alpar@906
|
1 |
/* -*- C++ -*-
|
alpar@906
|
2 |
*
|
alpar@1956
|
3 |
* This file is a part of LEMON, a generic C++ optimization library
|
alpar@1956
|
4 |
*
|
alpar@1956
|
5 |
* Copyright (C) 2003-2006
|
alpar@1956
|
6 |
* Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
|
alpar@1359
|
7 |
* (Egervary Research Group on Combinatorial Optimization, EGRES).
|
alpar@906
|
8 |
*
|
alpar@906
|
9 |
* Permission to use, modify and distribute this software is granted
|
alpar@906
|
10 |
* provided that this copyright notice appears in all copies. For
|
alpar@906
|
11 |
* precise terms see the accompanying LICENSE file.
|
alpar@906
|
12 |
*
|
alpar@906
|
13 |
* This software is provided "AS IS" with no warranty of any kind,
|
alpar@906
|
14 |
* express or implied, and with no claim as to its suitability for any
|
alpar@906
|
15 |
* purpose.
|
alpar@906
|
16 |
*
|
alpar@906
|
17 |
*/
|
alpar@906
|
18 |
|
alpar@921
|
19 |
#ifndef LEMON_MIN_COST_FLOW_H
|
alpar@921
|
20 |
#define LEMON_MIN_COST_FLOW_H
|
alpar@899
|
21 |
|
alpar@899
|
22 |
///\ingroup flowalgs
|
alpar@899
|
23 |
///\file
|
alpar@899
|
24 |
///\brief An algorithm for finding a flow of value \c k (for small values of \c k) having minimal total cost
|
alpar@899
|
25 |
|
alpar@899
|
26 |
|
alpar@921
|
27 |
#include <lemon/dijkstra.h>
|
alpar@1401
|
28 |
#include <lemon/graph_adaptor.h>
|
alpar@921
|
29 |
#include <lemon/maps.h>
|
alpar@899
|
30 |
#include <vector>
|
alpar@899
|
31 |
|
alpar@921
|
32 |
namespace lemon {
|
alpar@899
|
33 |
|
alpar@899
|
34 |
/// \addtogroup flowalgs
|
alpar@899
|
35 |
/// @{
|
alpar@899
|
36 |
|
alpar@899
|
37 |
///\brief Implementation of an algorithm for finding a flow of value \c k
|
alpar@899
|
38 |
///(for small values of \c k) having minimal total cost between 2 nodes
|
alpar@899
|
39 |
///
|
alpar@899
|
40 |
///
|
jacint@1270
|
41 |
/// The class \ref lemon::MinCostFlow "MinCostFlow" implements an
|
jacint@1270
|
42 |
/// algorithm for finding a flow of value \c k having minimal total
|
athos@1527
|
43 |
/// cost from a given source node to a given target node in a
|
athos@1527
|
44 |
/// directed graph with a cost function on the edges. To
|
athos@1527
|
45 |
/// this end, the edge-capacities and edge-costs have to be
|
athos@1527
|
46 |
/// nonnegative. The edge-capacities should be integers, but the
|
athos@1527
|
47 |
/// edge-costs can be integers, reals or of other comparable
|
athos@1527
|
48 |
/// numeric type. This algorithm is intended to be used only for
|
athos@1527
|
49 |
/// small values of \c k, since it is only polynomial in k, not in
|
athos@1527
|
50 |
/// the length of k (which is log k): in order to find the minimum
|
athos@1527
|
51 |
/// cost flow of value \c k it finds the minimum cost flow of value
|
athos@1527
|
52 |
/// \c i for every \c i between 0 and \c k.
|
alpar@899
|
53 |
///
|
alpar@899
|
54 |
///\param Graph The directed graph type the algorithm runs on.
|
alpar@899
|
55 |
///\param LengthMap The type of the length map.
|
alpar@899
|
56 |
///\param CapacityMap The capacity map type.
|
alpar@899
|
57 |
///
|
alpar@899
|
58 |
///\author Attila Bernath
|
alpar@899
|
59 |
template <typename Graph, typename LengthMap, typename CapacityMap>
|
alpar@899
|
60 |
class MinCostFlow {
|
alpar@899
|
61 |
|
alpar@987
|
62 |
typedef typename LengthMap::Value Length;
|
alpar@899
|
63 |
|
alpar@899
|
64 |
//Warning: this should be integer type
|
alpar@987
|
65 |
typedef typename CapacityMap::Value Capacity;
|
alpar@899
|
66 |
|
alpar@899
|
67 |
typedef typename Graph::Node Node;
|
alpar@899
|
68 |
typedef typename Graph::NodeIt NodeIt;
|
alpar@899
|
69 |
typedef typename Graph::Edge Edge;
|
alpar@899
|
70 |
typedef typename Graph::OutEdgeIt OutEdgeIt;
|
alpar@899
|
71 |
typedef typename Graph::template EdgeMap<int> EdgeIntMap;
|
alpar@899
|
72 |
|
alpar@1401
|
73 |
typedef ResGraphAdaptor<const Graph,int,CapacityMap,EdgeIntMap> ResGW;
|
marci@910
|
74 |
typedef typename ResGW::Edge ResGraphEdge;
|
alpar@899
|
75 |
|
marci@941
|
76 |
protected:
|
marci@941
|
77 |
|
marci@941
|
78 |
const Graph& g;
|
marci@941
|
79 |
const LengthMap& length;
|
marci@941
|
80 |
const CapacityMap& capacity;
|
marci@941
|
81 |
|
marci@941
|
82 |
EdgeIntMap flow;
|
marci@941
|
83 |
typedef typename Graph::template NodeMap<Length> PotentialMap;
|
marci@941
|
84 |
PotentialMap potential;
|
marci@941
|
85 |
|
marci@941
|
86 |
Node s;
|
marci@941
|
87 |
Node t;
|
marci@941
|
88 |
|
marci@941
|
89 |
Length total_length;
|
marci@941
|
90 |
|
alpar@899
|
91 |
class ModLengthMap {
|
alpar@899
|
92 |
typedef typename Graph::template NodeMap<Length> NodeMap;
|
marci@941
|
93 |
const ResGW& g;
|
marci@941
|
94 |
const LengthMap &length;
|
alpar@899
|
95 |
const NodeMap &pot;
|
alpar@899
|
96 |
public :
|
alpar@987
|
97 |
typedef typename LengthMap::Key Key;
|
alpar@987
|
98 |
typedef typename LengthMap::Value Value;
|
marci@941
|
99 |
|
marci@941
|
100 |
ModLengthMap(const ResGW& _g,
|
marci@941
|
101 |
const LengthMap &_length, const NodeMap &_pot) :
|
marci@941
|
102 |
g(_g), /*rev(_rev),*/ length(_length), pot(_pot) { }
|
alpar@899
|
103 |
|
alpar@987
|
104 |
Value operator[](typename ResGW::Edge e) const {
|
marci@941
|
105 |
if (g.forward(e))
|
alpar@986
|
106 |
return length[e]-(pot[g.target(e)]-pot[g.source(e)]);
|
alpar@899
|
107 |
else
|
alpar@986
|
108 |
return -length[e]-(pot[g.target(e)]-pot[g.source(e)]);
|
alpar@899
|
109 |
}
|
alpar@899
|
110 |
|
marci@941
|
111 |
}; //ModLengthMap
|
alpar@899
|
112 |
|
marci@941
|
113 |
ResGW res_graph;
|
marci@941
|
114 |
ModLengthMap mod_length;
|
marci@941
|
115 |
Dijkstra<ResGW, ModLengthMap> dijkstra;
|
alpar@899
|
116 |
|
alpar@899
|
117 |
public :
|
alpar@899
|
118 |
|
marci@941
|
119 |
/*! \brief The constructor of the class.
|
alpar@899
|
120 |
|
marci@941
|
121 |
\param _g The directed graph the algorithm runs on.
|
athos@1527
|
122 |
\param _length The length (cost) of the edges.
|
marci@941
|
123 |
\param _cap The capacity of the edges.
|
marci@941
|
124 |
\param _s Source node.
|
marci@941
|
125 |
\param _t Target node.
|
marci@941
|
126 |
*/
|
marci@941
|
127 |
MinCostFlow(Graph& _g, LengthMap& _length, CapacityMap& _cap,
|
marci@941
|
128 |
Node _s, Node _t) :
|
marci@941
|
129 |
g(_g), length(_length), capacity(_cap), flow(_g), potential(_g),
|
marci@941
|
130 |
s(_s), t(_t),
|
marci@941
|
131 |
res_graph(g, capacity, flow),
|
marci@941
|
132 |
mod_length(res_graph, length, potential),
|
marci@941
|
133 |
dijkstra(res_graph, mod_length) {
|
marci@941
|
134 |
reset();
|
marci@941
|
135 |
}
|
alpar@899
|
136 |
|
marci@941
|
137 |
/*! Tries to augment the flow between s and t by 1.
|
marci@941
|
138 |
The return value shows if the augmentation is successful.
|
marci@941
|
139 |
*/
|
marci@941
|
140 |
bool augment() {
|
marci@941
|
141 |
dijkstra.run(s);
|
marci@941
|
142 |
if (!dijkstra.reached(t)) {
|
alpar@899
|
143 |
|
marci@941
|
144 |
//Unsuccessful augmentation.
|
marci@941
|
145 |
return false;
|
marci@941
|
146 |
} else {
|
alpar@899
|
147 |
|
marci@941
|
148 |
//We have to change the potential
|
marci@941
|
149 |
for(typename ResGW::NodeIt n(res_graph); n!=INVALID; ++n)
|
marci@1027
|
150 |
potential.set(n, potential[n]+dijkstra.distMap()[n]);
|
alpar@899
|
151 |
|
jacint@1270
|
152 |
//Augmenting on the shortest path
|
alpar@899
|
153 |
Node n=t;
|
alpar@899
|
154 |
ResGraphEdge e;
|
alpar@899
|
155 |
while (n!=s){
|
deba@1763
|
156 |
e = dijkstra.predEdge(n);
|
alpar@899
|
157 |
n = dijkstra.predNode(n);
|
alpar@899
|
158 |
res_graph.augment(e,1);
|
alpar@899
|
159 |
//Let's update the total length
|
alpar@899
|
160 |
if (res_graph.forward(e))
|
alpar@899
|
161 |
total_length += length[e];
|
alpar@899
|
162 |
else
|
alpar@899
|
163 |
total_length -= length[e];
|
alpar@899
|
164 |
}
|
alpar@899
|
165 |
|
marci@941
|
166 |
return true;
|
alpar@899
|
167 |
}
|
marci@941
|
168 |
}
|
marci@941
|
169 |
|
marci@941
|
170 |
/*! \brief Runs the algorithm.
|
marci@941
|
171 |
|
marci@941
|
172 |
Runs the algorithm.
|
marci@941
|
173 |
Returns k if there is a flow of value at least k from s to t.
|
marci@941
|
174 |
Otherwise it returns the maximum value of a flow from s to t.
|
marci@941
|
175 |
|
marci@941
|
176 |
\param k The value of the flow we are looking for.
|
marci@941
|
177 |
|
marci@941
|
178 |
\todo May be it does make sense to be able to start with a nonzero
|
marci@941
|
179 |
feasible primal-dual solution pair as well.
|
marci@941
|
180 |
|
marci@941
|
181 |
\todo If the actual flow value is bigger than k, then everything is
|
marci@941
|
182 |
cleared and the algorithm starts from zero flow. Is it a good approach?
|
marci@941
|
183 |
*/
|
marci@941
|
184 |
int run(int k) {
|
marci@941
|
185 |
if (flowValue()>k) reset();
|
marci@941
|
186 |
while (flowValue()<k && augment()) { }
|
marci@941
|
187 |
return flowValue();
|
marci@941
|
188 |
}
|
alpar@899
|
189 |
|
marci@941
|
190 |
/*! \brief The class is reset to zero flow and potential.
|
marci@941
|
191 |
The class is reset to zero flow and potential.
|
marci@941
|
192 |
*/
|
marci@941
|
193 |
void reset() {
|
marci@941
|
194 |
total_length=0;
|
marci@941
|
195 |
for (typename Graph::EdgeIt e(g); e!=INVALID; ++e) flow.set(e, 0);
|
marci@941
|
196 |
for (typename Graph::NodeIt n(g); n!=INVALID; ++n) potential.set(n, 0);
|
marci@941
|
197 |
}
|
marci@941
|
198 |
|
marci@941
|
199 |
/*! Returns the value of the actual flow.
|
marci@941
|
200 |
*/
|
marci@941
|
201 |
int flowValue() const {
|
marci@941
|
202 |
int i=0;
|
marci@941
|
203 |
for (typename Graph::OutEdgeIt e(g, s); e!=INVALID; ++e) i+=flow[e];
|
marci@941
|
204 |
for (typename Graph::InEdgeIt e(g, s); e!=INVALID; ++e) i-=flow[e];
|
alpar@899
|
205 |
return i;
|
alpar@899
|
206 |
}
|
alpar@899
|
207 |
|
athos@1527
|
208 |
/// Total cost of the found flow.
|
alpar@899
|
209 |
|
athos@1527
|
210 |
/// This function gives back the total cost of the found flow.
|
alpar@899
|
211 |
Length totalLength(){
|
alpar@899
|
212 |
return total_length;
|
alpar@899
|
213 |
}
|
alpar@899
|
214 |
|
alpar@899
|
215 |
///Returns a const reference to the EdgeMap \c flow.
|
alpar@899
|
216 |
|
alpar@899
|
217 |
///Returns a const reference to the EdgeMap \c flow.
|
alpar@899
|
218 |
const EdgeIntMap &getFlow() const { return flow;}
|
alpar@899
|
219 |
|
marci@941
|
220 |
/*! \brief Returns a const reference to the NodeMap \c potential (the dual solution).
|
alpar@899
|
221 |
|
marci@941
|
222 |
Returns a const reference to the NodeMap \c potential (the dual solution).
|
marci@941
|
223 |
*/
|
alpar@899
|
224 |
const PotentialMap &getPotential() const { return potential;}
|
alpar@899
|
225 |
|
marci@941
|
226 |
/*! \brief Checking the complementary slackness optimality criteria.
|
alpar@899
|
227 |
|
marci@941
|
228 |
This function checks, whether the given flow and potential
|
jacint@1270
|
229 |
satisfy the complementary slackness conditions (i.e. these are optimal).
|
marci@941
|
230 |
This function only checks optimality, doesn't bother with feasibility.
|
marci@941
|
231 |
For testing purpose.
|
marci@941
|
232 |
*/
|
alpar@899
|
233 |
bool checkComplementarySlackness(){
|
alpar@899
|
234 |
Length mod_pot;
|
alpar@899
|
235 |
Length fl_e;
|
marci@941
|
236 |
for(typename Graph::EdgeIt e(g); e!=INVALID; ++e) {
|
alpar@899
|
237 |
//C^{\Pi}_{i,j}
|
alpar@986
|
238 |
mod_pot = length[e]-potential[g.target(e)]+potential[g.source(e)];
|
alpar@899
|
239 |
fl_e = flow[e];
|
alpar@899
|
240 |
if (0<fl_e && fl_e<capacity[e]) {
|
alpar@899
|
241 |
/// \todo better comparison is needed for real types, moreover,
|
alpar@899
|
242 |
/// this comparison here is superfluous.
|
alpar@899
|
243 |
if (mod_pot != 0)
|
alpar@899
|
244 |
return false;
|
alpar@899
|
245 |
}
|
alpar@899
|
246 |
else {
|
alpar@899
|
247 |
if (mod_pot > 0 && fl_e != 0)
|
alpar@899
|
248 |
return false;
|
alpar@899
|
249 |
if (mod_pot < 0 && fl_e != capacity[e])
|
alpar@899
|
250 |
return false;
|
alpar@899
|
251 |
}
|
alpar@899
|
252 |
}
|
alpar@899
|
253 |
return true;
|
alpar@899
|
254 |
}
|
alpar@899
|
255 |
|
alpar@899
|
256 |
}; //class MinCostFlow
|
alpar@899
|
257 |
|
alpar@899
|
258 |
///@}
|
alpar@899
|
259 |
|
alpar@921
|
260 |
} //namespace lemon
|
alpar@899
|
261 |
|
alpar@921
|
262 |
#endif //LEMON_MIN_COST_FLOW_H
|