +                  if (dn < heap[v]) {
1.193 +                    heap.decrease(v, dn);
1.194 +                    _pred[v] = e;
1.195 +                  }
1.196 +                  break;
1.197 +                case Heap::POST_HEAP:
1.198 +                  break;
1.199                }
1.200              }
1.201            }
1.202 @@ -205,7 +247,7 @@
1.203          // Update potentials of processed nodes
1.204          Length t_dist = heap.prio();
1.205          for (int i = 0; i < int(_proc_nodes.size()); ++i)
1.206 -          _potential[_proc_nodes[i]] += _dist[_proc_nodes[i]] - t_dist;
1.207 +          _pi[_proc_nodes[i]] += _dist[_proc_nodes[i]] - t_dist;
1.208          return true;
1.209        }
1.210
1.211 @@ -226,19 +268,16 @@
1.212      bool _local_potential;
1.213
1.214      // The source node
1.215 -    Node _source;
1.216 +    Node _s;
1.217      // The target node
1.218 -    Node _target;
1.219 +    Node _t;
1.220
1.221      // Container to store the found paths
1.222 -    std::vector< SimplePath<Digraph> > paths;
1.223 +    std::vector<Path> _paths;
1.224      int _path_num;
1.225
1.226      // The pred arc map
1.227      PredMap _pred;
1.228 -    // Implementation of the Dijkstra algorithm for finding augmenting
1.229 -    // shortest paths in the residual network
1.230 -    ResidualDijkstra *_dijkstra;
1.231
1.232    public:
1.233
1.234 @@ -258,7 +297,6 @@
1.235      ~Suurballe() {
1.236        if (_local_flow) delete _flow;
1.237        if (_local_potential) delete _potential;
1.238 -      delete _dijkstra;
1.239      }
1.240
1.241      /// \brief Set the flow map.
1.242 @@ -342,7 +380,7 @@
1.243      ///
1.244      /// \param s The source node.
1.245      void init(const Node& s) {
1.246 -      _source = s;
1.247 +      _s = s;
1.248
1.249        // Initialize maps
1.250        if (!_flow) {
1.251 @@ -372,20 +410,18 @@
1.252      ///
1.253      /// \pre \ref init() must be called before using this function.
1.254      int findFlow(const Node& t, int k = 2) {
1.255 -      _target = t;
1.256 -      _dijkstra =
1.257 -        new ResidualDijkstra( _graph, *_flow, _length, *_potential, _pred,
1.258 -                              _source, _target );
1.259 +      _t = t;
1.260 +      ResidualDijkstra dijkstra(*this);
1.261
1.262        // Find shortest paths
1.263        _path_num = 0;
1.264        while (_path_num < k) {
1.265          // Run Dijkstra
1.266 -        if (!_dijkstra->run()) break;
1.267 +        if (!dijkstra.run(_path_num)) break;
1.268          ++_path_num;
1.269
1.270          // Set the flow along the found shortest path
1.271 -        Node u = _target;
1.272 +        Node u = _t;
1.273          Arc e;
1.274          while ((e = _pred[u]) != INVALID) {
1.275            if (u == _graph.target(e)) {
1.276 @@ -402,8 +438,8 @@
1.277
1.278      /// \brief Compute the paths from the flow.
1.279      ///
1.280 -    /// This function computes the paths from the found minimum cost flow,
1.281 -    /// which is the union of some arc-disjoint paths.
1.282 +    /// This function computes arc-disjoint paths from the found minimum
1.283 +    /// cost flow, which is the union of them.
1.284      ///
1.285      /// \pre \ref init() and \ref findFlow() must be called before using
1.286      /// this function.
1.287 @@ -411,15 +447,15 @@
1.288        FlowMap res_flow(_graph);
1.289        for(ArcIt a(_graph); a != INVALID; ++a) res_flow[a] = (*_flow)[a];
1.290
1.291 -      paths.clear();
1.292 -      paths.resize(_path_num);
1.293 +      _paths.clear();
1.294 +      _paths.resize(_path_num);
1.295        for (int i = 0; i < _path_num; ++i) {
1.296 -        Node n = _source;
1.297 -        while (n != _target) {
1.298 +        Node n = _s;
1.299 +        while (n != _t) {
1.300            OutArcIt e(_graph, n);
1.301            for ( ; res_flow[e] == 0; ++e) ;
1.302            n = _graph.target(e);