src/work/athos/mincostflow.h
author alpar
Thu, 22 Jul 2004 20:06:40 +0000
changeset 733 240003bddaff
parent 671 708df4dc6ab6
child 921 818510fa3d99
permissions -rw-r--r--
Check StaticGraphSkeleton, as well.
     1 // -*- c++ -*-
     2 #ifndef HUGO_MINCOSTFLOW_H
     3 #define HUGO_MINCOSTFLOW_H
     4 
     5 ///\ingroup galgs
     6 ///\file
     7 ///\brief An algorithm for finding the minimum cost flow of given value in an uncapacitated network
     8 
     9 #include <hugo/dijkstra.h>
    10 #include <hugo/graph_wrapper.h>
    11 #include <hugo/maps.h>
    12 #include <vector>
    13 #include <list>
    14 #include <values.h>
    15 #include <hugo/for_each_macros.h>
    16 #include <hugo/unionfind.h>
    17 #include <hugo/bin_heap.h>
    18 #include <bfs_dfs.h>
    19 
    20 namespace hugo {
    21 
    22 /// \addtogroup galgs
    23 /// @{
    24 
    25   ///\brief Implementation of an algorithm for solving the minimum cost general
    26   /// flow problem in an uncapacitated network
    27   /// 
    28   ///
    29   /// The class \ref hugo::MinCostFlow "MinCostFlow" implements
    30   /// an algorithm for solving the following general minimum cost flow problem>
    31   /// 
    32   ///
    33   ///
    34   /// \warning It is assumed here that the problem has a feasible solution
    35   ///
    36   /// The range of the cost (weight) function is nonnegative reals but 
    37   /// the range of capacity function is the set of nonnegative integers. 
    38   /// It is not a polinomial time algorithm for counting the minimum cost
    39   /// maximal flow, since it counts the minimum cost flow for every value 0..M
    40   /// where \c M is the value of the maximal flow.
    41   ///
    42   ///\author Attila Bernath
    43   template <typename Graph, typename CostMap, typename SupplyDemandMap>
    44   class MinCostFlow {
    45 
    46     typedef typename CostMap::ValueType Cost;
    47 
    48 
    49     typedef typename SupplyDemandMap::ValueType SupplyDemand;
    50     
    51     typedef typename Graph::Node Node;
    52     typedef typename Graph::NodeIt NodeIt;
    53     typedef typename Graph::Edge Edge;
    54     typedef typename Graph::OutEdgeIt OutEdgeIt;
    55     typedef typename Graph::template EdgeMap<SupplyDemand> FlowMap;
    56     typedef ConstMap<Edge,SupplyDemand> ConstEdgeMap;
    57 
    58     //    typedef ConstMap<Edge,int> ConstMap;
    59 
    60     typedef ResGraphWrapper<const Graph,int,ConstEdgeMap,FlowMap> ResGraph;
    61     typedef typename ResGraph::Edge ResGraphEdge;
    62 
    63     class ModCostMap {   
    64       //typedef typename ResGraph::template NodeMap<Cost> NodeMap;
    65       typedef typename Graph::template NodeMap<Cost> NodeMap;
    66       const ResGraph& res_graph;
    67       //      const EdgeIntMap& rev;
    68       const CostMap &ol;
    69       const NodeMap &pot;
    70     public :
    71       typedef typename CostMap::KeyType KeyType;
    72       typedef typename CostMap::ValueType ValueType;
    73 	
    74       ValueType operator[](typename ResGraph::Edge e) const {     
    75 	if (res_graph.forward(e))
    76 	  return  ol[e]-(pot[res_graph.head(e)]-pot[res_graph.tail(e)]);   
    77 	else
    78 	  return -ol[e]-(pot[res_graph.head(e)]-pot[res_graph.tail(e)]);   
    79       }     
    80 	
    81       ModCostMap(const ResGraph& _res_graph,
    82 		   const CostMap &o,  const NodeMap &p) : 
    83 	res_graph(_res_graph), /*rev(_rev),*/ ol(o), pot(p){}; 
    84     };//ModCostMap
    85 
    86 
    87   protected:
    88     
    89     //Input
    90     const Graph& graph;
    91     const CostMap& cost;
    92     const SupplyDemandMap& supply_demand;//supply or demand of nodes
    93 
    94 
    95     //auxiliary variables
    96 
    97     //To store the flow
    98     FlowMap flow; 
    99     //To store the potential (dual variables)
   100     typedef typename Graph::template NodeMap<Cost> PotentialMap;
   101     PotentialMap potential;
   102     
   103 
   104     Cost total_cost;
   105 
   106 
   107   public :
   108 
   109 
   110    MinCostFlow(Graph& _graph, CostMap& _cost, SupplyDemandMap& _supply_demand):
   111      graph(_graph), 
   112      cost(_cost), 
   113      supply_demand(_supply_demand), 
   114      flow(_graph), 
   115      potential(_graph){ }
   116 
   117     
   118     ///Runs the algorithm.
   119 
   120     ///Runs the algorithm.
   121 
   122     ///\todo May be it does make sense to be able to start with a nonzero 
   123     /// feasible primal-dual solution pair as well.
   124     void run() {
   125 
   126       //To store excess-deficit values
   127       SupplyDemandMap excess_deficit(graph);
   128 
   129       //Resetting variables from previous runs
   130       //total_cost = 0;
   131 
   132 
   133       typedef typename Graph::template NodeMap<int> HeapMap;
   134       typedef BinHeap< Node, SupplyDemand, typename Graph::template NodeMap<int>,
   135 	std::greater<SupplyDemand> > 	HeapType;
   136 
   137       //A heap for the excess nodes
   138       HeapMap excess_nodes_map(graph,-1);
   139       HeapType excess_nodes(excess_nodes_map);
   140 
   141       //A heap for the deficit nodes
   142       HeapMap deficit_nodes_map(graph,-1);
   143       HeapType deficit_nodes(deficit_nodes_map);
   144 
   145       //A container to store nonabundant arcs
   146       std::list<Edge> nonabundant_arcs;
   147 
   148 	
   149       FOR_EACH_LOC(typename Graph::EdgeIt, e, graph){
   150 	flow.set(e,0);
   151 	nonabundant_arcs.push_back(e);
   152       }
   153 
   154       //Initial value for delta
   155       SupplyDemand delta = 0;
   156 
   157       typedef UnionFindEnum<Node, Graph::template NodeMap> UFE;
   158 
   159       //A union-find structure to store the abundant components
   160       typename UFE::MapType abund_comp_map(graph);
   161       UFE abundant_components(abund_comp_map);
   162 
   163 
   164 
   165       FOR_EACH_LOC(typename Graph::NodeIt, n, graph){
   166        	excess_deficit.set(n,supply_demand[n]);
   167 	//A supply node
   168 	if (excess_deficit[n] > 0){
   169 	  excess_nodes.push(n,excess_deficit[n]);
   170 	}
   171 	//A demand node
   172 	if (excess_deficit[n] < 0){
   173 	  deficit_nodes.push(n, - excess_deficit[n]);
   174 	}
   175 	//Finding out starting value of delta
   176 	if (delta < abs(excess_deficit[n])){
   177 	  delta = abs(excess_deficit[n]);
   178 	}
   179 	//Initialize the copy of the Dijkstra potential to zero
   180 	potential.set(n,0);
   181 	//Every single point is an abundant component initially 
   182 	abundant_components.insert(n);
   183       }
   184 
   185       //It'll be allright as an initial value, though this value 
   186       //can be the maximum deficit here
   187       SupplyDemand max_excess = delta;
   188       
   189       ///\bug This is a serious cheat here, before we have an uncapacitated ResGraph
   190       ConstEdgeMap const_inf_map(MAXINT);
   191       
   192       //We need a residual graph which is uncapacitated
   193       ResGraph res_graph(graph, const_inf_map, flow);
   194       
   195       //An EdgeMap to tell which arcs are abundant
   196       typename Graph::template EdgeMap<bool> abundant_arcs(graph);
   197 
   198       //Let's construct the sugraph consisting only of the abundant edges
   199       typedef ConstMap< typename Graph::Node, bool > ConstNodeMap;
   200 
   201       ConstNodeMap const_true_map(true);
   202       typedef SubGraphWrapper< const Graph, ConstNodeMap, 
   203 	 typename Graph::template EdgeMap<bool> > 
   204 	AbundantGraph;
   205       AbundantGraph abundant_graph(graph, const_true_map, abundant_arcs );
   206       
   207       //Let's construct the residual graph for the abundant graph
   208       typedef ResGraphWrapper<const AbundantGraph,int,ConstEdgeMap,FlowMap> 
   209 	ResAbGraph;
   210       //Again uncapacitated
   211       ResAbGraph res_ab_graph(abundant_graph, const_inf_map, flow);
   212       
   213       //We need things for the bfs
   214       typename ResAbGraph::template NodeMap<bool> bfs_reached(res_ab_graph);
   215       typename ResAbGraph::template NodeMap<typename ResAbGraph::Edge> 
   216 	bfs_pred(res_ab_graph); 
   217       NullMap<typename ResAbGraph::Node, int> bfs_dist_dummy;
   218       //Teszt celbol:
   219       //BfsIterator<ResAbGraph, typename ResAbGraph::template NodeMap<bool> > 
   220       //izebize(res_ab_graph, bfs_reached);
   221       //We want to run bfs-es (more) on this graph 'res_ab_graph'
   222       Bfs < const ResAbGraph , 
   223 	typename ResAbGraph::template NodeMap<bool>, 
   224 	typename ResAbGraph::template NodeMap<typename ResAbGraph::Edge>,
   225 	NullMap<typename ResAbGraph::Node, int> > 
   226 	bfs(res_ab_graph, bfs_reached, bfs_pred, bfs_dist_dummy);
   227       /*This is what Marci wants for a bfs
   228 	template <typename Graph, 
   229 	    typename ReachedMap=typename Graph::template NodeMap<bool>, 
   230 	    typename PredMap
   231 	    =typename Graph::template NodeMap<typename Graph::Edge>, 
   232 	    typename DistMap=typename Graph::template NodeMap<int> > 
   233 	    class Bfs : public BfsIterator<Graph, ReachedMap> {
   234 
   235        */
   236       
   237       ModCostMap mod_cost(res_graph, cost, potential);
   238 
   239       Dijkstra<ResGraph, ModCostMap> dijkstra(res_graph, mod_cost);
   240 
   241       //We will use the number of the nodes of the graph often
   242       int number_of_nodes = graph.nodeNum();
   243 
   244       while (max_excess > 0){
   245 
   246 	//Reset delta if still too big
   247 	if (8*number_of_nodes*max_excess <= delta){
   248 	  delta = max_excess;
   249 	  
   250 	}
   251 
   252 	/*
   253 	 * Beginning of the delta scaling phase 
   254 	*/
   255 	//Merge and stuff
   256 	{
   257 	  SupplyDemand buf=8*number_of_nodes*delta;
   258 	  typename std::list<Edge>::iterator i = nonabundant_arcs.begin();
   259 	  while ( i != nonabundant_arcs.end() ){
   260 	    if (flow[*i]>=buf){
   261 	      Node a = abundant_components.find(res_graph.head(*i));
   262 	      Node b = abundant_components.find(res_graph.tail(*i));
   263 	      //Merge
   264 	      if (a != b){
   265 		abundant_components.join(a,b);
   266 		//We want to push the smaller
   267 		//Which has greater absolut value excess/deficit
   268 		Node root=(abs(excess_deficit[a])>abs(excess_deficit[b]))?a:b;
   269 		//Which is the other
   270 		Node non_root = ( a == root ) ? b : a ;
   271 		abundant_components.makeRep(root);
   272 		SupplyDemand qty_to_augment = abs(excess_deficit[non_root]); 
   273 		//Push the positive value
   274 		if (excess_deficit[non_root] < 0)
   275 		  swap(root, non_root);
   276 		//If the non_root node has excess/deficit at all
   277 		if (qty_to_augment>0){
   278 		  //Find path and augment
   279 		  bfs.run(typename AbundantGraph::Node(non_root));
   280 		  //root should be reached
   281 		  
   282 		  //Augmenting on the found path
   283 		  Node n=root;
   284 		  ResGraphEdge e;
   285 		  while (n!=non_root){
   286 		    e = bfs_pred[n];
   287 		    n = res_graph.tail(e);
   288 		    res_graph.augment(e,qty_to_augment);
   289 		  }
   290 	  
   291 		  //We know that non_root had positive excess
   292 		  excess_nodes.set(non_root,
   293 				   excess_nodes[non_root] - qty_to_augment);
   294 		  //But what about root node
   295 		  //It might have been positive and so became larger
   296 		  if (excess_deficit[root]>0){
   297 		    excess_nodes.set(root, 
   298 				     excess_nodes[root] + qty_to_augment);
   299 		  }
   300 		  else{
   301 		    //Or negative but not turned into positive
   302 		    deficit_nodes.set(root, 
   303 				      deficit_nodes[root] - qty_to_augment);
   304 		  }
   305 
   306 		  //Update the excess_deficit map
   307 		  excess_deficit[non_root] -= qty_to_augment;
   308 		  excess_deficit[root] += qty_to_augment;
   309 
   310 		  
   311 		}
   312 	      }
   313 	      //What happens to i?
   314 	      //Marci and Zsolt says I shouldn't do such things
   315 	      nonabundant_arcs.erase(i++);
   316 	      abundant_arcs[*i] = true;
   317 	    }
   318 	    else
   319 	      ++i;
   320 	  }
   321 	}
   322 
   323 
   324 	Node s = excess_nodes.top(); 
   325 	max_excess = excess_nodes[s];
   326 	Node t = deficit_nodes.top(); 
   327 	if (max_excess < deficit_nodes[t]){
   328 	  max_excess = deficit_nodes[t];
   329 	}
   330 
   331 
   332 	while(max_excess > (number_of_nodes-1)*delta/number_of_nodes){
   333 	  
   334 	  
   335 	  //s es t valasztasa
   336 	  
   337 	  //Dijkstra part	
   338 	  dijkstra.run(s);
   339 	  
   340 	  /*We know from theory that t can be reached
   341 	  if (!dijkstra.reached(t)){
   342 	    //There are no k paths from s to t
   343 	    break;
   344 	  };
   345 	  */
   346 	  
   347 	  //We have to change the potential
   348 	  FOR_EACH_LOC(typename ResGraph::NodeIt, n, res_graph){
   349 	    potential[n] += dijkstra.distMap()[n];
   350 	  }
   351 
   352 
   353 	  //Augmenting on the sortest path
   354 	  Node n=t;
   355 	  ResGraphEdge e;
   356 	  while (n!=s){
   357 	    e = dijkstra.pred(n);
   358 	    n = dijkstra.predNode(n);
   359 	    res_graph.augment(e,delta);
   360 	    /*
   361 	    //Let's update the total cost
   362 	    if (res_graph.forward(e))
   363 	      total_cost += cost[e];
   364 	    else 
   365 	      total_cost -= cost[e];	    
   366 	    */
   367 	  }
   368 	  
   369 	  //Update the excess_deficit map
   370 	  excess_deficit[s] -= delta;
   371 	  excess_deficit[t] += delta;
   372 	  
   373 
   374 	  //Update the excess_nodes heap
   375 	  if (delta > excess_nodes[s]){
   376 	    if (delta > excess_nodes[s])
   377 	      deficit_nodes.push(s,delta - excess_nodes[s]);
   378 	    excess_nodes.pop();
   379 	    
   380 	  } 
   381 	  else{
   382 	    excess_nodes.set(s, excess_nodes[s] - delta);
   383 	  }
   384 	  //Update the deficit_nodes heap
   385 	  if (delta > deficit_nodes[t]){
   386 	    if (delta > deficit_nodes[t])
   387 	      excess_nodes.push(t,delta - deficit_nodes[t]);
   388 	    deficit_nodes.pop();
   389 	    
   390 	  } 
   391 	  else{
   392 	    deficit_nodes.set(t, deficit_nodes[t] - delta);
   393 	  }
   394 	  //Dijkstra part ends here
   395 	  
   396 	  //Choose s and t again
   397 	  s = excess_nodes.top(); 
   398 	  max_excess = excess_nodes[s];
   399 	  t = deficit_nodes.top(); 
   400 	  if (max_excess < deficit_nodes[t]){
   401 	    max_excess = deficit_nodes[t];
   402 	  }
   403 
   404 	}
   405 
   406 	/*
   407 	 * End of the delta scaling phase 
   408 	*/
   409 
   410 	//Whatever this means
   411 	delta = delta / 2;
   412 
   413 	/*This is not necessary here
   414 	//Update the max_excess
   415 	max_excess = 0;
   416 	FOR_EACH_LOC(typename Graph::NodeIt, n, graph){
   417 	  if (max_excess < excess_deficit[n]){
   418 	    max_excess = excess_deficit[n];
   419 	  }
   420 	}
   421 	*/
   422 
   423 	  
   424       }//while(max_excess > 0)
   425       
   426 
   427       //return i;
   428     }
   429 
   430 
   431 
   432 
   433     ///This function gives back the total cost of the found paths.
   434     ///Assumes that \c run() has been run and nothing changed since then.
   435     Cost totalCost(){
   436       return total_cost;
   437     }
   438 
   439     ///Returns a const reference to the EdgeMap \c flow. \pre \ref run() must
   440     ///be called before using this function.
   441     const FlowMap &getFlow() const { return flow;}
   442 
   443   ///Returns a const reference to the NodeMap \c potential (the dual solution).
   444     /// \pre \ref run() must be called before using this function.
   445     const PotentialMap &getPotential() const { return potential;}
   446 
   447     ///This function checks, whether the given solution is optimal
   448     ///Running after a \c run() should return with true
   449     ///In this "state of the art" this only checks optimality, doesn't bother with feasibility
   450     ///
   451     ///\todo Is this OK here?
   452     bool checkComplementarySlackness(){
   453       Cost mod_pot;
   454       Cost fl_e;
   455       FOR_EACH_LOC(typename Graph::EdgeIt, e, graph){
   456 	//C^{\Pi}_{i,j}
   457 	mod_pot = cost[e]-potential[graph.head(e)]+potential[graph.tail(e)];
   458 	fl_e = flow[e];
   459 	//	std::cout << fl_e << std::endl;
   460 	if (mod_pot > 0 && fl_e != 0)
   461 	  return false;
   462 
   463       }
   464       return true;
   465     }
   466 
   467     /*
   468     //For testing purposes only
   469     //Lists the node_properties
   470     void write_property_vector(const SupplyDemandMap& a,
   471 			       char* prop_name="property"){
   472       FOR_EACH_LOC(typename Graph::NodeIt, i, graph){
   473 	cout<<"Node id.: "<<graph.id(i)<<", "<<prop_name<<" value: "<<a[i]<<endl;
   474       }
   475       cout<<endl;
   476     }
   477     */
   478     bool checkFeasibility(){
   479       SupplyDemandMap supdem(graph);
   480       FOR_EACH_LOC(typename Graph::EdgeIt, e, graph){
   481 
   482 	if ( flow[e] < 0){
   483 
   484 	  return false;
   485 	}
   486 	supdem[graph.tail(e)] += flow[e];
   487 	supdem[graph.head(e)] -= flow[e];
   488       }
   489       //write_property_vector(supdem, "supdem");
   490       //write_property_vector(supply_demand, "supply_demand");
   491 
   492       FOR_EACH_LOC(typename Graph::NodeIt, n, graph){
   493 
   494 	if ( supdem[n] != supply_demand[n]){
   495 	  //cout<<"Node id.: "<<graph.id(n)<<" : "<<supdem[n]<<", should be: "<<supply_demand[n]<<endl;
   496 	  return false;
   497 	}
   498       }
   499 
   500       return true;
   501     }
   502 
   503     bool checkOptimality(){
   504       return checkFeasibility() && checkComplementarySlackness();
   505     }
   506 
   507   }; //class MinCostFlow
   508 
   509   ///@}
   510 
   511 } //namespace hugo
   512 
   513 #endif //HUGO_MINCOSTFLOW_H