src/work/athos/preflow_push.hh
author marci
Thu, 05 Feb 2004 15:06:45 +0000
changeset 64 72bd463289a9
child 77 69b2d279c8f0
permissions -rw-r--r--
.
     1 #ifndef PREFLOW_PUSH_HH
     2 #define PREFLOW_PUSH_HH
     3 
     4 #include <algorithm>
     5 #include <list>
     6 #include <vector>
     7 //#include "pf_hiba.hh"
     8 //#include <marci_list_graph.hh>
     9 #include <marci_graph_traits.hh>
    10 #include "reverse_bfs.hh"
    11 
    12 using namespace std;
    13 
    14 namespace marci {
    15 
    16   template <typename graph_type, typename T>
    17   class preflow_push {
    18 
    19     //Hasznos typedef-ek
    20     typedef graph_traits<graph_type>::node_iterator node_iterator;
    21     typedef graph_traits<graph_type>::edge_iterator edge_iterator;
    22     typedef graph_traits<graph_type>::each_node_iterator each_node_iterator;
    23     typedef graph_traits<graph_type>::each_edge_iterator each_edge_iterator;
    24     typedef graph_traits<graph_type>::out_edge_iterator out_edge_iterator;
    25     typedef graph_traits<graph_type>::in_edge_iterator in_edge_iterator;
    26     typedef graph_traits<graph_type>::sym_edge_iterator sym_edge_iterator;
    27 
    28     //---------------------------------------------
    29     //Parameters of the algorithm
    30     //---------------------------------------------
    31     //Fully examine an active node until excess becomes 0
    32     enum node_examination_t {examine_full, examine_to_relabel};
    33     //No more implemented yet:, examine_only_one_edge};
    34     node_examination_t node_examination;
    35     //Which implementation to be used
    36     enum implementation_t {impl_fifo, impl_highest_label};
    37     //No more implemented yet:};
    38     implementation_t implementation;
    39     //---------------------------------------------
    40     //Parameters of the algorithm
    41     //---------------------------------------------
    42  
    43   private:
    44     //input
    45     graph_type& G;
    46     node_iterator s;
    47     node_iterator t;
    48     edge_property_vector<graph_type, T> &capacity;
    49     //output
    50     edge_property_vector<graph_type, T> preflow;
    51     T maxflow_value;
    52   
    53     //auxiliary variables for computation
    54     int number_of_nodes;
    55     node_property_vector<graph_type, int> level;
    56     node_property_vector<graph_type, T> excess;
    57     
    58     //Number of nodes on each level
    59     vector<int> num_of_nodes_on_level;
    60     
    61     //For the FIFO implementation
    62     list<node_iterator> fifo_nodes;
    63     //For 'highest label' implementation
    64     int highest_active;
    65     //int second_highest_active;
    66     vector< list<node_iterator> > active_nodes;
    67 
    68   public:
    69   
    70     //Constructing the object using the graph, source, sink and capacity vector
    71     preflow_push(
    72 		      graph_type& _G, 
    73 		      node_iterator _s, 
    74 		      node_iterator _t, 
    75 		      edge_property_vector<graph_type, T>& _capacity)
    76       : G(_G), s(_s), t(_t), 
    77 	capacity(_capacity), 
    78 	preflow(_G),
    79 	//Counting the number of nodes
    80 	number_of_nodes(number_of(G.first_node())),
    81 	level(_G),
    82 	excess(_G)//,
    83         // Default constructor: active_nodes()
    84     { 
    85       //Simplest parameter settings
    86       node_examination = examine_full;//examine_to_relabel;//
    87       //Which implementation to be usedexamine_full
    88       implementation = impl_highest_label;//impl_fifo;
    89  
    90       //
    91       num_of_nodes_on_level.resize(2*number_of_nodes-1);
    92       num_of_nodes_on_level.clear();
    93 
    94       switch(implementation){
    95       case impl_highest_label :{
    96 	active_nodes.resize(2*number_of_nodes-1);
    97 	active_nodes.clear();
    98 	break;
    99       }
   100       default:
   101 	break;
   102       }
   103 
   104     }
   105 
   106     //Returns the value of a maximal flow 
   107     T run();
   108   
   109     edge_property_vector<graph_type, T> getmaxflow(){
   110       return preflow;
   111     }
   112 
   113 
   114   private:
   115     //For testing purposes only
   116     //Lists the node_properties
   117     void write_property_vector(node_property_vector<graph_type, T> a, 
   118 			       char* prop_name="property"){
   119       for(each_node_iterator i=G.first_node(); i.valid(); ++i) {
   120 	cout<<"Node id.: "<<G.id(i)<<", "<<prop_name<<" value: "<<a.get(i)<<endl;
   121       }
   122       cout<<endl;
   123     }
   124 
   125     //Modifies the excess of the node and makes sufficient changes
   126     void modify_excess(const node_iterator& a ,T v){
   127 	T old_value=excess.get(a);
   128 	excess.put(a,old_value+v);
   129     }
   130   
   131     //This private procedure is supposed to modify the preflow on edge j
   132     //by value v (which can be positive or negative as well) 
   133     //and maintain the excess on the head and tail
   134     //Here we do not check whether this is possible or not
   135     void modify_preflow(edge_iterator j, const T& v){
   136 
   137       //Auxiliary variable
   138       T old_value;
   139 	
   140       //Modifiyng the edge
   141       old_value=preflow.get(j);
   142       preflow.put(j,old_value+v);
   143 
   144 
   145       //Modifiyng the head
   146       modify_excess(G.head(j),v);
   147 	
   148       //Modifiyng the tail
   149       modify_excess(G.tail(j),-v);
   150 
   151     }
   152 
   153     //Gives the active node to work with 
   154     //(depending on the implementation to be used)
   155     node_iterator get_active_node(){
   156       //cout<<highest_active<<endl;
   157 
   158       switch(implementation) {
   159       case impl_highest_label : {
   160 
   161 	//First need to find the highest label for which there"s an active node
   162 	while( highest_active>=0 && active_nodes[highest_active].empty() ){ 
   163 	  --highest_active;
   164 	}
   165 
   166 	if( highest_active>=0) {
   167 	  node_iterator a=active_nodes[highest_active].front();
   168 	  active_nodes[highest_active].pop_front();
   169 	  return a;
   170 	}
   171 	else {
   172 	  return node_iterator();
   173 	}
   174 	
   175 	break;
   176 	
   177       }
   178       case impl_fifo : {
   179 
   180 	if( ! fifo_nodes.empty() ) {
   181 	  node_iterator a=fifo_nodes.front();
   182 	  fifo_nodes.pop_front();
   183 	  return a;
   184 	}
   185 	else {
   186 	  return node_iterator();
   187 	}
   188 	break;
   189       }
   190       }
   191       //
   192       return node_iterator();
   193     }
   194 
   195     //Puts node 'a' among the active nodes
   196     void make_active(const node_iterator& a){
   197       //s and t never become active
   198       if (a!=s && a!= t){
   199 	switch(implementation){
   200 	case impl_highest_label :
   201 	  active_nodes[level.get(a)].push_back(a);
   202 	  break;
   203 	case impl_fifo :
   204 	  fifo_nodes.push_back(a);
   205 	  break;
   206 	}
   207 
   208       }
   209 
   210       //Update highest_active label
   211       if (highest_active<level.get(a)){
   212 	highest_active=level.get(a);
   213       }
   214 
   215     }
   216 
   217     //Changes the level of node a and make sufficent changes
   218     void change_level_to(node_iterator a, int new_value){
   219       int seged = level.get(a);
   220       level.put(a,new_value);
   221       --num_of_nodes_on_level[seged];
   222       ++num_of_nodes_on_level[new_value];
   223     }
   224 
   225     //Collection of things useful (or necessary) to do before running
   226     void preprocess(){
   227 
   228       //---------------------------------------
   229       //Initialize parameters
   230       //---------------------------------------
   231 
   232       //Setting starting preflow, level and excess values to zero
   233       //This can be important, if the algorithm is run more then once
   234       for(each_node_iterator i=G.first_node(); i.valid(); ++i) {
   235         level.put(i,0);
   236         excess.put(i,0);
   237 	for(out_edge_iterator j=G.first_out_edge(i); j.valid(); ++j) 
   238 	  preflow.put(j, 0);
   239       }
   240       num_of_nodes_on_level[0]=number_of_nodes;
   241       highest_active=0;
   242       //---------------------------------------
   243       //Initialize parameters
   244       //---------------------------------------
   245 
   246       
   247       //------------------------------------
   248       //This is the only part that uses BFS
   249       //------------------------------------
   250       //Setting starting level values using reverse bfs
   251       reverse_bfs<graph_type> rev_bfs(G,t);
   252       rev_bfs.run();
   253       //write_property_vector(rev_bfs.dist,"rev_bfs");
   254       for(each_node_iterator i=G.first_node(); i.valid(); ++i) {
   255         change_level_to(i,rev_bfs.dist(i));
   256 	//level.put(i,rev_bfs.dist.get(i));
   257       }
   258       //------------------------------------
   259       //This is the only part that uses BFS
   260       //------------------------------------
   261       
   262       
   263       //Starting level of s
   264       change_level_to(s,number_of_nodes);
   265       //level.put(s,number_of_nodes);
   266       
   267       
   268       //we push as much preflow from s as possible to start with
   269       for(out_edge_iterator j=G.first_out_edge(s); j.valid(); ++j){ 
   270 	modify_preflow(j,capacity.get(j) );
   271 	make_active(G.head(j));
   272 	int lev=level.get(G.head(j));
   273 	if(highest_active<lev){
   274 	  highest_active=lev;
   275 	}
   276       }
   277       //cout<<highest_active<<endl;
   278     } 
   279 
   280     
   281     //If the preflow is less than the capacity on the given edge
   282     //then it is an edge in the residual graph
   283     bool is_admissible_forward_edge(out_edge_iterator j, int& new_level){
   284       if (capacity.get(j)>preflow.get(j)){
   285 	if(level.get(G.tail(j))==level.get(G.head(j))+1){
   286 	  return true;
   287 	}
   288 	else{
   289 	  if (level.get(G.head(j)) < new_level)
   290 	    new_level=level.get(G.head(j));
   291 	}
   292       }
   293       return false;
   294     }
   295 
   296     //If the preflow is greater than 0 on the given edge
   297     //then the edge reversd is an edge in the residual graph
   298     bool is_admissible_backward_edge(in_edge_iterator j, int& new_level){
   299       if (0<preflow.get(j)){
   300 	if(level.get(G.tail(j))==level.get(G.head(j))-1){
   301 	  return true;
   302 	}
   303 	else{
   304 	  if (level.get(G.tail(j)) < new_level)
   305 	    new_level=level.get(G.tail(j));
   306 	}
   307 	
   308       }
   309       return false;
   310     }
   311 
   312  
   313   };  //class preflow_push  
   314 
   315   template<typename graph_type, typename T>
   316     T preflow_push<graph_type, T>::run() {
   317     
   318     preprocess();
   319     
   320     T e,v;
   321     node_iterator a;
   322     while (a=get_active_node(), a.valid()){
   323       //cout<<G.id(a)<<endl;
   324       //write_property_vector(excess,"excess");
   325       //write_property_vector(level,"level");
   326 
   327       //Initial value for the new level for the active node we are dealing with
   328       int new_level=2*number_of_nodes;
   329 
   330       bool go_to_next_node=false;
   331       e = excess.get(a);
   332       while (!go_to_next_node){
   333 	
   334 	//Out edges from node a
   335 	{
   336 	  out_edge_iterator j=G.first_out_edge(a);
   337 	  while (j.valid() && e){
   338 
   339 	    if (is_admissible_forward_edge(j,new_level)){
   340 	      v=min(e,capacity.get(j) - preflow.get(j));
   341 	      e -= v;
   342 	      //New node might become active
   343 	      if (excess.get(G.head(j))==0){
   344 		make_active(G.head(j));
   345 	      }
   346 	      modify_preflow(j,v);
   347 	    }
   348 	    ++j;
   349 	  }
   350 	}
   351 	//In edges to node a
   352 	{
   353 	  in_edge_iterator j=G.first_in_edge(a);
   354 	  while (j.valid() && e){
   355 	    if (is_admissible_backward_edge(j,new_level)){
   356 	      v=min(e,preflow.get(j));
   357 	      e -= v;
   358 	      //New node might become active
   359 	      if (excess.get(G.tail(j))==0){
   360 		make_active(G.tail(j));
   361 	      }
   362 	      modify_preflow(j,-v);
   363 	    }
   364 	    ++j;
   365 	  }
   366 	}
   367 
   368 	//cout<<G.id(a)<<" "<<new_level<<endl;
   369 
   370 	if (0==e){
   371 	  //Saturating push
   372 	  go_to_next_node=true;
   373 	}
   374 	else{//If there is still excess in node a
   375 
   376 	  //Level remains empty
   377 	  if (num_of_nodes_on_level[level.get(a)]==1){
   378 	    change_level_to(a,number_of_nodes);
   379 	    //go_to_next_node=True;
   380 	  }
   381 	  else{
   382 	    change_level_to(a,new_level+1);
   383 	    //increase_level(a);
   384 	  }
   385 
   386     
   387 	  
   388 
   389 	  switch(node_examination){
   390 	  case examine_to_relabel:
   391 	    make_active(a);
   392 
   393 	    go_to_next_node = true;
   394 	    break;
   395 	  default:
   396 	    break;
   397 	  }
   398 	  
   399     
   400 	
   401 	}//if (0==e)
   402       }
   403     }
   404     maxflow_value = excess.get(t);
   405     return maxflow_value;
   406   }//run
   407 
   408 
   409 }//namespace marci
   410 
   411 #endif //PREFLOW_PUSH_HH