1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
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
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
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
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
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
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
|
"""PageRank analysis of graph structure. """
from warnings import warn
import networkx as nx
__all__ = ["pagerank", "google_matrix"]
@nx.dispatch
def pagerank(
G,
alpha=0.85,
personalization=None,
max_iter=100,
tol=1.0e-6,
nstart=None,
weight="weight",
dangling=None,
):
"""Returns the PageRank of the nodes in the graph.
PageRank computes a ranking of the nodes in the graph G based on
the structure of the incoming links. It was originally designed as
an algorithm to rank web pages.
Parameters
----------
G : graph
A NetworkX graph. Undirected graphs will be converted to a directed
graph with two directed edges for each undirected edge.
alpha : float, optional
Damping parameter for PageRank, default=0.85.
personalization: dict, optional
The "personalization vector" consisting of a dictionary with a
key some subset of graph nodes and personalization value each of those.
At least one personalization value must be non-zero.
If not specfiied, a nodes personalization value will be zero.
By default, a uniform distribution is used.
max_iter : integer, optional
Maximum number of iterations in power method eigenvalue solver.
tol : float, optional
Error tolerance used to check convergence in power method solver.
nstart : dictionary, optional
Starting value of PageRank iteration for each node.
weight : key, optional
Edge data key to use as weight. If None weights are set to 1.
dangling: dict, optional
The outedges to be assigned to any "dangling" nodes, i.e., nodes without
any outedges. The dict key is the node the outedge points to and the dict
value is the weight of that outedge. By default, dangling nodes are given
outedges according to the personalization vector (uniform if not
specified). This must be selected to result in an irreducible transition
matrix (see notes under google_matrix). It may be common to have the
dangling dict to be the same as the personalization dict.
Returns
-------
pagerank : dictionary
Dictionary of nodes with PageRank as value
Examples
--------
>>> G = nx.DiGraph(nx.path_graph(4))
>>> pr = nx.pagerank(G, alpha=0.9)
Notes
-----
The eigenvector calculation is done by the power iteration method
and has no guarantee of convergence. The iteration will stop after
an error tolerance of ``len(G) * tol`` has been reached. If the
number of iterations exceed `max_iter`, a
:exc:`networkx.exception.PowerIterationFailedConvergence` exception
is raised.
The PageRank algorithm was designed for directed graphs but this
algorithm does not check if the input graph is directed and will
execute on undirected graphs by converting each edge in the
directed graph to two edges.
See Also
--------
google_matrix
Raises
------
PowerIterationFailedConvergence
If the algorithm fails to converge to the specified tolerance
within the specified number of iterations of the power iteration
method.
References
----------
.. [1] A. Langville and C. Meyer,
"A survey of eigenvector methods of web information retrieval."
http://citeseer.ist.psu.edu/713792.html
.. [2] Page, Lawrence; Brin, Sergey; Motwani, Rajeev and Winograd, Terry,
The PageRank citation ranking: Bringing order to the Web. 1999
http://dbpubs.stanford.edu:8090/pub/showDoc.Fulltext?lang=en&doc=1999-66&format=pdf
"""
return _pagerank_scipy(
G, alpha, personalization, max_iter, tol, nstart, weight, dangling
)
def _pagerank_python(
G,
alpha=0.85,
personalization=None,
max_iter=100,
tol=1.0e-6,
nstart=None,
weight="weight",
dangling=None,
):
if len(G) == 0:
return {}
D = G.to_directed()
# Create a copy in (right) stochastic form
W = nx.stochastic_graph(D, weight=weight)
N = W.number_of_nodes()
# Choose fixed starting vector if not given
if nstart is None:
x = dict.fromkeys(W, 1.0 / N)
else:
# Normalized nstart vector
s = sum(nstart.values())
x = {k: v / s for k, v in nstart.items()}
if personalization is None:
# Assign uniform personalization vector if not given
p = dict.fromkeys(W, 1.0 / N)
else:
s = sum(personalization.values())
p = {k: v / s for k, v in personalization.items()}
if dangling is None:
# Use personalization vector if dangling vector not specified
dangling_weights = p
else:
s = sum(dangling.values())
dangling_weights = {k: v / s for k, v in dangling.items()}
dangling_nodes = [n for n in W if W.out_degree(n, weight=weight) == 0.0]
# power iteration: make up to max_iter iterations
for _ in range(max_iter):
xlast = x
x = dict.fromkeys(xlast.keys(), 0)
danglesum = alpha * sum(xlast[n] for n in dangling_nodes)
for n in x:
# this matrix multiply looks odd because it is
# doing a left multiply x^T=xlast^T*W
for _, nbr, wt in W.edges(n, data=weight):
x[nbr] += alpha * xlast[n] * wt
x[n] += danglesum * dangling_weights.get(n, 0) + (1.0 - alpha) * p.get(n, 0)
# check convergence, l1 norm
err = sum(abs(x[n] - xlast[n]) for n in x)
if err < N * tol:
return x
raise nx.PowerIterationFailedConvergence(max_iter)
def google_matrix(
G, alpha=0.85, personalization=None, nodelist=None, weight="weight", dangling=None
):
"""Returns the Google matrix of the graph.
Parameters
----------
G : graph
A NetworkX graph. Undirected graphs will be converted to a directed
graph with two directed edges for each undirected edge.
alpha : float
The damping factor.
personalization: dict, optional
The "personalization vector" consisting of a dictionary with a
key some subset of graph nodes and personalization value each of those.
At least one personalization value must be non-zero.
If not specfiied, a nodes personalization value will be zero.
By default, a uniform distribution is used.
nodelist : list, optional
The rows and columns are ordered according to the nodes in nodelist.
If nodelist is None, then the ordering is produced by G.nodes().
weight : key, optional
Edge data key to use as weight. If None weights are set to 1.
dangling: dict, optional
The outedges to be assigned to any "dangling" nodes, i.e., nodes without
any outedges. The dict key is the node the outedge points to and the dict
value is the weight of that outedge. By default, dangling nodes are given
outedges according to the personalization vector (uniform if not
specified) This must be selected to result in an irreducible transition
matrix (see notes below). It may be common to have the dangling dict to
be the same as the personalization dict.
Returns
-------
A : 2D NumPy ndarray
Google matrix of the graph
Notes
-----
The array returned represents the transition matrix that describes the
Markov chain used in PageRank. For PageRank to converge to a unique
solution (i.e., a unique stationary distribution in a Markov chain), the
transition matrix must be irreducible. In other words, it must be that
there exists a path between every pair of nodes in the graph, or else there
is the potential of "rank sinks."
This implementation works with Multi(Di)Graphs. For multigraphs the
weight between two nodes is set to be the sum of all edge weights
between those nodes.
See Also
--------
pagerank
"""
import numpy as np
if nodelist is None:
nodelist = list(G)
A = nx.to_numpy_array(G, nodelist=nodelist, weight=weight)
N = len(G)
if N == 0:
return A
# Personalization vector
if personalization is None:
p = np.repeat(1.0 / N, N)
else:
p = np.array([personalization.get(n, 0) for n in nodelist], dtype=float)
if p.sum() == 0:
raise ZeroDivisionError
p /= p.sum()
# Dangling nodes
if dangling is None:
dangling_weights = p
else:
# Convert the dangling dictionary into an array in nodelist order
dangling_weights = np.array([dangling.get(n, 0) for n in nodelist], dtype=float)
dangling_weights /= dangling_weights.sum()
dangling_nodes = np.where(A.sum(axis=1) == 0)[0]
# Assign dangling_weights to any dangling nodes (nodes with no out links)
A[dangling_nodes] = dangling_weights
A /= A.sum(axis=1)[:, np.newaxis] # Normalize rows to sum to 1
return alpha * A + (1 - alpha) * p
def _pagerank_numpy(
G, alpha=0.85, personalization=None, weight="weight", dangling=None
):
"""Returns the PageRank of the nodes in the graph.
PageRank computes a ranking of the nodes in the graph G based on
the structure of the incoming links. It was originally designed as
an algorithm to rank web pages.
Parameters
----------
G : graph
A NetworkX graph. Undirected graphs will be converted to a directed
graph with two directed edges for each undirected edge.
alpha : float, optional
Damping parameter for PageRank, default=0.85.
personalization: dict, optional
The "personalization vector" consisting of a dictionary with a
key some subset of graph nodes and personalization value each of those.
At least one personalization value must be non-zero.
If not specfiied, a nodes personalization value will be zero.
By default, a uniform distribution is used.
weight : key, optional
Edge data key to use as weight. If None weights are set to 1.
dangling: dict, optional
The outedges to be assigned to any "dangling" nodes, i.e., nodes without
any outedges. The dict key is the node the outedge points to and the dict
value is the weight of that outedge. By default, dangling nodes are given
outedges according to the personalization vector (uniform if not
specified) This must be selected to result in an irreducible transition
matrix (see notes under google_matrix). It may be common to have the
dangling dict to be the same as the personalization dict.
Returns
-------
pagerank : dictionary
Dictionary of nodes with PageRank as value.
Examples
--------
>>> from networkx.algorithms.link_analysis.pagerank_alg import _pagerank_numpy
>>> G = nx.DiGraph(nx.path_graph(4))
>>> pr = _pagerank_numpy(G, alpha=0.9)
Notes
-----
The eigenvector calculation uses NumPy's interface to the LAPACK
eigenvalue solvers. This will be the fastest and most accurate
for small graphs.
This implementation works with Multi(Di)Graphs. For multigraphs the
weight between two nodes is set to be the sum of all edge weights
between those nodes.
See Also
--------
pagerank, google_matrix
References
----------
.. [1] A. Langville and C. Meyer,
"A survey of eigenvector methods of web information retrieval."
http://citeseer.ist.psu.edu/713792.html
.. [2] Page, Lawrence; Brin, Sergey; Motwani, Rajeev and Winograd, Terry,
The PageRank citation ranking: Bringing order to the Web. 1999
http://dbpubs.stanford.edu:8090/pub/showDoc.Fulltext?lang=en&doc=1999-66&format=pdf
"""
import numpy as np
if len(G) == 0:
return {}
M = google_matrix(
G, alpha, personalization=personalization, weight=weight, dangling=dangling
)
# use numpy LAPACK solver
eigenvalues, eigenvectors = np.linalg.eig(M.T)
ind = np.argmax(eigenvalues)
# eigenvector of largest eigenvalue is at ind, normalized
largest = np.array(eigenvectors[:, ind]).flatten().real
norm = largest.sum()
return dict(zip(G, map(float, largest / norm)))
def _pagerank_scipy(
G,
alpha=0.85,
personalization=None,
max_iter=100,
tol=1.0e-6,
nstart=None,
weight="weight",
dangling=None,
):
"""Returns the PageRank of the nodes in the graph.
PageRank computes a ranking of the nodes in the graph G based on
the structure of the incoming links. It was originally designed as
an algorithm to rank web pages.
Parameters
----------
G : graph
A NetworkX graph. Undirected graphs will be converted to a directed
graph with two directed edges for each undirected edge.
alpha : float, optional
Damping parameter for PageRank, default=0.85.
personalization: dict, optional
The "personalization vector" consisting of a dictionary with a
key some subset of graph nodes and personalization value each of those.
At least one personalization value must be non-zero.
If not specfiied, a nodes personalization value will be zero.
By default, a uniform distribution is used.
max_iter : integer, optional
Maximum number of iterations in power method eigenvalue solver.
tol : float, optional
Error tolerance used to check convergence in power method solver.
nstart : dictionary, optional
Starting value of PageRank iteration for each node.
weight : key, optional
Edge data key to use as weight. If None weights are set to 1.
dangling: dict, optional
The outedges to be assigned to any "dangling" nodes, i.e., nodes without
any outedges. The dict key is the node the outedge points to and the dict
value is the weight of that outedge. By default, dangling nodes are given
outedges according to the personalization vector (uniform if not
specified) This must be selected to result in an irreducible transition
matrix (see notes under google_matrix). It may be common to have the
dangling dict to be the same as the personalization dict.
Returns
-------
pagerank : dictionary
Dictionary of nodes with PageRank as value
Examples
--------
>>> from networkx.algorithms.link_analysis.pagerank_alg import _pagerank_scipy
>>> G = nx.DiGraph(nx.path_graph(4))
>>> pr = _pagerank_scipy(G, alpha=0.9)
Notes
-----
The eigenvector calculation uses power iteration with a SciPy
sparse matrix representation.
This implementation works with Multi(Di)Graphs. For multigraphs the
weight between two nodes is set to be the sum of all edge weights
between those nodes.
See Also
--------
pagerank
Raises
------
PowerIterationFailedConvergence
If the algorithm fails to converge to the specified tolerance
within the specified number of iterations of the power iteration
method.
References
----------
.. [1] A. Langville and C. Meyer,
"A survey of eigenvector methods of web information retrieval."
http://citeseer.ist.psu.edu/713792.html
.. [2] Page, Lawrence; Brin, Sergey; Motwani, Rajeev and Winograd, Terry,
The PageRank citation ranking: Bringing order to the Web. 1999
http://dbpubs.stanford.edu:8090/pub/showDoc.Fulltext?lang=en&doc=1999-66&format=pdf
"""
import numpy as np
import scipy as sp
import scipy.sparse # call as sp.sparse
N = len(G)
if N == 0:
return {}
nodelist = list(G)
A = nx.to_scipy_sparse_array(G, nodelist=nodelist, weight=weight, dtype=float)
S = A.sum(axis=1)
S[S != 0] = 1.0 / S[S != 0]
# TODO: csr_array
Q = sp.sparse.csr_array(sp.sparse.spdiags(S.T, 0, *A.shape))
A = Q @ A
# initial vector
if nstart is None:
x = np.repeat(1.0 / N, N)
else:
x = np.array([nstart.get(n, 0) for n in nodelist], dtype=float)
x /= x.sum()
# Personalization vector
if personalization is None:
p = np.repeat(1.0 / N, N)
else:
p = np.array([personalization.get(n, 0) for n in nodelist], dtype=float)
if p.sum() == 0:
raise ZeroDivisionError
p /= p.sum()
# Dangling nodes
if dangling is None:
dangling_weights = p
else:
# Convert the dangling dictionary into an array in nodelist order
dangling_weights = np.array([dangling.get(n, 0) for n in nodelist], dtype=float)
dangling_weights /= dangling_weights.sum()
is_dangling = np.where(S == 0)[0]
# power iteration: make up to max_iter iterations
for _ in range(max_iter):
xlast = x
x = alpha * (x @ A + sum(x[is_dangling]) * dangling_weights) + (1 - alpha) * p
# check convergence, l1 norm
err = np.absolute(x - xlast).sum()
if err < N * tol:
return dict(zip(nodelist, map(float, x)))
raise nx.PowerIterationFailedConvergence(max_iter)
|