def current_flow_closeness_centrality(G,
weight=None,
dtype=float,
solver="lu"):
"""Compute current-flow closeness centrality for nodes.
Current-flow closeness centrality is variant of closeness
centrality based on effective resistance between nodes in
a network. This metric is also known as information centrality.
Parameters
----------
G : graph
A NetworkX graph.
weight : None or string, optional (default=None)
If None, all edge weights are considered equal.
Otherwise holds the name of the edge attribute used as weight.
The weight reflects the capacity or the strength of the
edge.
dtype: data type (default=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 current flow closeness centrality as the value.
See Also
--------
closeness_centrality
Notes
-----
The algorithm is from Brandes [1]_.
See also [2]_ for the original definition of information centrality.
References
----------
.. [1] Ulrik Brandes and Daniel Fleischer,
Centrality Measures Based on Current Flow.
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] Karen Stephenson and Marvin Zelen:
Rethinking centrality: Methods and examples.
Social Networks 11(1):1-37, 1989.
https://doi.org/10.1016/0378-8733(89)90016-6
"""
if not nx.is_connected(G):
raise nx.NetworkXError("Graph not connected.")
solvername = {
"full": FullInverseLaplacian,
"lu": SuperLUInverseLaplacian,
"cg": CGInverseLaplacian,
}
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
H = nx.relabel_nodes(G, dict(zip(ordering, range(n))))
betweenness = dict.fromkeys(H, 0.0) # b[v]=0 for v in H
n = H.number_of_nodes()
L = laplacian_sparse_matrix(H,
nodelist=range(n),
weight=weight,
dtype=dtype,
format="csc")
C2 = solvername[solver](L, width=1, dtype=dtype) # initialize solver
for v in H:
col = C2.get_row(v)
for w in H:
betweenness[v] += col[v] - 2 * col[w]
betweenness[w] += col[v]
for v in H:
betweenness[v] = 1.0 / (betweenness[v])
return {ordering[k]: float(v) for k, v in betweenness.items()}
def approximate_current_flow_betweenness_centrality(G,
normalized=True,
weight=None,
dtype=float,
solver='full',
epsilon=0.5,
kmax=10000,
seed=None):
r"""Compute the approximate current-flow betweenness centrality for nodes.
Approximates the current-flow betweenness centrality within absolute
error of epsilon with high probability [1]_.
Parameters
----------
G : graph
A NetworkX graph
normalized : bool, optional (default=True)
If True the betweenness values are normalized by 2/[(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).
epsilon: float
Absolute error tolerance.
kmax: int
Maximum number of sample node pairs to use for approximation.
seed : integer, random_state, or None (default)
Indicator of random number generation state.
See :ref:`Randomness<randomness>`.
Returns
-------
nodes : dictionary
Dictionary of nodes with betweenness centrality as the value.
See Also
--------
current_flow_betweenness_centrality
Notes
-----
The running time is $O((1/\epsilon^2)m{\sqrt k} \log n)$
and the space required is $O(m)$ for $n$ nodes and $m$ edges.
If the edges have a 'weight' attribute they will be used as
weights in this algorithm. Unspecified weights are set to 1.
References
----------
.. [1] Ulrik Brandes and Daniel Fleischer:
Centrality Measures Based on Current Flow.
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
"""
try:
import numpy as np
except ImportError:
raise ImportError(
'current_flow_betweenness_centrality requires NumPy ',
'http://scipy.org/')
try:
from scipy import sparse
from scipy.sparse import linalg
except ImportError:
raise ImportError(
'current_flow_betweenness_centrality requires SciPy ',
'http://scipy.org/')
if not nx.is_connected(G):
raise nx.NetworkXError("Graph not connected.")
solvername = {
"full": FullInverseLaplacian,
"lu": SuperLUInverseLaplacian,
"cg": CGInverseLaplacian
}
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
H = nx.relabel_nodes(G, dict(zip(ordering, range(n))))
L = laplacian_sparse_matrix(H,
nodelist=range(n),
weight=weight,
dtype=dtype,
format='csc')
C = solvername[solver](L, dtype=dtype) # initialize solver
betweenness = dict.fromkeys(H, 0.0)
nb = (n - 1.0) * (n - 2.0) # normalization factor
cstar = n * (n - 1) / nb
l = 1 # parameter in approximation, adjustable
k = l * int(np.ceil((cstar / epsilon)**2 * np.log(n)))
if k > kmax:
msg = 'Number random pairs k>kmax (%d>%d) ' % (k, kmax)
raise nx.NetworkXError(msg, 'Increase kmax or epsilon')
cstar2k = cstar / (2 * k)
for i in range(k):
s, t = seed.sample(range(n), 2)
b = np.zeros(n, dtype=dtype)
b[s] = 1
b[t] = -1
p = C.solve(b)
for v in H:
if v == s or v == t:
continue
for nbr in H[v]:
w = H[v][nbr].get(weight, 1.0)
betweenness[v] += w * np.abs(p[v] - p[nbr]) * cstar2k
if normalized:
factor = 1.0
else:
factor = nb / 2.0
# remap to original node names and "unnormalize" if required
return dict(
(ordering[k], float(v * factor)) for k, v in betweenness.items())
