quickstart.rst 27.5 KB
Newer Older
1
2
3
Quick start using `graph-tool`
==============================

4
5
6
7
The :mod:`graph_tool` module provides a :class:`~graph_tool.Graph` class
and several algorithms that operate on it. The internals of this class,
and of most algorithms, are written in C++ for performance, using the
`Boost Graph Library <http://www.boost.org>`_.
8

9
10
11
The module must be of course imported before it can be used. The package is
subdivided into several sub-modules. To import everything from all of them, one
can do:
12

13
14
15
16
17
.. testsetup::

   np.random.seed(42)
   gt.seed_rng(42)

18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
.. doctest::

   >>> from graph_tool.all import *

In the following, it will always be assumed that the previous line was run.

Creating and manipulating graphs
--------------------------------

An empty graph can be created by instantiating a :class:`~graph_tool.Graph`
class:

.. doctest::

   >>> g = Graph()

By default, newly created graphs are always directed. To construct undirected
35
graphs, one must pass a value to the ``directed`` parameter:
36
37
38
39
40

.. doctest::

   >>> ug = Graph(directed=False)

41
42
43
44
A graph can always be switched *on-the-fly* from directed to undirected
(and vice versa), with the :meth:`~graph_tool.Graph.set_directed`
method. The "directedness" of the graph can be queried with the
:meth:`~graph_tool.Graph.is_directed` method,
45
46
47
48
49
50
51

.. doctest::

   >>> ug = Graph()
   >>> ug.set_directed(False)
   >>> assert(ug.is_directed() == False)

52
53
54
A graph can also be created by providing another graph, in which case
the entire graph (and its internal property maps, see
:ref:`sec_property_maps`) is copied:
55
56
57
58

.. doctest::

   >>> g1 = Graph()
59
   >>> # ... construct g1 ...
60
61
   >>> g2 = Graph(g1)                 # g1 and g2 are copies

62
63
64
65
66
67
68
69
70
Above, ``g2`` is a "deep" copy of ``g1``, i.e. any modification of
``g2`` will not affect ``g1``.

Once a graph is created, it can be populated with vertices and edges. A
vertex can be added with the :meth:`~graph_tool.Graph.add_vertex`
method, which returns an instance of a :class:`~graph_tool.Vertex`
class, also called a *vertex descriptor*. For instance, the following
code creates two vertices, and returns vertex descriptors stored in the
variables ``v1`` and ``v2``.
71
72
73

.. doctest::

74
75
   >>> v1 = g.add_vertex()
   >>> v2 = g.add_vertex()
76

77
78
79
Edges can be added in an analogous manner, by calling the
:meth:`~graph_tool.Graph.add_edge` method, which returns an edge
descriptor (an instance of the :class:`~graph_tool.Edge` class):
80
81
82

.. doctest::

83
   >>> e = g.add_edge(v1, v2)
84

85
86
87
The above code creates a directed edge from ``v1`` to ``v2``. We can
visualize the graph we created so far with the
:func:`~graph_tool.draw.graph_draw` function.
88
89
90

.. doctest::

91
   >>> graph_draw(g, vertex_text=g.vertex_index, vertex_font_size=18,
Tiago Peixoto's avatar
Tiago Peixoto committed
92
   ...            output_size=(200, 200), output="two-nodes.png")
93
94
   <...>

95
96
97
98
.. doctest::
   :hide:

   graph_draw(g, vertex_text=g.vertex_index, vertex_font_size=18,
Tiago Peixoto's avatar
Tiago Peixoto committed
99
              output_size=(200, 200), output="two-nodes.pdf")
100
101


102
.. figure:: two-nodes.*
103
104
105
106
   :align: center

   A simple directed graph with two vertices and one edge, created by
   the commands above.
107

108
109
110
111
With vertex and edge descriptors, one can examine and manipulate the
graph in an arbitrary manner. For instance, in order to obtain the
out-degree of a vertex, we can simply call the
:meth:`~graph_tool.Vertex.out_degree` method:
112
113
114

.. doctest::

Tiago Peixoto's avatar
Tiago Peixoto committed
115
   >>> print(v1.out_degree())
116
117
118
119
   1

Analogously, we could have used the :meth:`~graph_tool.Vertex.in_degree`
method to query the in-degree.
120

121
122
123
124
.. note::

   For undirected graphs, the "out-degree" is synonym for degree, and
   in this case the in-degree of a vertex is always zero.
125

126
127
128
Edge descriptors have two useful methods, :meth:`~graph_tool.Edge.source`
and :meth:`~graph_tool.Edge.target`, which return the source and target
vertex of an edge, respectively.
129
130
131

.. doctest::

Tiago Peixoto's avatar
Tiago Peixoto committed
132
   >>> print(e.source(), e.target())
133
   0 1
134

135
136
The :meth:`~graph_tool.Graph.add_vertex` method also accepts an optional
parameter which specifies the number of vertices to create. If this
137
138
value is greater than 1, it returns an iterator on the added vertex
descriptors:
139
140
141

.. doctest::

142
   >>> vlist = g.add_vertex(10)
143
   >>> print(len(list(vlist)))
144
145
   10

146
147
148
149
150
151
152
153
154
155
156
Each vertex in a graph has an unique index, which is *always* between
:math:``0`` and :math:``N-1``, where :math:``N`` is the number of
vertices. This index can be obtained by using the
:attr:`~graph_tool.Graph.vertex_index` attribute of the graph (which is
a *property map*, see :ref:`sec_property_maps`), or by converting the
vertex descriptor to an ``int``.

.. doctest::

   >>> v = g.add_vertex()
   >>> print(g.vertex_index[v])
Tiago Peixoto's avatar
Tiago Peixoto committed
157
   12
158
   >>> print(int(v))
Tiago Peixoto's avatar
Tiago Peixoto committed
159
   12
160
161

   
162
163
164
165
166
167
168
169
Edges and vertices can also be removed at any time with the
:meth:`~graph_tool.Graph.remove_vertex` and :meth:`~graph_tool.Graph.remove_edge` methods,

.. doctest::

   >>> g.remove_edge(e)                               # e no longer exists
   >>> g.remove_vertex(v2)                # the second vertex is also gone

170
   
171
.. note::
172

173
174
175
176
177
178
179
180
181
182
   Removing a vertex is typically an :math:`O(N)` operation. The
   vertices are internally stored in a `STL vector
   <http://en.wikipedia.org/wiki/Vector_%28STL%29>`_, so removing an
   element somewhere in the middle of the list requires the shifting of
   the rest of the list. Thus, fast :math:`O(1)` removals are only
   possible either if one can guarantee that only vertices in the end of
   the list are removed (the ones last added to the graph), or if the
   relative vertex ordering is invalidated. This last behavior can be
   achieved by passing the option ``fast == True``, to
   :meth:`~graph_tool.Graph.remove_vertex`, which causes the vertex
Tiago Peixoto's avatar
Tiago Peixoto committed
183
184
185
186
   being deleted to be 'swapped' with the last vertex (i.e. with the
   largest index), which will in turn inherit the index of the vertex
   being deleted.

187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
.. warning::

   Because of the above, removing a vertex with an index smaller than
   :math:`N-1` will **invalidate either the last** (``fast = True``)
   **or all** (``fast = False``) **descriptors pointing to vertices with
   higher index**.

   As a consequence, if more than one vertex is to be removed at a given
   time, they should **always** be removed in decreasing index order:
   
   .. code::

       # 'del_list' is a list of vertex descriptors
       for v in reversed(sorted(del_list)):
           g.remove_vertex(v)

   Alternatively (and preferably), a list (or any iterable) may be
   passed directly as the ``vertex`` parameter of the
   :meth:`~graph_tool.Graph.remove_vertex` function, and the above is
   performed internally (in C++).

   Note that property map values (see :ref:`sec_property_maps`) are
   unaffected by the index changes due to vertex removal.

.. note::
212

213
   Removing an edge is an :math:`O(k_{s} + k_{t})` operation, where
Tiago Peixoto's avatar
Tiago Peixoto committed
214
   :math:`k_{s}` is the out-degree of the source vertex, and
Tiago Peixoto's avatar
Tiago Peixoto committed
215
216
217
218
   :math:`k_{t}` is the in-degree of the target vertex. This can be made
   faster by setting :meth:`~graph_tool.Graph.set_fast_edge_removal` to
   `True`, in which case it becomes :math:`O(1)`, at the expense of
   additional data of size :math:`O(E)`.
219

220
   No edge descriptors are ever invalidated after edge removal.
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249

Since vertices are uniquely identifiable by their indexes, there is no
need to keep the vertex descriptor lying around to access them at a
later point. If we know its index, one can obtain the descriptor of a
vertex with a given index using the :meth:`~graph_tool.Graph.vertex`
method,

.. doctest::

   >>> v = g.vertex(8)

which takes an index, and returns a vertex descriptor. Edges cannot be
directly obtained by its index, but if the source and target vertices of
a given edge is known, it can be obtained with the
:meth:`~graph_tool.Graph.edge` method

.. doctest::

   >>> g.add_edge(g.vertex(2), g.vertex(3))
   <...>
   >>> e = g.edge(2, 3)


Another way to obtain edge or vertex descriptors is to *iterate* through
them, as described in section :ref:`sec_iteration`. This is in fact the
most useful way of obtaining vertex and edge descriptors.

Like vertices, edges also have unique indexes, which are given by the
:attr:`~graph_tool.Graph.edge_index` property:
250
251
252
253

.. doctest::

   >>> e = g.add_edge(g.vertex(0), g.vertex(1))
Tiago Peixoto's avatar
Tiago Peixoto committed
254
   >>> print(g.edge_index[e])
255
256
257
258
259
260
261
262
263
   1

Differently from vertices, edge indexes do not necessarily conform to
any specific range. If no edges are ever removed, the indexes will be in
the range :math:`[0, E-1]`, where :math:`E` is the number of edges, and
edges added earlier have lower indexes. However if an edge is removed,
its index will be "vacant", and the remaining indexes will be left
unmodified, and thus will not lie in the range :math:`[0, E-1]`.  If a
new edge is added, it will reuse old indexes, in an increasing order.
264

265

266
267
268
.. _sec_iteration:

Iterating over vertices and edges
269
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
270

271
272
273
274
Algorithms must often iterate through vertices, edges, out-edges of a
vertex, etc. The :class:`~graph_tool.Graph` and
:class:`~graph_tool.Vertex` classes provide different types of iterators
for doing so. The iterators always point to edge or vertex descriptors.
275

276
277
278
279
280
281
Iterating over all vertices or edges
""""""""""""""""""""""""""""""""""""

In order to iterate through all the vertices or edges of a graph, the
:meth:`~graph_tool.Graph.vertices` and :meth:`~graph_tool.Graph.edges`
methods should be used:
282
283
284
285

.. doctest::

   for v in g.vertices():
Tiago Peixoto's avatar
Tiago Peixoto committed
286
       print(v)
Tiago Peixoto's avatar
Tiago Peixoto committed
287
   for e in g.edges():
Tiago Peixoto's avatar
Tiago Peixoto committed
288
       print(e)
289
290
291
292

The code above will print the vertices and edges of the graph in the order they
are found.

293
294
295
Iterating over the neighbourhood of a vertex
""""""""""""""""""""""""""""""""""""""""""""

296
297
298
The out- and in-edges of a vertex, as well as the out- and in-neighbours can be
iterated through with the :meth:`~graph_tool.Vertex.out_edges`,
:meth:`~graph_tool.Vertex.in_edges`, :meth:`~graph_tool.Vertex.out_neighbours`
299
and :meth:`~graph_tool.Vertex.in_neighbours` methods, respectively.
300
301
302
303
304
305

.. doctest::

   from itertools import izip
   for v in g.vertices():
      for e in v.out_edges():
Tiago Peixoto's avatar
Tiago Peixoto committed
306
          print(e)
307
      for w in v.out_neighbours():
Tiago Peixoto's avatar
Tiago Peixoto committed
308
          print(w)
309
310
311
312
313

      # the edge and neighbours order always match
      for e,w in izip(v.out_edges(), v.out_neighbours()):
          assert(e.target() == w)

314
315
316
317
318
319
320
321
322
323
324
325
326
The code above will print the out-edges and out-neighbours of all
vertices in the graph.

.. note::

   The ordering of the vertices and edges visited by the iterators is
   always the same as the order in which they were added to the graph
   (with the exception of the iterator returned by
   :meth:`~graph_tool.Graph.edges`). Usually, algorithms do not care
   about this order, but if it is ever needed, this inherent ordering
   can be relied upon.

.. warning::
327

328
329
330
331
332
   You should never remove vertex or edge descriptors when iterating
   over them, since this invalidates the iterators. If you plan to
   remove vertices or edges during iteration, you must first store them
   somewhere (such as in a list) and remove them only after no iterator
   is being used. Removal during iteration will cause bad things to
333
   happen.
334
   
335

336
337
338
339
340
.. _sec_property_maps:

Property maps
-------------

341
342
343
344
345
346
Property maps are a way of associating additional information to the
vertices, edges or to the graph itself. There are thus three types of
property maps: vertex, edge and graph. All of them are handled by the
same class, :class:`~graph_tool.PropertyMap`. Each created property map
has an associated *value type*, which must be chosen from the predefined
set:
347
348
349
350
351

.. tabularcolumns:: |l|l|

.. table::

Tiago Peixoto's avatar
Tiago Peixoto committed
352
353
354
355
    ========================     ======================
     Type name                   Alias
    ========================     ======================
    ``bool``                     ``uint8_t``
356
    ``int16_t``                  ``short``
Tiago Peixoto's avatar
Tiago Peixoto committed
357
358
359
    ``int32_t``                  ``int``
    ``int64_t``                  ``long``, ``long long``
    ``double``                   ``float``
Tiago Peixoto's avatar
Tiago Peixoto committed
360
361
    ``long double``
    ``string``
Tiago Peixoto's avatar
Tiago Peixoto committed
362
    ``vector<bool>``             ``vector<uint8_t>``
363
    ``vector<int16_t>``          ``vector<short>``
Tiago Peixoto's avatar
Tiago Peixoto committed
364
365
366
    ``vector<int32_t>``          ``vector<int>``
    ``vector<int64_t>``          ``vector<long>``, ``vector<long long>``
    ``vector<double>``           ``vector<float>``
367
368
    ``vector<long double>``
    ``vector<string>``
Tiago Peixoto's avatar
Tiago Peixoto committed
369
370
    ``python::object``           ``object``
    ========================     ======================
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387

New property maps can be created for a given graph by calling the
:meth:`~graph_tool.Graph.new_vertex_property`, :meth:`~graph_tool.Graph.new_edge_property`, or
:meth:`~graph_tool.Graph.new_graph_property`, for each map type. The values are then
accessed by vertex or edge descriptors, or the graph itself, as such:

.. doctest::

    from itertools import izip
    from numpy.random import randint

    g = Graph()
    g.add_vertex(100)
    # insert some random links
    for s,t in izip(randint(0, 100, 100), randint(0, 100, 100)):
        g.add_edge(g.vertex(s), g.vertex(t))

388
    vprop_double = g.new_vertex_property("double")            # Double-precision floating point
389
390
    vprop_double[g.vertex(10)] = 3.1416

391
    vprop_vint = g.new_vertex_property("vector<int>")         # Vector of ints
392
393
    vprop_vint[g.vertex(40)] = [1, 3, 42, 54]
    
394
395
    eprop_dict = g.new_edge_property("object")                # Arbitrary python object.
    eprop_dict[g.edges().next()] = {"foo": "bar", "gnu": 42}  # In this case, a dict.
396

397
    gprop_bool = g.new_edge_property("bool")                  # Boolean
398
399
    gprop_bool[g] = True

400
401
Property maps with scalar value types can also be accessed as a
:class:`numpy.ndarray`, with the
Tiago Peixoto's avatar
Tiago Peixoto committed
402
:meth:`~graph_tool.PropertyMap.get_array` method, or the
403
:attr:`~graph_tool.PropertyMap.a` attribute, i.e.,
404
405
406
407
408

.. doctest::

    from numpy.random import random

409
    # this assigns random values to the vertex properties
410
411
    vprop_double.get_array()[:] = random(g.num_vertices())

412
413
414
    # or more conveniently (this is equivalent to the above)
    vprop_double.a = random(g.num_vertices())

415
416
.. _sec_internal_props:

417
Internal property maps
418
^^^^^^^^^^^^^^^^^^^^^^
419

420
421
422
423
424
Any created property map can be made "internal" to the corresponding
graph. This means that it will be copied and saved to a file together
with the graph. Properties are internalized by including them in the
graph's dictionary-like attributes
:attr:`~graph_tool.Graph.vertex_properties`,
425
:attr:`~graph_tool.Graph.edge_properties` or
426
427
428
:attr:`~graph_tool.Graph.graph_properties` (or their aliases,
:attr:`~graph_tool.Graph.vp`, :attr:`~graph_tool.Graph.ep` or
:attr:`~graph_tool.Graph.gp`, respectively). When inserted in the graph,
429
430
the property maps must have an unique name (between those of the same
type):
431
432
433
434
435
436
437
438

.. doctest::

    >>> eprop = g.new_edge_property("string")
    >>> g.edge_properties["some name"] = eprop
    >>> g.list_properties()
    some name      (edge)    (type: string)

439
440
441
442
443
444
445
446
447
448
449
450
451
Internal graph property maps behave slightly differently. Instead of
returning the property map object, the value itself is returned from the
dictionaries:

.. doctest::

    >>> gprop = g.new_graph_property("int")
    >>> g.graph_properties["foo"] = gprop   # this sets the actual property map
    >>> g.graph_properties["foo"] = 42      # this sets its value
    >>> print(g.graph_properties["foo"])
    42
    >>> del g.graph_properties["foo"]       # the property map entry is deleted from the dictionary

452
453
454
455
456
For convenience, the internal property maps can also be accessed via
attributes:

.. doctest::

Tiago Peixoto's avatar
Tiago Peixoto committed
457
458
459
460
461
462
    >>> vprop = g.new_vertex_property("double")
    >>> g.vp.foo = vprop                        # equivalent to g.vertex_properties["foo"] = vprop
    >>> v = g.vertex(0)
    >>> g.vp.foo[v] = 3.14
    >>> print(g.vp.foo[v])
    3.14
463

464
.. _sec_graph_io:
465
466
467
468

Graph I/O
---------

469
Graphs can be saved and loaded in four formats: `graphml
470
<http://graphml.graphdrawing.org/>`_, `dot
471
472
<http://www.graphviz.org/doc/info/lang.html>`_, `gml
<http://www.fim.uni-passau.de/en/fim/faculty/chairs/theoretische-informatik/projects.html>`_
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
and a custom binary format ``gt`` (see :ref:`sec_gt_format`). 

.. warning::

    The binary format ``gt`` and ``graphml`` are the preferred formats,
    since they are by far the most complete. Both these formats are
    equally complete, but the ``gt`` format is faster and requires less
    storage.

    The ``dot`` and ``gml`` formats are fully supported, but since they
    contain no precise type information, all properties are read as
    strings (or also as double, in the case of ``gml``), and must be
    converted by hand to the desired type. Therefore you should always
    use either ``gt`` or ``graphml``, since they implement an exact
    bit-for-bit representation of all supported :ref:`sec_property_maps`
    types, except when interfacing with other software, or existing
    data, which uses ``dot`` or ``gml``.
490
491
492

A graph can be saved or loaded to a file with the :attr:`~graph_tool.Graph.save`
and :attr:`~graph_tool.Graph.load` methods, which take either a file name or a
493
file-like object. A graph can also be loaded from disc with the
494
495
496
497
498
499
500
501
:func:`~graph_tool.load_graph` function, as such:

.. doctest::

    g = Graph()
    #  ... fill the graph ...
    g.save("my_graph.xml.gz")    
    g2 = load_graph("my_graph.xml.gz")
502
    # g and g2 should be copies of each other
503
504
505
506
507
508
509

Graph classes can also be pickled with the :mod:`pickle` module.


An Example: Building a Price Network
------------------------------------

510
511
A Price network is the first known model of a "scale-free" graph,
invented in 1976 by `de Solla Price
512
<http://en.wikipedia.org/wiki/Derek_J._de_Solla_Price>`_. It is defined
513
514
515
516
517
518
519
520
521
522
523
524
dynamically, where at each time step a new vertex is added to the graph,
and connected to an old vertex, with probability proportional to its
in-degree. The following program implements this construction using
``graph-tool``.

.. note::

   Note that it would be much faster just to use the
   :func:`~graph_tool.generation.price_network` function, which is
   implemented in C++, as opposed to the script below which is in pure
   python. The code below is merely a demonstration on how to use the
   library.
525

526
527
528
.. literalinclude:: price.py
   :linenos:

529
530
The following is what should happen when the program is run.

531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
.. testcode::
   :hide:

   from price import *
   clf()

.. testoutput::

    vertex: 36063 in-degree: 0 out-degree: 1 age: 36063
    vertex: 9075 in-degree: 4 out-degree: 1 age: 9075
    vertex: 5967 in-degree: 3 out-degree: 1 age: 5967
    vertex: 1113 in-degree: 7 out-degree: 1 age: 1113
    vertex: 25 in-degree: 84 out-degree: 1 age: 25
    vertex: 10 in-degree: 541 out-degree: 1 age: 10
    vertex: 5 in-degree: 140 out-degree: 1 age: 5
    vertex: 2 in-degree: 459 out-degree: 1 age: 2
    vertex: 1 in-degree: 520 out-degree: 1 age: 1
    vertex: 0 in-degree: 210 out-degree: 0 age: 0
    Nowhere else to go... We found the main hub!

Tiago Peixoto's avatar
Tiago Peixoto committed
551
552
553
Below is the degree distribution, with 100000 nodes. If you want to see
a broader power law, try to increase the number of vertices to something
like :math:`10^6` or :math:`10^7`.
554

Tiago Peixoto's avatar
Tiago Peixoto committed
555
.. figure:: price-deg-dist.*
556
   :align: center
557

Tiago Peixoto's avatar
Tiago Peixoto committed
558
   In-degree distribution of a price network with :math:`10^5` nodes.
559

560
561
562
563
564
565
566

We can draw the graph to see some other features of its topology. For that we
use the :func:`~graph_tool.draw.graph_draw` function.

.. testcode::

   g = load_graph("price.xml.gz")
567
   age = g.vertex_properties["age"]
568

569
570
   pos = sfdp_layout(g)
   graph_draw(g, pos, output_size=(1000, 1000), vertex_color=[1,1,1,0],
571
              vertex_fill_color=age, vertex_size=1, edge_pen_width=1.2,
572
              vcmap=matplotlib.cm.gist_heat_r, output="price.png")
573

574
.. figure:: price.*
575
576
   :align: center

577
   A Price network with :math:`10^5` nodes. The vertex colors represent
578
   the age of the vertex, from older (red) to newer (black).
579

580
581
.. _sec_graph_filtering:

582
583
584
Graph filtering
---------------

585
586
587
588
One of the very nice features of ``graph-tool`` is the "on-the-fly" filtering of
edges and/or vertices. Filtering means the temporary masking of vertices/edges,
which are in fact not really removed, and can be easily recovered. Vertices or
edges which are to be filtered should be marked with a
589
590
591
592
593
594
595
596
:class:`~graph_tool.PropertyMap` with value type ``bool``, and then set with
:meth:`~graph_tool.Graph.set_vertex_filter` or
:meth:`~graph_tool.Graph.set_edge_filter` methods. By default, vertex or edges
with value "1" are `kept` in the graphs, and those with value "0" are filtered
out. This behaviour can be modified with the ``inverted`` parameter of the
respective functions. All manipulation functions and algorithms will work as if
the marked edges or vertices were removed from the graph, with minimum overhead.

Tiago Peixoto's avatar
Tiago Peixoto committed
597
598
599
600
601
602
.. note::

    It is important to emphasize that the filtering functionality does not add
    any overhead when the graph is not being filtered. In this case, the
    algorithms run just as fast as if the filtering functionality didn't exist.

603
604
Here is an example which obtains the minimum spanning tree of a graph,
using the function :func:`~graph_tool.topology.min_spanning_tree` and
605
606
607
608
609
edge filtering.

.. testcode::
   :hide:

610
   from numpy.random import *
611
612
613
614
   seed(42)

.. testcode::

615
   g, pos = triangulation(random((500, 2)) * 4, type="delaunay")
616
   tree = min_spanning_tree(g)
617
618
619
620
621
   graph_draw(g, pos=pos, edge_color=tree, output="min_tree.pdf")

.. testcode::
   :hide:

622
   graph_draw(g, pos=pos, edge_color=tree, output_size=(400, 400),
623
624
              output="min_tree.png")

625
626
627
628
629

The ``tree`` property map has a bool type, with value "1" if the edge belongs to
the tree, and "0" otherwise. Below is an image of the original graph, with the
marked edges.

630
.. figure:: min_tree.*
631
632
633
634
635
636
   :align: center

We can now filter out the edges which don't belong to the minimum spanning tree.

.. testcode::

637
638
639
640
641
642
643
   g.set_edge_filter(tree)
   graph_draw(g, pos=pos, output="min_tree_filtered.pdf")

.. testcode::
   :hide:

   graph_draw(g, pos=pos, output_size=(400, 400), output="min_tree_filtered.png")
644
645
646

This is how the graph looks when filtered:

647
.. figure:: min_tree_filtered.*
648
649
650
   :align: center

Everything should work transparently on the filtered graph, simply as if the
Tiago Peixoto's avatar
Tiago Peixoto committed
651
652
653
masked edges were removed. For instance, the following code will calculate the
:func:`~graph_tool.centrality.betweenness` centrality of the edges and vertices,
and draws them as colors and line thickness in the graph.
654
655
656

.. testcode::

Tiago Peixoto's avatar
Tiago Peixoto committed
657
    bv, be = betweenness(g)
658
659
    be.a /= be.a.max() / 5
    graph_draw(g, pos=pos, vertex_fill_color=bv, edge_pen_width=be,
660
661
662
663
664
665
666
               output="filtered-bt.pdf")

.. testcode::
   :hide:

   graph_draw(g, pos=pos, vertex_fill_color=bv, edge_pen_width=be,
              output_size=(400, 400), output="filtered-bt.png")
667

668
.. figure:: filtered-bt.*
Tiago Peixoto's avatar
Tiago Peixoto committed
669
   :align: center
670
671
672
673
674
675
676


The original graph can be recovered by setting the edge filter to ``None``.

.. testcode::

    g.set_edge_filter(None)
Tiago Peixoto's avatar
Tiago Peixoto committed
677
    bv, be = betweenness(g)
678
679
    be.a /= be.a.max() / 5
    graph_draw(g, pos=pos, vertex_fill_color=bv, edge_pen_width=be,
680
681
682
683
684
685
686
               output="nonfiltered-bt.pdf")

.. testcode::
   :hide:

   graph_draw(g, pos=pos, vertex_fill_color=bv, edge_pen_width=be,
              output_size=(400, 400), output="nonfiltered-bt.png")
687

688
.. figure:: nonfiltered-bt.*
Tiago Peixoto's avatar
Tiago Peixoto committed
689
   :align: center
690
691
692
693
694
695
696
697

Everything works in analogous fashion with vertex filtering.

Additionally, the graph can also have its edges reversed with the
:meth:`~graph_tool.Graph.set_reversed` method. This is also an :math:`O(1)`
operation, which does not really modify the graph.

As mentioned previously, the directedness of the graph can also be changed
698
699
"on-the-fly" with the :meth:`~graph_tool.Graph.set_directed` method.

700
701
.. _sec_graph_views:

702
703
704
705
706
707
708
709
710
711
712
Graph views
^^^^^^^^^^^

It is often desired to work with filtered and unfiltered graphs
simultaneously, or to temporarily create a filtered version of graph for
some specific task. For these purposes, graph-tool provides a
:class:`~graph_tool.GraphView` class, which represents a filtered "view"
of a graph, and behaves as an independent graph object, which shares the
underlying data with the original graph. Graph views are constructed by
instantiating a :class:`~graph_tool.GraphView` class, and passing a
graph object which is supposed to be filtered, together with the desired
Tiago Peixoto's avatar
Tiago Peixoto committed
713
714
filter parameters. For example, to create a directed view of the graph
``g`` constructed above, one should do:
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739

.. doctest::

    >>> ug = GraphView(g, directed=True)
    >>> ug.is_directed()
    True

Graph views also provide a much more direct and convenient approach to
vertex/edge filtering: To construct a filtered minimum spanning tree
like in the example above, one must only pass the filter property as the
"efilter" parameter:

.. doctest::

    >>> tv = GraphView(g, efilt=tree)

Note that this is an :math:`O(1)` operation, since it is equivalent (in
speed) to setting the filter in graph ``g`` directly, but in this case
the object ``g`` remains unmodified.

Like above, the result should be the isolated minimum spanning tree:

.. doctest::

    >>> bv, be = betweenness(tv)
740
741
    >>> be.a /= be.a.max() / 5
    >>> graph_draw(tv, pos=pos, vertex_fill_color=bv,
742
    ...            edge_pen_width=be, output="mst-view.pdf")
743
744
    <...>

745
746
747
748
749
750
.. testcode::
   :hide:

   graph_draw(tv, pos=pos, vertex_fill_color=bv,
              edge_pen_width=be, output_size=(400, 400),
              output="mst-view.png")
751

752
.. figure:: mst-view.*
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
   :align: center

   A view of the Delaunay graph, isolating only the minimum spanning tree.

.. note::

   :class:`~graph_tool.GraphView` objects behave *exactly* like regular
   :class:`~graph_tool.Graph` objects. In fact,
   :class:`~graph_tool.GraphView` is a subclass of
   :class:`~graph_tool.Graph`. The only difference is that a
   :class:`~graph_tool.GraphView` object shares its internal data with
   its parent :class:`~graph_tool.Graph` class. Therefore, if the
   original :class:`~graph_tool.Graph` object is modified, this
   modification will be reflected immediately in the
   :class:`~graph_tool.GraphView` object, and vice-versa.

For even more convenience, one can supply a function as filter
parameter, which takes a vertex or an edge as single parameter, and
returns ``True`` if the vertex/edge should be kept and ``False``
otherwise. For instance, if we want to keep only the most "central"
edges, we can do:

.. doctest::

    >>> bv, be = betweenness(g)
778
    >>> u = GraphView(g, efilt=lambda e: be[e] > be.a.max() / 2)
779
780

This creates a graph view ``u`` which contains only the edges of ``g``
781
782
783
784
785
786
which have a normalized betweenness centrality larger than half of the
maximum value. Note that, differently from the case above, this is an
:math:`O(E)` operation, where :math:`E` is the number of edges, since
the supplied function must be called :math:`E` times to construct a
filter property map. Thus, supplying a constructed filter map is always
faster, but supplying a function can be more convenient.
787
788
789
790
791

The graph view constructed above can be visualized as

.. doctest::

792
    >>> be.a /= be.a.max() / 5
793
    >>> graph_draw(u, pos=pos, vertex_fill_color=bv, output="central-edges-view.pdf")
794
795
    <...>

796
797
798
799
800
801
802
.. testcode::
   :hide:

   graph_draw(u, pos=pos, vertex_fill_color=bv, output_size=(400, 400),
              output="central-edges-view.png")


803
.. figure:: central-edges-view.*
804
805
806
807
808
809
810
811
812
   :align: center

   A view of the Delaunay graph, isolating only the edges with
   normalized betweenness centrality larger than 0.01.

Composing graph views
"""""""""""""""""""""

Since graph views are regular graphs, one can just as easily create
813
814
815
816
graph views `of graph views`. This provides a convenient way of
composing filters. For instance, in order to isolate the minimum
spanning tree of all vertices of the example above which have a degree
larger than four, one can do:
817
818
819
820
821
822
823
824
825
826


    >>> u = GraphView(g, vfilt=lambda v: v.out_degree() > 4)
    >>> tree = min_spanning_tree(u)
    >>> u = GraphView(u, efilt=tree)

The resulting graph view can be visualized as

.. doctest::

827
    >>> graph_draw(u, pos=pos, output="composed-filter.pdf")
828
829
    <...>

830
831
832
833
834
.. testcode::
   :hide:

   graph_draw(u, pos=pos, output_size=(400, 400), output="composed-filter.png")

835
.. figure:: composed-filter.*
836
837
838
839
   :align: center

   A composed view, obtained as the minimum spanning tree of all vertices
   in the graph which have a degree larger than four.