doc/basic_concepts.dox
author kpeter
Fri, 06 Feb 2009 21:52:34 +0000
changeset 2634 e98bbe64cca4
parent 2476 059dcdda37c5
permissions -rw-r--r--
Rework Network Simplex
Use simpler and faster graph implementation instead of SmartGraph
alpar@2391
     1
/* -*- C++ -*-
alpar@2391
     2
 *
alpar@2391
     3
 * This file is a part of LEMON, a generic C++ optimization library
alpar@2391
     4
 *
alpar@2553
     5
 * Copyright (C) 2003-2008
alpar@2391
     6
 * Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
alpar@2391
     7
 * (Egervary Research Group on Combinatorial Optimization, EGRES).
alpar@2391
     8
 *
alpar@2391
     9
 * Permission to use, modify and distribute this software is granted
alpar@2391
    10
 * provided that this copyright notice appears in all copies. For
alpar@2391
    11
 * precise terms see the accompanying LICENSE file.
alpar@2391
    12
 *
alpar@2391
    13
 * This software is provided "AS IS" with no warranty of any kind,
alpar@2391
    14
 * express or implied, and with no claim as to its suitability for any
alpar@2391
    15
 * purpose.
alpar@2391
    16
 *
alpar@2391
    17
 */
alpar@2391
    18
alpar@2350
    19
namespace lemon {
alpar@2350
    20
alpar@2195
    21
/**
alpar@2195
    22
\page basic_concepts Basic concepts
alpar@2195
    23
alpar@2195
    24
\section basic_graph The graph classes
athos@2288
    25
The most important classes in LEMON are the graph classes. An instance of a graph
alpar@2195
    26
class is the representation of the mathematical graph. It holds the topology and
alpar@2195
    27
every structural information of the graph. The structural manipulations are also
alpar@2195
    28
provided by the graph object. There is no universal graph class instead we have
alpar@2195
    29
different classes for different purposes. They can differ in many ways, but all
alpar@2195
    30
have to satisfy one or more \ref concept "graph concepts" which are standardized
athos@2288
    31
interfaces to work with the rest of the library. The most basic concept is the
kpeter@2476
    32
\ref concepts::Graph "Graph".<br>
alpar@2195
    33
A good example is the \ref ListGraph which we already know from Hello World and
alpar@2195
    34
will be used in our examples as well.
alpar@2195
    35
alpar@2195
    36
One main advantage of the templates are, that you can write your own graph classes.
alpar@2195
    37
As long as they provide the interface a concept is defining all the LEMON algorithms
alpar@2195
    38
and classes will work with it properly - no representation or implementation is
alpar@2195
    39
written into stone.
alpar@2195
    40
alpar@2195
    41
alpar@2195
    42
\subsection basic_node Nodes
alpar@2195
    43
To refer to a node of a graph we need some kind of typed variable. Graph classes
alpar@2195
    44
have the Node public type for this purpose. Stacking by the last example:
alpar@2195
    45
\code lemon::ListGraph::Node \endcode
alpar@2195
    46
alpar@2195
    47
If the graph fits the ExtendableGraphComponent concept, then you can add new nodes
alpar@2195
    48
to the graph with the addNode() member function. It returns the newly added node
athos@2288
    49
(as value). So if you need the new node to do something useful with, for example
kpeter@2476
    50
create an edge, assign a value to it through \ref maps1 maps.
alpar@2195
    51
\code lemon::ListGraph::Node  new_node = graph.addNode(); \endcode
alpar@2195
    52
athos@2288
    53
If the graph fits into the ErasableGraphComponent concept you can also remove nodes
alpar@2195
    54
from the graph with the erase() member function.
alpar@2195
    55
\code graph.erase( new_node ); \endcode
alpar@2195
    56
alpar@2195
    57
You don't have to store every node in a variable, you can access individual nodes
alpar@2195
    58
with node iterators discussed in the next section. But how do you know which
alpar@2195
    59
node is which?<br>
alpar@2195
    60
The graph class has the id( Node n ) member function providing an unique identifier
alpar@2195
    61
assigned to every node.
alpar@2195
    62
alpar@2195
    63
alpar@2195
    64
\subsection basic_edge Edges
alpar@2195
    65
An Edge is what you think it is. It goes from one node to another node (they can
athos@2288
    66
be identical if the edge is a loop). If the graph class is directed, the Edge is directed too. Otherwise
alpar@2195
    67
the edge is considered undirected and called UEdge.
alpar@2195
    68
\code lemon::ListUGraph::UEdge \endcode
alpar@2195
    69
alpar@2195
    70
The addEdge() member function will create a new edge. It has two arguments, the
alpar@2195
    71
source node and the target node. The graph class must be extendable.
alpar@2195
    72
\code lemon::ListGraph::Edge  new_edge = graph.addEdge( src_node, trg_node ); \endcode
athos@2288
    73
You can handle edges similar as nodes. The erase() member function has an edge taking
alpar@2195
    74
overload too.
alpar@2195
    75
alpar@2195
    76
You can ask for the source or target node of the edge by the corresponding member
alpar@2195
    77
functions:
alpar@2195
    78
\code
alpar@2195
    79
graph.source( new_edge );
alpar@2195
    80
lemon::ListGraph::Node  n = graph.target( new_edge ); \endcode
alpar@2195
    81
These functions are always legal even if the graph is undirected. UEdge has a
alpar@2195
    82
default direction.
alpar@2195
    83
alpar@2195
    84
alpar@2195
    85
\section basic_iterators Iterators
alpar@2195
    86
Graphs are some kind of containers. And as you expect they have iterator types.
athos@2288
    87
One for nodes and a couple for edges - and special classes can have additional
alpar@2195
    88
iterators too. An example:
alpar@2195
    89
\code lemon::ListGraph::NodeIt \endcode
athos@2288
    90
This is a node iterator. Every iterator type starts with a name that refers to
athos@2288
    91
the iterated object, and ends with 'It'.
alpar@2195
    92
athos@2288
    93
LEMON style iterators differ from \c stl or \c boost iterators in a very tasty
alpar@2195
    94
way. A graph has no begin or end - or at least a generic graph class has none.
alpar@2195
    95
If by some topology you could pick a good begin node, it would be misleading and
alpar@2195
    96
incorrect. A LEMON style iterator must be initialized at construction time.
alpar@2195
    97
The constructor takes the needed parameters - by a node iterator it's the graph
alpar@2195
    98
object. And will be compared to the lemon::INVALID to check if it's still valid.
alpar@2195
    99
Every iterator can be compared to INVALID. No \c begin() or \c end() needed.<br>
alpar@2195
   100
Let's see these things working together:
alpar@2195
   101
\code
alpar@2195
   102
for( ListGraph::NodeIt n(graph); n != INVALID; ++n )
athos@2288
   103
    do_useful_things_with_node(n);
alpar@2195
   104
\endcode
alpar@2195
   105
Note that the function \c do_useful_things_with_node() expects a Node type argument
alpar@2195
   106
ad we just gave him the iterator. LEMON style iterators must provide "on demand
alpar@2195
   107
dereferencing". For example a NodeIt can be used everywhere a Node could. (In some
alpar@2195
   108
graph classes Node is the base class of NodeIt. But in other cases this is implemented
alpar@2195
   109
through typecast operator.)
alpar@2195
   110
alpar@2195
   111
<b>Very important!</b> The iteration has no defined order. There is absolutely no
athos@2288
   112
warranty that the next time the iteration will give us the nodes in the same order.
alpar@2195
   113
Don't use this order to identify nodes! Use the \c id() member function of the
alpar@2195
   114
graph class described above. (There is a powerful technique using maps right in
alpar@2195
   115
the next page.)
alpar@2195
   116
kpeter@2476
   117
The \ref concepts::Graph::EdgeIt "EdgeIt" works exactly the same - nothing more to say.
kpeter@2476
   118
But there are \ref concepts::Graph::InEdgeIt "InEdgeIt" and
kpeter@2476
   119
\ref concepts::Graph::OutEdgeIt "OutEdgeIt" by directed graphs and
kpeter@2476
   120
\ref concepts::UGraph::IncEdgeIt "IncEdgeIt" by undirected graphs.
alpar@2195
   121
They take two arguments. The first is a graph, the second is certain node of the
alpar@2195
   122
graph. InEdgeIt iterates on the incoming edges of that node and OutEdgeIt does it
alpar@2195
   123
on the outgoing edges. The IncEdgeIt of course iterates every edge connecting to
alpar@2195
   124
the given node.
alpar@2195
   125
alpar@2195
   126
\code
alpar@2195
   127
for( ListGraph::NodeIt n(graph); n != INVALID; ++n ) {
alpar@2195
   128
  int in = 0, out = 0;
alpar@2195
   129
  for( ListGraph::InEdgeIt e(graph,n); e != INVALID; ++e ) ++in;
alpar@2195
   130
  for( ListGraph::OutEdgeIt e(graph,n); e != INVALID; ++e ) ++out;
alpar@2195
   131
alpar@2195
   132
  std::cout << "#" << graph.id(n) << " node has " << in << " incoming and "
alpar@2195
   133
    << out << "outgoing edges." << std::endl;
alpar@2195
   134
}
alpar@2195
   135
\endcode
alpar@2195
   136
alpar@2195
   137
alpar@2195
   138
\section basic_ListGraph ListGraph - a versatile directed graph
alpar@2195
   139
As you see ListGraph satisfies most of the basic concepts and ideal for general
alpar@2195
   140
graph representations. It has an undirected version too: ListUGraph.
alpar@2195
   141
*/
alpar@2350
   142
alpar@2391
   143
}