__init__.py 39.6 KB
Newer Older
Tiago Peixoto's avatar
Tiago Peixoto committed
1
#! /usr/bin/env python
2
# -*- coding: utf-8 -*-
Tiago Peixoto's avatar
Tiago Peixoto committed
3
#
4 5
# graph_tool -- a general graph manipulation python module
#
Tiago Peixoto's avatar
Tiago Peixoto committed
6
# Copyright (C) 2006-2013 Tiago de Paula Peixoto <tiago@skewed.de>
Tiago Peixoto's avatar
Tiago Peixoto committed
7 8 9 10 11 12 13 14 15 16 17 18 19 20
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.

21
"""
22 23
``graph_tool.centrality`` - Centrality measures
-----------------------------------------------
24 25

This module includes centrality-related algorithms.
26 27 28 29 30 31 32 33 34 35

Summary
+++++++

.. autosummary::
   :nosignatures:

   pagerank
   betweenness
   central_point_dominance
Tiago Peixoto's avatar
Tiago Peixoto committed
36
   closeness
37
   eigenvector
Tiago Peixoto's avatar
Tiago Peixoto committed
38
   katz
39
   hits
40
   eigentrust
41
   trust_transitivity
42 43 44

Contents
++++++++
45 46
"""

47 48
from __future__ import division, absolute_import, print_function

Tiago Peixoto's avatar
Tiago Peixoto committed
49
from .. dl_import import dl_import
50
dl_import("from . import libgraph_tool_centrality")
Tiago Peixoto's avatar
Tiago Peixoto committed
51

52
from .. import _prop, ungroup_vector_property
Tiago Peixoto's avatar
Tiago Peixoto committed
53 54
import sys
import numpy
Tiago Peixoto's avatar
Tiago Peixoto committed
55

Tiago Peixoto's avatar
Tiago Peixoto committed
56 57
__all__ = ["pagerank", "betweenness", "central_point_dominance", "closeness",
           "eigentrust", "eigenvector", "katz", "hits", "trust_transitivity"]
Tiago Peixoto's avatar
Tiago Peixoto committed
58

Tiago Peixoto's avatar
Tiago Peixoto committed
59

60 61
def pagerank(g, damping=0.85, pers=None, weight=None, prop=None, epsilon=1e-6,
             max_iter=None, ret_iter=False):
62 63 64 65 66
    r"""
    Calculate the PageRank of each vertex.

    Parameters
    ----------
67
    g : :class:`~graph_tool.Graph`
68
        Graph to be used.
69
    damping : float, optional (default: 0.85)
70
        Damping factor.
71 72 73 74 75
    pers : :class:`~graph_tool.PropertyMap`, optional (default: None)
        Personalization vector. If omitted, a constant value of :math:`1/N`
        will be used.
    weight : :class:`~graph_tool.PropertyMap`, optional (default: None)
        Edge weights. If omitted, a constant value of 1 will be used.
76
    prop : :class:`~graph_tool.PropertyMap`, optional (default: None)
77 78
        Vertex property map to store the PageRank values. If supplied, it will
        be used uninitialized.
Tiago Peixoto's avatar
Tiago Peixoto committed
79
    epsilon : float, optional (default: 1e-6)
80 81 82 83 84 85 86 87 88
        Convergence condition. The iteration will stop if the total delta of all
        vertices are below this value.
    max_iter : int, optional (default: None)
        If supplied, this will limit the total number of iterations.
    ret_iter : bool, optional (default: False)
        If true, the total number of iterations is also returned.

    Returns
    -------
89 90
    pagerank : :class:`~graph_tool.PropertyMap`
        A vertex property map containing the PageRank values.
91 92 93 94 95

    See Also
    --------
    betweenness: betweenness centrality
    eigentrust: eigentrust centrality
96 97
    eigenvector: eigenvector centrality
    hits: hubs and authority centralities
98
    trust_transitivity: pervasive trust transitivity
99 100 101

    Notes
    -----
Tiago Peixoto's avatar
Tiago Peixoto committed
102 103
    The value of PageRank [pagerank-wikipedia]_ of vertex v, :math:`PR(v)`, is
    given iteratively by the relation:
104 105

    .. math::
106

107 108
        PR(v) = \frac{1-d}{N} + d \sum_{u \in \Gamma^{-}(v)}
                \frac{PR (u)}{d^{+}(u)}
109 110 111 112

    where :math:`\Gamma^{-}(v)` are the in-neighbours of v, :math:`d^{+}(w)` is
    the out-degree of w, and d is a damping factor.

113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130
    If a personalization property :math:`p(v)` is given, the definition becomes:

    .. math::

        PR(v) = (1-d)p(v) + d \sum_{u \in \Gamma^{-}(v)}
                \frac{PR (u)}{d^{+}(u)}

    If edge weights are also given, the equation is then generalized to:

    .. math::

        PR(v) = (1-d)p(v) + d \sum_{u \in \Gamma^{-}(v)}
                \frac{PR (u) w_{u\to v}}{d^{+}(u)}

    where :math:`d^{+}(u)=\sum_{y}A_{u,y}w_{u\to y}` is redefined to be the sum
    of the weights of the out-going edges from u.

    The implemented algorithm progressively iterates the above equations, until
Tiago Peixoto's avatar
Tiago Peixoto committed
131
    it no longer changes, according to the parameter epsilon. It has a
132 133 134 135 136 137
    topology-dependent running time.

    If enabled during compilation, this algorithm runs in parallel.

    Examples
    --------
138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160

    .. doctest:: pagerank

       >>> g = gt.collection.data["polblogs"]
       >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
       >>> pr = gt.pagerank(g)
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=pr,
       ...               vertex_size=gt.prop_to_size(pr, mi=5, ma=15),
       ...               vorder=pr, output="polblogs_pr.pdf")
       <...>

    .. testcode:: pagerank
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=pr,
                     vertex_size=gt.prop_to_size(pr, mi=5, ma=15),
                     vorder=pr, output="polblogs_pr.png")


    .. figure:: polblogs_pr.*
       :align: center

       PageRank values of the a political blogs network of [adamic-polblogs]_.
161 162 163

    Now with a personalization vector, and edge weights:

164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
    .. doctest:: pagerank

       >>> d = g.degree_property_map("total")
       >>> periphery = d.a <= 2
       >>> p = g.new_vertex_property("double")
       >>> p.a[periphery] = 100
       >>> pr = gt.pagerank(g, pers=p)
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=pr,
       ...               vertex_size=gt.prop_to_size(pr, mi=5, ma=15),
       ...               vorder=pr, output="polblogs_pr_pers.pdf")
       <...>

    .. testcode:: pagerank
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=pr,
                     vertex_size=gt.prop_to_size(pr, mi=5, ma=15),
                     vorder=pr, output="polblogs_pr_pers.png")


    .. figure:: polblogs_pr_pers.*
       :align: center

       Personalized PageRank values of the a political blogs network of
       [adamic-polblogs]_, where vertices with very low degree are given
       artificially high scores.
190 191 192

    References
    ----------
193 194
    .. [pagerank-wikipedia] http://en.wikipedia.org/wiki/Pagerank
    .. [lawrence-pagerank-1998] P. Lawrence, B. Sergey, M. Rajeev, W. Terry,
195
       "The pagerank citation ranking: Bringing order to the web", Technical
196
       report, Stanford University, 1998
197 198 199
    .. [Langville-survey-2005] A. N. Langville, C. D. Meyer, "A Survey of
       Eigenvector Methods for Web Information Retrieval", SIAM Review, vol. 47,
       no. 1, pp. 135-161, 2005, :DOI:`10.1137/S0036144503424786`
200 201 202
    .. [adamic-polblogs] L. A. Adamic and N. Glance, "The political blogosphere
       and the 2004 US Election", in Proceedings of the WWW-2005 Workshop on the
       Weblogging Ecosystem (2005). :DOI:`10.1145/1134271.1134277`
203 204 205 206
    """

    if max_iter == None:
        max_iter = 0
Tiago Peixoto's avatar
Tiago Peixoto committed
207 208
    if prop == None:
        prop = g.new_vertex_property("double")
209 210
        N = len(prop.a)
        prop.a = pers.a[:N] if pers is not None else 1. / g.num_vertices()
Tiago Peixoto's avatar
Tiago Peixoto committed
211
    ic = libgraph_tool_centrality.\
212 213 214
            get_pagerank(g._Graph__graph, _prop("v", g, prop),
                         _prop("v", g, pers), _prop("e", g, weight),
                         damping, epsilon, max_iter)
Tiago Peixoto's avatar
Tiago Peixoto committed
215 216 217 218 219
    if ret_iter:
        return prop, ic
    else:
        return prop

Tiago Peixoto's avatar
Tiago Peixoto committed
220

221 222 223 224 225 226
def betweenness(g, vprop=None, eprop=None, weight=None, norm=True):
    r"""
    Calculate the betweenness centrality for each vertex and edge.

    Parameters
    ----------
227
    g : :class:`~graph_tool.Graph`
228
        Graph to be used.
229
    vprop : :class:`~graph_tool.PropertyMap`, optional (default: None)
230
        Vertex property map to store the vertex betweenness values.
231
    eprop : :class:`~graph_tool.PropertyMap`, optional (default: None)
232
        Edge property map to store the edge betweenness values.
233
    weight : :class:`~graph_tool.PropertyMap`, optional (default: None)
234 235 236 237 238 239
        Edge property map corresponding to the weight value of each edge.
    norm : bool, optional (default: True)
        Whether or not the betweenness values should be normalized.

    Returns
    -------
Tiago Peixoto's avatar
Tiago Peixoto committed
240 241
    vertex_betweenness : A vertex property map with the vertex betweenness values.
    edge_betweenness : An edge property map with the edge betweenness values.
242 243 244 245 246 247

    See Also
    --------
    central_point_dominance: central point dominance of the graph
    pagerank: PageRank centrality
    eigentrust: eigentrust centrality
248 249
    eigenvector: eigenvector centrality
    hits: hubs and authority centralities
250
    trust_transitivity: pervasive trust transitivity
251 252 253 254 255

    Notes
    -----
    Betweenness centrality of a vertex :math:`C_B(v)` is defined as,

256 257
    .. math::

258 259 260 261 262 263 264 265 266
        C_B(v)= \sum_{s \neq v \neq t \in V \atop s \neq t}
                \frac{\sigma_{st}(v)}{\sigma_{st}}

    where :math:`\sigma_{st}` is the number of shortest geodesic paths from s to
    t, and :math:`\sigma_{st}(v)` is the number of shortest geodesic paths from
    s to t that pass through a vertex v.  This may be normalised by dividing
    through the number of pairs of vertices not including v, which is
    :math:`(n-1)(n-2)/2`.

267
    The algorithm used here is defined in [brandes-faster-2001]_, and has a
268 269 270 271 272 273 274
    complexity of :math:`O(VE)` for unweighted graphs and :math:`O(VE + V(V+E)
    \log V)` for weighted graphs. The space complexity is :math:`O(VE)`.

    If enabled during compilation, this algorithm runs in parallel.

    Examples
    --------
275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299

    .. doctest:: betweenness

       >>> g = gt.collection.data["polblogs"]
       >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
       >>> vp, ep = gt.betweenness(g)
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=vp,
       ...               vertex_size=gt.prop_to_size(vp, mi=5, ma=15),
       ...               edge_pen_width=gt.prop_to_size(ep, mi=0.5, ma=5),
       ...               vorder=vp, output="polblogs_betweenness.pdf")
       <...>

    .. testcode:: betweenness
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=vp,
                     vertex_size=gt.prop_to_size(vp, mi=5, ma=15),
                     edge_pen_width=gt.prop_to_size(ep, mi=0.5, ma=5),
                     vorder=vp, output="polblogs_betweenness.png")


    .. figure:: polblogs_betweenness.*
       :align: center

       Betweenness values of the a political blogs network of [adamic-polblogs]_.
300 301 302

    References
    ----------
303 304
    .. [betweenness-wikipedia] http://en.wikipedia.org/wiki/Centrality#Betweenness_centrality
    .. [brandes-faster-2001] U. Brandes, "A faster algorithm for betweenness
Tiago Peixoto's avatar
Tiago Peixoto committed
305
       centrality", Journal of Mathematical Sociology, 2001, :doi:`10.1080/0022250X.2001.9990249`
306 307 308
    .. [adamic-polblogs] L. A. Adamic and N. Glance, "The political blogosphere
       and the 2004 US Election", in Proceedings of the WWW-2005 Workshop on the
       Weblogging Ecosystem (2005). :DOI:`10.1145/1134271.1134277`
309
    """
Tiago Peixoto's avatar
Tiago Peixoto committed
310 311 312 313 314 315 316 317 318 319 320 321 322
    if vprop == None:
        vprop = g.new_vertex_property("double")
    if eprop == None:
        eprop = g.new_edge_property("double")
    if weight != None and weight.value_type() != eprop.value_type():
        nw = g.new_edge_property(eprop.value_type())
        g.copy_property(weight, nw)
        weight = nw
    libgraph_tool_centrality.\
            get_betweenness(g._Graph__graph, _prop("e", g, weight),
                            _prop("e", g, eprop), _prop("v", g, vprop), norm)
    return vprop, eprop

Tiago Peixoto's avatar
Tiago Peixoto committed
323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448
def closeness(g, weight=None, source=None, vprop=None, norm=True, harmonic=False):
    r"""
    Calculate the closeness centrality for each vertex.

    Parameters
    ----------
    g : :class:`~graph_tool.Graph`
        Graph to be used.
    weight : :class:`~graph_tool.PropertyMap`, optional (default: None)
        Edge property map corresponding to the weight value of each edge.
    source : :class:`~graph_tool.Vertex`, optional (default: ``None``)
        If specified, the centrality is computed for this vertex alone.
    vprop : :class:`~graph_tool.PropertyMap`, optional (default: ``None``)
        Vertex property map to store the vertex centrality values.
    norm : bool, optional (default: ``True``)
        Whether or not the centrality values should be normalized.
    harmonic : bool, optional (default: ``False``)
        If true, the sum of the inverse of the distances will be computed,
        instead of the inverse of the sum.

    Returns
    -------
    vertex_closeness : :class:`~graph_tool.PropertyMap`
        A vertex property map with the vertex closeness values.

    See Also
    --------
    central_point_dominance: central point dominance of the graph
    pagerank: PageRank centrality
    eigentrust: eigentrust centrality
    eigenvector: eigenvector centrality
    hits: hubs and authority centralities
    trust_transitivity: pervasive trust transitivity

    Notes
    -----
    The closeness centrality of a vertex :math:`i` is defined as,

    .. math::

        c_i = \frac{1}{\sum_j d_{ij}}

    where :math:`d_{ij}` is the (possibly directed and/or weighted) distance
    from :math:`i` to :math:`j`. In case there is no path between the two
    vertices, here the distance is taken to be zero.

    If ``harmonic == True``, the definition becomes

    .. math::

        c_i = \sum_j\frac{1}{d_{ij}},

    but now, in case there is no path between the two vertices, we take
    :math:`d_{ij} \to\infty` such that :math:`1/d_{ij}=0`.

    If ``norm == True``, the values of :math:`c_i` are normalized by
    :math:`n_i-1` where :math:`n_i` is the size of the (out-) component of
    :math:`i`. If ``harmonic == True``, they are instead simply normalized by
    :math:`N-1`.

    The algorithm complexity of :math:`O(N(N + E))` for unweighted graphs and
    :math:`O(N(N+E) \log N)` for weighted graphs. If the option ``source`` is
    specified, this drops to :math:`O(N + E)` and :math:`O((N+E)\log N)`
    respectively.

    If enabled during compilation, this algorithm runs in parallel.

    Examples
    --------

    .. doctest:: closeness

       >>> g = gt.collection.data["polblogs"]
       >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
       >>> c = gt.closeness(g)
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=c,
       ...               vertex_size=gt.prop_to_size(c, mi=5, ma=15),
       ...               vorder=c, output="polblogs_closeness.pdf")
       <...>

    .. testcode:: closeness
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=c,
                     vertex_size=gt.prop_to_size(c, mi=5, ma=15),
                     vorder=c, output="polblogs_closeness.png")


    .. figure:: polblogs_closeness.*
       :align: center

       Closeness values of the a political blogs network of [adamic-polblogs]_.

    References
    ----------
    .. [closeness-wikipedia] https://en.wikipedia.org/wiki/Closeness_centrality
    .. [opsahl-node-2010] Opsahl, T., Agneessens, F., Skvoretz, J., "Node
       centrality in weighted networks: Generalizing degree and shortest
       paths". Social Networks 32, 245-251, 2010 :DOI:`10.1016/j.socnet.2010.03.006`
    .. [adamic-polblogs] L. A. Adamic and N. Glance, "The political blogosphere
       and the 2004 US Election", in Proceedings of the WWW-2005 Workshop on the
       Weblogging Ecosystem (2005). :DOI:`10.1145/1134271.1134277`

    """
    if source is None:
        if vprop == None:
            vprop = g.new_vertex_property("double")
        libgraph_tool_centrality.\
            closeness(g._Graph__graph, _prop("e", g, weight),
                      _prop("v", g, vprop), harmonic, norm)
        return vprop
    else:
        max_dist = g.num_vertices() + 1
        dist = shortest_distance(g, source=source, weight=weight,
                                 max_dist=max_dist)
        if harmonic:
            dists = dist.fa[(dist.fa < max_dist) * (dist.fa > 0)]
            c = (1. / dists).sum()
            if norm:
                c /= g.num_vertices() - 1
        else:
            dists = dist.fa[(dist.fa < max_dist) * (dist.fa > 0)]
            c = 1. / dists.sum()
            if norm:
                c /= len(dists)

Tiago Peixoto's avatar
Tiago Peixoto committed
449

Tiago Peixoto's avatar
Tiago Peixoto committed
450
def central_point_dominance(g, betweenness):
451 452 453 454 455 456
    r"""
    Calculate the central point dominance of the graph, given the betweenness
    centrality of each vertex.

    Parameters
    ----------
457
    g : :class:`~graph_tool.Graph`
458
        Graph to be used.
459
    betweenness : :class:`~graph_tool.PropertyMap`
460 461 462 463 464
        Vertex property map with the betweenness centrality values. The values
        must be normalized.

    Returns
    -------
465 466
    cp : float
        The central point dominance.
467 468 469 470 471 472 473 474

    See Also
    --------
    betweenness: betweenness centrality

    Notes
    -----
    Let :math:`v^*` be the vertex with the largest relative betweenness
475
    centrality; then, the central point dominance [freeman-set-1977]_ is defined
476 477
    as:

478 479
    .. math::

480 481 482 483 484 485 486 487 488
        C'_B = \frac{1}{|V|-1} \sum_{v} C_B(v^*) - C_B(v)

    where :math:`C_B(v)` is the normalized betweenness centrality of vertex
    v. The value of :math:`C_B` lies in the range [0,1].

    The algorithm has a complexity of :math:`O(V)`.

    Examples
    --------
489 490 491 492 493 494

    >>> g = gt.collection.data["polblogs"]
    >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
    >>> vp, ep = gt.betweenness(g)
    >>> print(gt.central_point_dominance(g, vp))
    0.11610685614353008
495 496 497

    References
    ----------
498
    .. [freeman-set-1977] Linton C. Freeman, "A Set of Measures of Centrality
Tiago Peixoto's avatar
Tiago Peixoto committed
499 500
       Based on Betweenness", Sociometry, Vol. 40, No. 1,  pp. 35-41, 1977,
       `http://www.jstor.org/stable/3033543 <http://www.jstor.org/stable/3033543>`_
501 502
    """

Tiago Peixoto's avatar
Tiago Peixoto committed
503
    return libgraph_tool_centrality.\
504
           get_central_point_dominance(g._Graph__graph,
Tiago Peixoto's avatar
Tiago Peixoto committed
505 506
                                       _prop("v", g, betweenness))

507

508 509 510 511 512 513 514 515 516
def eigenvector(g, weight=None, vprop=None, epsilon=1e-6, max_iter=None):
    r"""
    Calculate the eigenvector centrality of each vertex in the graph, as well as
    the largest eigenvalue.

    Parameters
    ----------
    g : :class:`~graph_tool.Graph`
        Graph to be used.
517
    weight : :class:`~graph_tool.PropertyMap` (optional, default: ``None``)
518 519
        Edge property map with the edge weights.
    vprop : :class:`~graph_tool.PropertyMap`, optional (default: ``None``)
520 521
        Vertex property map where the values of eigenvector must be stored. If
        provided, it will be used uninitialized.
522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
    epsilon : float, optional (default: ``1e-6``)
        Convergence condition. The iteration will stop if the total delta of all
        vertices are below this value.
    max_iter : int, optional (default: ``None``)
        If supplied, this will limit the total number of iterations.

    Returns
    -------
    eigenvalue : float
        The largest eigenvalue of the (weighted) adjacency matrix.
    eigenvector : :class:`~graph_tool.PropertyMap`
        A vertex property map containing the eigenvector values.

    See Also
    --------
    betweenness: betweenness centrality
    pagerank: PageRank centrality
539
    hits: hubs and authority centralities
540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566
    trust_transitivity: pervasive trust transitivity

    Notes
    -----

    The eigenvector centrality :math:`\mathbf{x}` is the eigenvector of the
    (weighted) adjacency matrix with the largest eigenvalue :math:`\lambda`,
    i.e. it is the solution of

    .. math::

        \mathbf{A}\mathbf{x} = \lambda\mathbf{x},


    where :math:`\mathbf{A}` is the (weighted) adjacency matrix and
    :math:`\lambda` is the largest eigenvalue.

    The algorithm uses the power method which has a topology-dependent complexity of
    :math:`O\left(N\times\frac{-\log\epsilon}{\log|\lambda_1/\lambda_2|}\right)`,
    where :math:`N` is the number of vertices, :math:`\epsilon` is the ``epsilon``
    parameter, and :math:`\lambda_1` and :math:`\lambda_2` are the largest and
    second largest eigenvalues of the (weighted) adjacency matrix, respectively.

    If enabled during compilation, this algorithm runs in parallel.

    Examples
    --------
567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597
    .. testsetup:: eigenvector

       np.random.seed(42)

    .. doctest:: eigenvector

       >>> g = gt.collection.data["polblogs"]
       >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
       >>> w = g.new_edge_property("double")
       >>> w.a = np.random.random(len(w.a)) * 42
       >>> ee, x = gt.eigenvector(g, w)
       >>> print(ee)
       0.0013713102236792602
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=x,
       ...               vertex_size=gt.prop_to_size(x, mi=5, ma=15),
       ...               vorder=x, output="polblogs_eigenvector.pdf")
       <...>

    .. testcode:: eigenvector
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=x,
                     vertex_size=gt.prop_to_size(x, mi=5, ma=15),
                     vorder=x, output="polblogs_eigenvector.png")


    .. figure:: polblogs_eigenvector.*
       :align: center

       Eigenvector values of the a political blogs network of
       [adamic-polblogs]_, with random weights attributed to the edges.
598 599 600 601 602 603 604 605 606

    References
    ----------

    .. [eigenvector-centrality] http://en.wikipedia.org/wiki/Centrality#Eigenvector_centrality
    .. [power-method] http://en.wikipedia.org/wiki/Power_iteration
    .. [langville-survey-2005] A. N. Langville, C. D. Meyer, "A Survey of
       Eigenvector Methods for Web Information Retrieval", SIAM Review, vol. 47,
       no. 1, pp. 135-161, 2005, :DOI:`10.1137/S0036144503424786`
607 608 609
    .. [adamic-polblogs] L. A. Adamic and N. Glance, "The political blogosphere
       and the 2004 US Election", in Proceedings of the WWW-2005 Workshop on the
       Weblogging Ecosystem (2005). :DOI:`10.1145/1134271.1134277`
610 611 612

    """

613
    if vprop is None:
614
        vprop = g.new_vertex_property("double")
615
        vprop.a = 1. / g.num_vertices()
616 617 618 619 620 621 622 623
    if max_iter is None:
        max_iter = 0
    ee = libgraph_tool_centrality.\
         get_eigenvector(g._Graph__graph, _prop("e", g, weight),
                         _prop("v", g, vprop), epsilon, max_iter)
    return ee, vprop


Tiago Peixoto's avatar
Tiago Peixoto committed
624 625
def katz(g, alpha=0.01, beta=None, weight=None, vprop=None, epsilon=1e-6, max_iter=None):
    r"""
Tiago Peixoto's avatar
Tiago Peixoto committed
626
    Calculate the Katz centrality of each vertex in the graph.
Tiago Peixoto's avatar
Tiago Peixoto committed
627 628 629 630 631 632 633 634

    Parameters
    ----------
    g : :class:`~graph_tool.Graph`
        Graph to be used.
    weight : :class:`~graph_tool.PropertyMap` (optional, default: ``None``)
        Edge property map with the edge weights.
    alpha : float, optional (default: ``0.01``)
635 636
        Free parameter :math:`\alpha`. This must be smaller than the inverse of
        the largest eigenvalue of the adjacency matrix.
Tiago Peixoto's avatar
Tiago Peixoto committed
637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683
    beta : :class:`~graph_tool.PropertyMap`, optional (default: ``None``)
        Vertex property map where the local personalization values. If not
        provided, the global value of 1 will be used.
    vprop : :class:`~graph_tool.PropertyMap`, optional (default: ``None``)
        Vertex property map where the values of eigenvector must be stored. If
        provided, it will be used uninitialized.
    epsilon : float, optional (default: ``1e-6``)
        Convergence condition. The iteration will stop if the total delta of all
        vertices are below this value.
    max_iter : int, optional (default: ``None``)
        If supplied, this will limit the total number of iterations.

    Returns
    -------
    centrality : :class:`~graph_tool.PropertyMap`
        A vertex property map containing the Katz centrality values.

    See Also
    --------
    betweenness: betweenness centrality
    pagerank: PageRank centrality
    eigenvector: eigenvector centrality
    hits: hubs and authority centralities
    trust_transitivity: pervasive trust transitivity

    Notes
    -----

    The Katz centrality :math:`\mathbf{x}` is the solution of the nonhomogeneous
    linear system

    .. math::

        \mathbf{x} = \alpha\mathbf{A}\mathbf{x} + \mathbf{\beta},


    where :math:`\mathbf{A}` is the (weighted) adjacency matrix and
    :math:`\mathbf{\beta}` is the personalization vector (if not supplied,
    :math:`\mathbf{\beta} = \mathbf{1}` is assumed).

    The algorithm uses successive iterations of the equation above, which has a
    topology-dependent convergence complexity.

    If enabled during compilation, this algorithm runs in parallel.

    Examples
    --------
684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712
    .. testsetup:: katz

       np.random.seed(42)

    .. doctest:: katz

       >>> g = gt.collection.data["polblogs"]
       >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
       >>> w = g.new_edge_property("double")
       >>> w.a = np.random.random(len(w.a)) * 42
       >>> x = gt.katz(g, weight=w)
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=x,
       ...               vertex_size=gt.prop_to_size(x, mi=5, ma=15),
       ...               vorder=x, output="polblogs_katz.pdf")
       <...>

    .. testcode:: katz
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=x,
                     vertex_size=gt.prop_to_size(x, mi=5, ma=15),
                     vorder=x, output="polblogs_katz.png")


    .. figure:: polblogs_katz.*
       :align: center

       Katz centrality values of the a political blogs network of
       [adamic-polblogs]_, with random weights attributed to the edges.
Tiago Peixoto's avatar
Tiago Peixoto committed
713 714 715 716 717 718 719

    References
    ----------

    .. [katz-centrality] http://en.wikipedia.org/wiki/Katz_centrality
    .. [katz-new] L. Katz, "A new status index derived from sociometric analysis",
       Psychometrika 18, Number 1, 39-43, 1953, :DOI:`10.1007/BF02289026`
720 721 722
    .. [adamic-polblogs] L. A. Adamic and N. Glance, "The political blogosphere
       and the 2004 US Election", in Proceedings of the WWW-2005 Workshop on the
       Weblogging Ecosystem (2005). :DOI:`10.1145/1134271.1134277`
Tiago Peixoto's avatar
Tiago Peixoto committed
723 724 725 726 727 728 729 730 731 732 733 734 735 736
    """

    if vprop == None:
        vprop = g.new_vertex_property("double")
        N = len(vprop.a)
        vprop.a = beta.a[:N] if beta is not None else 1.
    if max_iter is None:
        max_iter = 0
    ee = libgraph_tool_centrality.\
         get_katz(g._Graph__graph, _prop("e", g, weight), _prop("v", g, vprop),
         _prop("v", beta, vprop), float(alpha), epsilon, max_iter)
    return vprop


737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802
def hits(g, weight=None, xprop=None, yprop=None, epsilon=1e-6, max_iter=None):
    r"""
    Calculate the authority and hub centralities of each vertex in the graph.

    Parameters
    ----------
    g : :class:`~graph_tool.Graph`
        Graph to be used.
    weight : :class:`~graph_tool.PropertyMap` (optional, default: ``None``)
        Edge property map with the edge weights.
    xprop : :class:`~graph_tool.PropertyMap`, optional (default: ``None``)
        Vertex property map where the authority centrality must be stored.
    yprop : :class:`~graph_tool.PropertyMap`, optional (default: ``None``)
        Vertex property map where the hub centrality must be stored.
    epsilon : float, optional (default: ``1e-6``)
        Convergence condition. The iteration will stop if the total delta of all
        vertices are below this value.
    max_iter : int, optional (default: ``None``)
        If supplied, this will limit the total number of iterations.

    Returns
    -------
    eig : `float`
        The largest eigenvalue of the cocitation matrix.
    x : :class:`~graph_tool.PropertyMap`
        A vertex property map containing the authority centrality values.
    y : :class:`~graph_tool.PropertyMap`
        A vertex property map containing the hub centrality values.

    See Also
    --------
    betweenness: betweenness centrality
    eigenvector: eigenvector centrality
    pagerank: PageRank centrality
    trust_transitivity: pervasive trust transitivity

    Notes
    -----

    The Hyperlink-Induced Topic Search (HITS) centrality assigns hub
    (:math:`\mathbf{y}`) and authority (:math:`\mathbf{x}`) centralities to the
    vertices, following:

    .. math::

        \begin{align}
            \mathbf{x} &= \alpha\mathbf{A}\mathbf{y} \\
            \mathbf{y} &= \beta\mathbf{A}^T\mathbf{x}
        \end{align}


    where :math:`\mathbf{A}` is the (weighted) adjacency matrix and
    :math:`\lambda = 1/(\alpha\beta)` is the largest eigenvalue of the
    cocitation matrix, :math:`\mathbf{A}\mathbf{A}^T`. (Without loss of
    generality, we set :math:`\beta=1` in the algorithm.)

    The algorithm uses the power method which has a topology-dependent complexity of
    :math:`O\left(N\times\frac{-\log\epsilon}{\log|\lambda_1/\lambda_2|}\right)`,
    where :math:`N` is the number of vertices, :math:`\epsilon` is the ``epsilon``
    parameter, and :math:`\lambda_1` and :math:`\lambda_2` are the largest and
    second largest eigenvalues of the (weighted) cocitation matrix, respectively.

    If enabled during compilation, this algorithm runs in parallel.

    Examples
    --------
803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838

    .. doctest:: hits

       >>> g = gt.collection.data["polblogs"]
       >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
       >>> ee, x, y = gt.hits(g)
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=x,
       ...               vertex_size=gt.prop_to_size(x, mi=5, ma=15),
       ...               vorder=x, output="polblogs_hits_auths.pdf")
       <...>
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=y,
       ...               vertex_size=gt.prop_to_size(y, mi=5, ma=15),
       ...               vorder=y, output="polblogs_hits_hubs.pdf")
       <...>

    .. testcode:: hits
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=x,
                     vertex_size=gt.prop_to_size(x, mi=5, ma=15),
                     vorder=x, output="polblogs_hits_auths.png")
       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=y,
                     vertex_size=gt.prop_to_size(y, mi=5, ma=15),
                     vorder=y, output="polblogs_hits_hubs.png")


    .. figure:: polblogs_hits_auths.*
       :align: left

       HITS authority values of the a political blogs network of
       [adamic-polblogs]_.

    .. figure:: polblogs_hits_hubs.*
       :align: right

       HITS hub values of the a political blogs network of [adamic-polblogs]_.
839 840 841 842 843 844

    References
    ----------

    .. [hits-algorithm] http://en.wikipedia.org/wiki/HITS_algorithm
    .. [kleinberg-authoritative] J. Kleinberg, "Authoritative sources in a
845
       hyperlinked environment", Journal of the ACM 46 (5): 604-632, 1999,
846 847
       :DOI:`10.1145/324133.324140`.
    .. [power-method] http://en.wikipedia.org/wiki/Power_iteration
848 849 850
    .. [adamic-polblogs] L. A. Adamic and N. Glance, "The political blogosphere
       and the 2004 US Election", in Proceedings of the WWW-2005 Workshop on the
       Weblogging Ecosystem (2005). :DOI:`10.1145/1134271.1134277`
851 852 853 854 855 856 857 858 859 860 861 862 863 864
    """

    if xprop is None:
        xprop = g.new_vertex_property("double")
    if yprop is None:
        yprop = g.new_vertex_property("double")
    if max_iter is None:
        max_iter = 0
    l = libgraph_tool_centrality.\
         get_hits(g._Graph__graph, _prop("e", g, weight), _prop("v", g, xprop),
                  _prop("v", g, yprop), epsilon, max_iter)
    return 1. / l, xprop, yprop


Tiago Peixoto's avatar
Tiago Peixoto committed
865
def eigentrust(g, trust_map, vprop=None, norm=False, epsilon=1e-6, max_iter=0,
Tiago Peixoto's avatar
Tiago Peixoto committed
866
               ret_iter=False):
867 868 869 870 871
    r"""
    Calculate the eigentrust centrality of each vertex in the graph.

    Parameters
    ----------
872
    g : :class:`~graph_tool.Graph`
873
        Graph to be used.
874
    trust_map : :class:`~graph_tool.PropertyMap`
875
        Edge property map with the values of trust associated with each
876
        edge. The values must lie in the range [0,1].
877
    vprop : :class:`~graph_tool.PropertyMap`, optional (default: ``None``)
878
        Vertex property map where the values of eigentrust must be stored.
879
    norm : bool, optional (default:  ``False``)
880
        Norm eigentrust values so that the total sum equals 1.
881
    epsilon : float, optional (default: ``1e-6``)
882 883
        Convergence condition. The iteration will stop if the total delta of all
        vertices are below this value.
884
    max_iter : int, optional (default: ``None``)
885
        If supplied, this will limit the total number of iterations.
886
    ret_iter : bool, optional (default: ``False``)
887 888 889 890
        If true, the total number of iterations is also returned.

    Returns
    -------
891 892
    eigentrust : :class:`~graph_tool.PropertyMap`
        A vertex property map containing the eigentrust values.
893 894 895 896 897

    See Also
    --------
    betweenness: betweenness centrality
    pagerank: PageRank centrality
898
    trust_transitivity: pervasive trust transitivity
899 900 901

    Notes
    -----
902
    The eigentrust [kamvar-eigentrust-2003]_ values :math:`t_i` correspond the
903 904
    following limit

905 906
    .. math::

907 908 909 910 911
        \mathbf{t} = \lim_{n\to\infty} \left(C^T\right)^n \mathbf{c}

    where :math:`c_i = 1/|V|` and the elements of the matrix :math:`C` are the
    normalized trust values:

912 913
    .. math::

914 915 916 917 918 919 920 921
        c_{ij} = \frac{\max(s_{ij},0)}{\sum_{j} \max(s_{ij}, 0)}

    The algorithm has a topology-dependent complexity.

    If enabled during compilation, this algorithm runs in parallel.

    Examples
    --------
922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952

    .. testsetup:: eigentrust

       np.random.seed(42)

    .. doctest:: eigentrust

       >>> g = gt.collection.data["polblogs"]
       >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
       >>> w = g.new_edge_property("double")
       >>> w.a = np.random.random(len(w.a)) * 42
       >>> t = gt.eigentrust(g, w)
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=t,
       ...               vertex_size=gt.prop_to_size(t, mi=5, ma=15),
       ...               vorder=t, output="polblogs_eigentrust.pdf")
       <...>

    .. testcode:: eigentrust
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=t,
                     vertex_size=gt.prop_to_size(t, mi=5, ma=15),
                     vorder=t, output="polblogs_eigentrust.png")


    .. figure:: polblogs_eigentrust.*
       :align: center

       Eigentrust values of the a political blogs network of
       [adamic-polblogs]_, with random weights attributed to the edges.

953 954 955

    References
    ----------
956
    .. [kamvar-eigentrust-2003] S. D. Kamvar, M. T. Schlosser, H. Garcia-Molina
957 958
       "The eigentrust algorithm for reputation management in p2p networks",
       Proceedings of the 12th international conference on World Wide Web,
Tiago Peixoto's avatar
Tiago Peixoto committed
959
       Pages: 640 - 651, 2003, :doi:`10.1145/775152.775242`
960 961 962
    .. [adamic-polblogs] L. A. Adamic and N. Glance, "The political blogosphere
       and the 2004 US Election", in Proceedings of the WWW-2005 Workshop on the
       Weblogging Ecosystem (2005). :DOI:`10.1145/1134271.1134277`
963 964
    """

Tiago Peixoto's avatar
Tiago Peixoto committed
965 966
    if vprop == None:
        vprop = g.new_vertex_property("double")
967 968
    i = libgraph_tool_centrality.\
           get_eigentrust(g._Graph__graph, _prop("e", g, trust_map),
Tiago Peixoto's avatar
Tiago Peixoto committed
969
                          _prop("v", g, vprop), epsilon, max_iter)
970 971 972 973 974 975 976 977
    if norm:
        vprop.get_array()[:] /= sum(vprop.get_array())

    if ret_iter:
        return vprop, i
    else:
        return vprop

Tiago Peixoto's avatar
Tiago Peixoto committed
978

979
def trust_transitivity(g, trust_map, source=None, target=None, vprop=None):
980
    r"""
981 982
    Calculate the pervasive trust transitivity between chosen (or all) vertices
    in the graph.
983 984 985

    Parameters
    ----------
986
    g : :class:`~graph_tool.Graph`
987
        Graph to be used.
988
    trust_map : :class:`~graph_tool.PropertyMap`
989 990
        Edge property map with the values of trust associated with each
        edge. The values must lie in the range [0,1].
Tiago Peixoto's avatar
Tiago Peixoto committed
991
    source : :class:`~graph_tool.Vertex` (optional, default: None)
992
        Source vertex. All trust values are computed relative to this vertex.
993
        If left unspecified, the trust values for all sources are computed.
Tiago Peixoto's avatar
Tiago Peixoto committed
994
    target : :class:`~graph_tool.Vertex` (optional, default: None)
995 996 997
        The only target for which the trust value will be calculated. If left
        unspecified, the trust values for all targets are computed.
    vprop : :class:`~graph_tool.PropertyMap` (optional, default: None)
998 999
        A vertex property map where the values of transitive trust must be
        stored.
1000 1001 1002

    Returns
    -------
1003 1004 1005 1006 1007 1008 1009 1010
    trust_transitivity : :class:`~graph_tool.PropertyMap` or float
        A vertex vector property map containing, for each source vertex, a
        vector with the trust values for the other vertices. If only one of
        `source` or `target` is specified, this will be a single-valued vertex
        property map containing the trust vector from/to the source/target
        vertex to/from the rest of the network. If both `source` and `target`
        are specified, the result is a single float, with the corresponding
        trust value for the target.
1011

1012 1013 1014 1015 1016 1017 1018 1019
    See Also
    --------
    eigentrust: eigentrust centrality
    betweenness: betweenness centrality
    pagerank: PageRank centrality

    Notes
    -----
Tiago Peixoto's avatar
Tiago Peixoto committed
1020
    The pervasive trust transitivity between vertices i and j is defined as
1021

1022 1023
    .. math::

1024 1025
        t_{ij} = \frac{\sum_m A_{m,j} w^2_{G\setminus\{j\}}(i\to m)c_{m,j}}
                 {\sum_m A_{m,j} w_{G\setminus\{j\}}(i\to m)}
1026

1027 1028 1029
    where :math:`A_{ij}` is the adjacency matrix, :math:`c_{ij}` is the direct
    trust from i to j, and :math:`w_G(i\to j)` is the weight of the path with
    maximum weight from i to j, computed as
Tiago Peixoto's avatar
Tiago Peixoto committed
1030

1031 1032
    .. math::

1033
       w_G(i\to j) = \prod_{e\in i\to j} c_e.
1034

1035 1036
    The algorithm measures the transitive trust by finding the paths with
    maximum weight, using Dijkstra's algorithm, to all in-neighbours of a given
1037
    target. This search needs to be performed repeatedly for every target, since
1038 1039 1040 1041 1042 1043 1044
    it needs to be removed from the graph first. For each given source, the
    resulting complexity is therefore :math:`O(N^2\log N)` for all targets, and
    :math:`O(N\log N)` for a single target. For a given target, the complexity
    for obtaining the trust from all given sources is :math:`O(kN\log N)`, where
    :math:`k` is the in-degree of the target. Thus, the complexity for obtaining
    the complete trust matrix is :math:`O(EN\log N)`, where :math:`E` is the
    number of edges in the network.
1045 1046 1047 1048 1049

    If enabled during compilation, this algorithm runs in parallel.

    Examples
    --------
1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082
    .. testsetup:: trust_transitivity

       np.random.seed(42)

    .. doctest:: trust_transitivity

       >>> g = gt.collection.data["polblogs"]
       >>> g = gt.GraphView(g, vfilt=gt.label_largest_component(g))
       >>> g = gt.Graph(g, prune=True)
       >>> w = g.new_edge_property("double")
       >>> w.a = np.random.random(len(w.a))
       >>> g.vp["label"][g.vertex(42)]
       'blogforamerica.com'
       >>> t = gt.trust_transitivity(g, w, source=g.vertex(42))
       >>> gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=t,
       ...               vertex_size=gt.prop_to_size(t, mi=5, ma=15),
       ...               vorder=t, output="polblogs_trust_transitivity.pdf")
       <...>

    .. testcode:: trust_transitivity
       :hide:

       gt.graph_draw(g, pos=g.vp["pos"], vertex_fill_color=t,
                     vertex_size=gt.prop_to_size(t, mi=5, ma=15),
                     vorder=t, output="polblogs_trust_transitivity.png")


    .. figure:: polblogs_trust_transitivity.*
       :align: center

       Trust transitivity values from source vertex 42 of the a political blogs
       network of [adamic-polblogs]_, with random weights attributed to the
       edges.
Tiago Peixoto's avatar
Tiago Peixoto committed
1083 1084 1085

    References
    ----------
1086 1087 1088
    .. [richters-trust-2010] Oliver Richters and Tiago P. Peixoto, "Trust
       Transitivity in Social Networks," PLoS ONE 6, no. 4:
       e1838 (2011), :doi:`10.1371/journal.pone.0018384`
1089 1090 1091
    .. [adamic-polblogs] L. A. Adamic and N. Glance, "The political blogosphere
       and the 2004 US Election", in Proceedings of the WWW-2005 Workshop on the
       Weblogging Ecosystem (2005). :DOI:`10.1145/1134271.1134277`
Tiago Peixoto's avatar
Tiago Peixoto committed
1092

1093
    """
Tiago Peixoto's avatar
Tiago Peixoto committed
1094 1095

    if vprop == None:
1096
        vprop = g.new_vertex_property("vector<double>")
1097

1098 1099 1100 1101
    if target == None:
        target = -1
    else:
        target = g.vertex_index[target]
1102

1103 1104 1105 1106 1107
    if source == None:
        source = -1
    else:
        source = g.vertex_index[source]

1108
    libgraph_tool_centrality.\
1109 1110 1111 1112
            get_trust_transitivity(g._Graph__graph, source, target,
                                   _prop("e", g, trust_map),
                                   _prop("v", g, vprop))
    if target != -1 or source != -1:
1113
        vprop = ungroup_vector_property(vprop, [0])[0]
1114
    if target != -1 and source != -1:
1115
        return vprop.a[target]
1116
    return vprop