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
|
"""Current-flow betweenness centrality measures for subsets of nodes."""
import networkx as nx
from networkx.algorithms.centrality.flow_matrix import flow_matrix_row
from networkx.utils import not_implemented_for, reverse_cuthill_mckee_ordering
__all__ = [
"current_flow_betweenness_centrality_subset",
"edge_current_flow_betweenness_centrality_subset",
]
@not_implemented_for("directed")
def current_flow_betweenness_centrality_subset(
G, sources, targets, normalized=True, weight=None, dtype=float, solver="lu"
):
r"""Compute current-flow betweenness centrality for subsets of nodes.
Current-flow betweenness centrality uses an electrical current
model for information spreading in contrast to betweenness
centrality which uses shortest paths.
Current-flow betweenness centrality is also known as
random-walk betweenness centrality [2]_.
Parameters
----------
G : graph
A NetworkX graph
sources: list of nodes
Nodes to use as sources for current
targets: list of nodes
Nodes to use as sinks for current
normalized : bool, optional (default=True)
If True the betweenness values are normalized by b=b/(n-1)(n-2) where
n is the number of nodes in G.
weight : string or None, optional (default=None)
Key for edge data used as the edge weight.
If None, then use 1 as each edge weight.
dtype: data type (float)
Default data type for internal matrices.
Set to np.float32 for lower memory consumption.
solver: string (default='lu')
Type of linear solver to use for computing the flow matrix.
Options are "full" (uses most memory), "lu" (recommended), and
"cg" (uses least memory).
Returns
-------
nodes : dictionary
Dictionary of nodes with betweenness centrality as the value.
See Also
--------
approximate_current_flow_betweenness_centrality
betweenness_centrality
edge_betweenness_centrality
edge_current_flow_betweenness_centrality
Notes
-----
Current-flow betweenness can be computed in $O(I(n-1)+mn \log n)$
time [1]_, where $I(n-1)$ is the time needed to compute the
inverse Laplacian. For a full matrix this is $O(n^3)$ but using
sparse methods you can achieve $O(nm{\sqrt k})$ where $k$ is the
Laplacian matrix condition number.
The space required is $O(nw)$ where $w$ is the width of the sparse
Laplacian matrix. Worse case is $w=n$ for $O(n^2)$.
If the edges have a 'weight' attribute they will be used as
weights in this algorithm. Unspecified weights are set to 1.
References
----------
.. [1] Centrality Measures Based on Current Flow.
Ulrik Brandes and Daniel Fleischer,
Proc. 22nd Symp. Theoretical Aspects of Computer Science (STACS '05).
LNCS 3404, pp. 533-544. Springer-Verlag, 2005.
http://algo.uni-konstanz.de/publications/bf-cmbcf-05.pdf
.. [2] A measure of betweenness centrality based on random walks,
M. E. J. Newman, Social Networks 27, 39-54 (2005).
"""
from networkx.utils import reverse_cuthill_mckee_ordering
import numpy as np
if not nx.is_connected(G):
raise nx.NetworkXError("Graph not connected.")
n = G.number_of_nodes()
ordering = list(reverse_cuthill_mckee_ordering(G))
# make a copy with integer labels according to rcm ordering
# this could be done without a copy if we really wanted to
mapping = dict(zip(ordering, range(n)))
H = nx.relabel_nodes(G, mapping)
betweenness = dict.fromkeys(H, 0.0) # b[v]=0 for v in H
for row, (s, t) in flow_matrix_row(H, weight=weight, dtype=dtype, solver=solver):
for ss in sources:
i = mapping[ss]
for tt in targets:
j = mapping[tt]
betweenness[s] += 0.5 * np.abs(row[i] - row[j])
betweenness[t] += 0.5 * np.abs(row[i] - row[j])
if normalized:
nb = (n - 1.0) * (n - 2.0) # normalization factor
else:
nb = 2.0
for v in H:
betweenness[v] = betweenness[v] / nb + 1.0 / (2 - n)
return {ordering[k]: v for k, v in betweenness.items()}
@not_implemented_for("directed")
def edge_current_flow_betweenness_centrality_subset(
G, sources, targets, normalized=True, weight=None, dtype=float, solver="lu"
):
r"""Compute current-flow betweenness centrality for edges using subsets
of nodes.
Current-flow betweenness centrality uses an electrical current
model for information spreading in contrast to betweenness
centrality which uses shortest paths.
Current-flow betweenness centrality is also known as
random-walk betweenness centrality [2]_.
Parameters
----------
G : graph
A NetworkX graph
sources: list of nodes
Nodes to use as sources for current
targets: list of nodes
Nodes to use as sinks for current
normalized : bool, optional (default=True)
If True the betweenness values are normalized by b=b/(n-1)(n-2) where
n is the number of nodes in G.
weight : string or None, optional (default=None)
Key for edge data used as the edge weight.
If None, then use 1 as each edge weight.
dtype: data type (float)
Default data type for internal matrices.
Set to np.float32 for lower memory consumption.
solver: string (default='lu')
Type of linear solver to use for computing the flow matrix.
Options are "full" (uses most memory), "lu" (recommended), and
"cg" (uses least memory).
Returns
-------
nodes : dict
Dictionary of edge tuples with betweenness centrality as the value.
See Also
--------
betweenness_centrality
edge_betweenness_centrality
current_flow_betweenness_centrality
Notes
-----
Current-flow betweenness can be computed in $O(I(n-1)+mn \log n)$
time [1]_, where $I(n-1)$ is the time needed to compute the
inverse Laplacian. For a full matrix this is $O(n^3)$ but using
sparse methods you can achieve $O(nm{\sqrt k})$ where $k$ is the
Laplacian matrix condition number.
The space required is $O(nw)$ where $w$ is the width of the sparse
Laplacian matrix. Worse case is $w=n$ for $O(n^2)$.
If the edges have a 'weight' attribute they will be used as
weights in this algorithm. Unspecified weights are set to 1.
References
----------
.. [1] Centrality Measures Based on Current Flow.
Ulrik Brandes and Daniel Fleischer,
Proc. 22nd Symp. Theoretical Aspects of Computer Science (STACS '05).
LNCS 3404, pp. 533-544. Springer-Verlag, 2005.
http://algo.uni-konstanz.de/publications/bf-cmbcf-05.pdf
.. [2] A measure of betweenness centrality based on random walks,
M. E. J. Newman, Social Networks 27, 39-54 (2005).
"""
import numpy as np
if not nx.is_connected(G):
raise nx.NetworkXError("Graph not connected.")
n = G.number_of_nodes()
ordering = list(reverse_cuthill_mckee_ordering(G))
# make a copy with integer labels according to rcm ordering
# this could be done without a copy if we really wanted to
mapping = dict(zip(ordering, range(n)))
H = nx.relabel_nodes(G, mapping)
edges = (tuple(sorted((u, v))) for u, v in H.edges())
betweenness = dict.fromkeys(edges, 0.0)
if normalized:
nb = (n - 1.0) * (n - 2.0) # normalization factor
else:
nb = 2.0
for row, (e) in flow_matrix_row(H, weight=weight, dtype=dtype, solver=solver):
for ss in sources:
i = mapping[ss]
for tt in targets:
j = mapping[tt]
betweenness[e] += 0.5 * np.abs(row[i] - row[j])
betweenness[e] /= nb
return {(ordering[s], ordering[t]): v for (s, t), v in betweenness.items()}
|