src/work/athos/preflow_push.hh
author alpar
Fri, 23 Jul 2004 17:13:23 +0000
changeset 737 2d867176d10e
parent 331 f5461f8bc59b
permissions -rw-r--r--
Several changes in Kruskal alg.
- Input object interface was changed to an STL compatible one.
- template parameters of class KruskalPairVec has been simplified.
- (the most of) the names meet the naming conventions.
- a lot of (but still not enough) documentation has been added.
- class KruskalMapVec has been commented out.
     1 #ifndef HUGO_PREFLOW_PUSH_HH
     2 #define HUGO_PREFLOW_PUSH_HH
     3 
     4 //#include <algorithm>
     5 #include <list>
     6 #include <vector>
     7 #include <queue>
     8 //#include "pf_hiba.hh"
     9 //#include <marci_list_graph.hh>
    10 //#include <marci_graph_traits.hh>
    11 #include <invalid.h>
    12 #include <graph_wrapper.h>
    13 //#include <reverse_bfs.hh>
    14 
    15 using namespace std;
    16 
    17 namespace hugo {
    18 
    19   template <typename Graph, typename T>
    20   class preflow_push {
    21 
    22     //Useful typedefs
    23     typedef typename Graph::Node Node;
    24     typedef typename Graph::NodeIt NodeIt;
    25     typedef typename Graph::Edge Edge;
    26     typedef typename Graph::OutEdgeIt OutEdgeIt;
    27     typedef typename Graph::InEdgeIt InEdgeIt;
    28     typedef typename Graph::EdgeMap<T> CapacityType;
    29 
    30     typedef ResGraphWrapper<const Graph,int,CapacityType,CapacityType> ResGraphType;
    31 
    32 
    33     //---------------------------------------------
    34     //Parameters of the algorithm
    35     //---------------------------------------------
    36     //Fully examine an active node until excess becomes 0
    37     enum node_examination_t {examine_full, examine_to_relabel};
    38     //No more implemented yet:, examine_only_one_edge};
    39     node_examination_t node_examination;
    40     //Which implementation to be used
    41     enum implementation_t {impl_fifo, impl_highest_label};
    42     //No more implemented yet:};
    43     implementation_t implementation;
    44     //---------------------------------------------
    45     //Parameters of the algorithm
    46     //---------------------------------------------
    47  
    48   private:
    49     //input
    50     Graph& G;
    51     Node s;
    52     Node t;
    53     CapacityType &capacity;
    54 
    55     //output
    56     CapacityType preflow;
    57     T maxflow_value;
    58   
    59     //auxiliary variables for computation
    60     //The number of the nodes
    61     int number_of_nodes;
    62     //A nodemap for the level
    63     typename Graph::NodeMap<int> level;
    64     //A nodemap for the excess
    65     typename Graph::NodeMap<T> excess;
    66     
    67     //Number of nodes on each level
    68     vector<int> num_of_nodes_on_level;
    69     
    70     //For the FIFO implementation
    71     list<Node> fifo_nodes;
    72     //For 'highest label' implementation
    73     int highest_active;
    74     //int second_highest_active;
    75     vector< list<Node> > active_nodes;
    76 
    77   public:
    78   
    79     //Constructing the object using the graph, source, sink and capacity vector
    80     preflow_push(
    81 		      Graph& _G, 
    82 		      Node _s, 
    83 		      Node _t, 
    84 		      typename Graph::EdgeMap<T> & _capacity)
    85       : G(_G), s(_s), t(_t), 
    86 	capacity(_capacity), 
    87 	preflow(_G),
    88 	//Counting the number of nodes
    89 	//number_of_nodes(count(G.first<EachNodeIt>())),
    90 	number_of_nodes(G.nodeNum()),
    91 
    92 	level(_G),
    93 	excess(_G)//,
    94         // Default constructor: active_nodes()
    95     { 
    96       //Simplest parameter settings
    97       node_examination = examine_full;//examine_to_relabel;//
    98       //Which implementation to be usedexamine_full
    99       implementation = impl_highest_label;//impl_fifo;
   100  
   101       //
   102       num_of_nodes_on_level.resize(2*number_of_nodes-1);
   103       num_of_nodes_on_level.clear();
   104 
   105       switch(implementation){
   106       case impl_highest_label :{
   107 	active_nodes.clear();
   108 	active_nodes.resize(2*number_of_nodes-1);
   109 	
   110 	break;
   111       }
   112       default:
   113 	break;
   114       }
   115 
   116     }
   117 
   118     //Returns the value of a maximal flow 
   119     T run();
   120   
   121     typename Graph::EdgeMap<T>  getmaxflow(){
   122       return preflow;
   123     }
   124 
   125 
   126   private:
   127     //For testing purposes only
   128     //Lists the node_properties
   129     void write_property_vector(typename Graph::NodeMap<T> a,
   130 			       //node_property_vector<Graph, T> a, 
   131 			       char* prop_name="property"){
   132       for(NodeIt i=G.template first<NodeIt>(); G.valid(i); G.next(i)) {
   133 	cout<<"Node id.: "<<G.id(i)<<", "<<prop_name<<" value: "<<a[i]<<endl;
   134       }
   135       cout<<endl;
   136     }
   137     /*
   138     //Modifies the excess of the node and makes sufficient changes
   139     void modify_excess(const Node& a ,T v){
   140       //T old_value=excess[a];
   141       excess[a] += v;
   142     }
   143   
   144     //This private procedure is supposed to modify the preflow on edge j
   145     //by value v (which can be positive or negative as well) 
   146     //and maintain the excess on the head and tail
   147     //Here we do not check whether this is possible or not
   148     void modify_preflow(Edge j, const T& v){
   149 
   150       //Modifiyng the edge
   151       preflow[j] += v;
   152 
   153 
   154       //Modifiyng the head
   155       modify_excess(G.head(j),v);
   156 	
   157       //Modifiyng the tail
   158       modify_excess(G.tail(j),-v);
   159 
   160     }
   161     */
   162     //Gives the active node to work with 
   163     //(depending on the implementation to be used)
   164     Node get_active_node(){
   165       
   166 
   167       switch(implementation) {
   168       case impl_highest_label : {
   169 
   170 	//First need to find the highest label for which there's an active node
   171 	while( highest_active>=0 && active_nodes[highest_active].empty() ){ 
   172 	  --highest_active;
   173 	}
   174 
   175 	if( highest_active>=0) {
   176 	  
   177 
   178 	  Node a=active_nodes[highest_active].front();
   179 	  active_nodes[highest_active].pop_front();
   180 	  
   181 	  return a;
   182 	}
   183 	else {
   184 	  return INVALID;
   185 	}
   186 	
   187 	break;
   188 	
   189       }
   190       case impl_fifo : {
   191 
   192 	if( ! fifo_nodes.empty() ) {
   193 	  Node a=fifo_nodes.front();
   194 	  fifo_nodes.pop_front();
   195 	  return a;
   196 	}
   197 	else {
   198 	  return INVALID;
   199 	}
   200 	break;
   201       }
   202       }
   203       //
   204       return INVALID;
   205     }
   206 
   207     //Puts node 'a' among the active nodes
   208     void make_active(const Node& a){
   209       //s and t never become active
   210       if (a!=s && a!= t){
   211 	switch(implementation){
   212 	case impl_highest_label :
   213 	  active_nodes[level[a]].push_back(a);
   214 	  break;
   215 	case impl_fifo :
   216 	  fifo_nodes.push_back(a);
   217 	  break;
   218 	}
   219 
   220       }
   221 
   222       //Update highest_active label
   223       if (highest_active<level[a]){
   224 	highest_active=level[a];
   225       }
   226 
   227     }
   228 
   229     //Changes the level of node a and make sufficent changes
   230     void change_level_to(Node a, int new_value){
   231       int seged = level[a];
   232       level.set(a,new_value);
   233       --num_of_nodes_on_level[seged];
   234       ++num_of_nodes_on_level[new_value];
   235     }
   236 
   237     //Collection of things useful (or necessary) to do before running
   238 
   239     void preprocess(){
   240 
   241       //---------------------------------------
   242       //Initialize parameters
   243       //---------------------------------------
   244 
   245       //Setting starting preflow, level and excess values to zero
   246       //This can be important, if the algorithm is run more then once
   247       for(NodeIt i=G.template first<NodeIt>(); G.valid(i); G.next(i)) {
   248         level.set(i,0);
   249         excess.set(i,0);
   250 	for(OutEdgeIt j=G.template first<OutEdgeIt>(i); G.valid(j); G.next(j)) 
   251 	  preflow.set(j, 0);
   252       }
   253       num_of_nodes_on_level[0]=number_of_nodes;
   254       highest_active=0;
   255       //---------------------------------------
   256       //Initialize parameters
   257       //---------------------------------------
   258 
   259       
   260       //------------------------------------
   261       //This is the only part that uses BFS
   262       //------------------------------------
   263 
   264       /*Reverse_bfs from t, to find the starting level.*/
   265       //Copyright: Jacint
   266       change_level_to(t,0);
   267 
   268       std::queue<Node> bfs_queue;
   269       bfs_queue.push(t);
   270 
   271       while (!bfs_queue.empty()) {
   272 
   273 	Node v=bfs_queue.front();	
   274 	bfs_queue.pop();
   275 	int l=level[v]+1;
   276 
   277 	InEdgeIt e;
   278 	for(G.first(e,v); G.valid(e); G.next(e)) {
   279 	  Node w=G.tail(e);
   280 	  if ( level[w] == number_of_nodes && w != s ) {
   281 	    bfs_queue.push(w);
   282 	    //Node first=level_list[l];
   283 	    //if ( G.valid(first) ) left.set(first,w);
   284 	    //right.set(w,first);
   285 	    //level_list[l]=w;
   286 	    change_level_to(w, l);
   287 	    //level.set(w, l);
   288 	  }
   289 	}
   290       }
   291       change_level_to(s,number_of_nodes);
   292       //level.set(s,number_of_nodes);
   293 
   294       /*
   295       //Setting starting level values using reverse bfs
   296       reverse_bfs<Graph> rev_bfs(G,t);
   297       rev_bfs.run();
   298       //write_property_vector(rev_bfs.dist,"rev_bfs");
   299       for(NodeIt i=G.template first<NodeIt>(); G.valid(i); G.next(i)) {
   300         change_level_to(i,rev_bfs.dist(i));
   301 	//level.put(i,rev_bfs.dist.get(i));
   302       }
   303       */
   304       //------------------------------------
   305       //This is the only part that uses BFS
   306       //------------------------------------
   307       
   308       
   309       //Starting level of s
   310       change_level_to(s,number_of_nodes);
   311       //level.put(s,number_of_nodes);
   312       
   313       
   314       //we push as much preflow from s as possible to start with
   315       for(OutEdgeIt j=G.template first<OutEdgeIt>(s); G.valid(j); G.next(j)){ 
   316 	modify_preflow(j,capacity[j] );
   317 	make_active(G.head(j));
   318 	int lev=level[G.head(j)];
   319 	if(highest_active<lev){
   320 	  highest_active=lev;
   321 	}
   322       }
   323       //cout<<highest_active<<endl;
   324     } 
   325 
   326     
   327     //If the preflow is less than the capacity on the given edge
   328     //then it is an edge in the residual graph
   329     bool is_admissible_forward_edge(Edge j, int& new_level){
   330 
   331       if (capacity[j]>preflow[j]){
   332 	if(level[G.tail(j)]==level[G.head(j)]+1){
   333 	  return true;
   334 	}
   335 	else{
   336 	  if (level[G.head(j)] < new_level)
   337 	    new_level=level[G.head(j)];
   338 	}
   339       }
   340       return false;
   341     }
   342 
   343     //If the preflow is greater than 0 on the given edge
   344     //then the edge reversd is an edge in the residual graph
   345     bool is_admissible_backward_edge(Edge j, int& new_level){
   346       
   347       if (0<preflow[j]){
   348 	if(level[G.tail(j)]==level[G.head(j)]-1){
   349 	 
   350 	  return true;
   351 	}
   352 	else{
   353 	  if (level[G.tail(j)] < new_level)
   354 	    new_level=level[G.tail(j)];
   355 	}
   356 	
   357       }
   358       return false;
   359     }
   360 
   361  
   362   };  //class preflow_push  
   363 
   364   template<typename Graph, typename T>
   365     T preflow_push<Graph, T>::run() {
   366     
   367     //We need a residual graph
   368     ResGraphType res_graph(G, preflow, capacity);
   369     
   370     preprocess();
   371     //write_property_vector(level,"level");
   372     T e,v;
   373     Node a;
   374     while (a=get_active_node(), G.valid(a)){
   375       
   376       bool go_to_next_node=false;
   377       e = excess[a];
   378       while (!go_to_next_node){
   379 
   380 	//Initial value for the new level for the active node we are dealing with
   381 	int new_level=2*number_of_nodes;
   382 
   383 
   384 	//Out edges from node a
   385 	{
   386 	  ResGraphType::OutEdgeIt j=res_graph.first(j,a);
   387 	  while (res_graph.valid(j) && e){
   388 	    if (is_admissible_forward_edge(j,new_level)){
   389 	      v=min(e,res_graph.resCap(j));
   390 	      e -= v;
   391 	      //New node might become active
   392 	      if (excess[res_graph.head(j)]==0){
   393 		make_active(res_graph.head(j));
   394 	      }
   395 	      res_graph.augment(j,v);
   396 	      excess[res_graph.tail(j)] -= v;
   397 	      excess[res_graph.head(j)] += v;
   398 	    }
   399 	    res_graph.next(j);
   400 	  }
   401 	}
   402 
   403 	/*
   404 	//Out edges from node a
   405 	{
   406 	  OutEdgeIt j=G.template first<OutEdgeIt>(a);
   407 	  while (G.valid(j) && e){
   408 
   409 	    if (is_admissible_forward_edge(j,new_level)){
   410 	      v=min(e,capacity[j] - preflow[j]);
   411 	      e -= v;
   412 	      //New node might become active
   413 	      if (excess[G.head(j)]==0){
   414 		make_active(G.head(j));
   415 	      }
   416 	      modify_preflow(j,v);
   417 	    }
   418 	    G.next(j);
   419 	  }
   420 	}
   421 	//In edges to node a
   422 	{
   423 	  InEdgeIt j=G.template first<InEdgeIt>(a);
   424 	  while (G.valid(j) && e){
   425 	    if (is_admissible_backward_edge(j,new_level)){
   426 	      v=min(e,preflow[j]);
   427 	      e -= v;
   428 	      //New node might become active
   429 	      if (excess[G.tail(j)]==0){
   430 		make_active(G.tail(j));
   431 	      }
   432 	      modify_preflow(j,-v);
   433 	    }
   434 	    G.next(j);
   435 	  }
   436 	}
   437 	*/
   438 
   439 	//if (G.id(a)==999)
   440 	//cout<<new_level<<" e: "<<e<<endl;
   441 	//cout<<G.id(a)<<" "<<new_level<<endl;
   442 
   443 	if (0==e){
   444 	  //Saturating push
   445 	  go_to_next_node=true;
   446 	}
   447 	else{//If there is still excess in node a
   448 	  
   449 	  //change_level_to(a,new_level+1);
   450 	  
   451 	  //Level remains empty
   452 	  if (num_of_nodes_on_level[level[a]]==1){
   453 	    change_level_to(a,number_of_nodes);
   454 	    //go_to_next_node=True;
   455 	  }
   456 	  else{
   457 	    change_level_to(a,new_level+1);
   458 	    //increase_level(a);
   459 	  }
   460 	  
   461     
   462 	  
   463 
   464 	  switch(node_examination){
   465 	  case examine_to_relabel:
   466 	    make_active(a);
   467 
   468 	    go_to_next_node = true;
   469 	    break;
   470 	  default:
   471 	    break;
   472 	  }
   473 	  
   474     
   475 	
   476 	}//if (0==e)
   477       }
   478     }
   479     maxflow_value = excess[t];
   480     return maxflow_value;
   481   }//run
   482 
   483 
   484 }//namespace hugo
   485 
   486 #endif //PREFLOW_PUSH_HH