def test_counts():
n = 100
p = 0.1
A = nsr.erdos_renyi(n, p)
print("Motifs (fast):")
f = nsm.motifs(A)
print("\tf = %s" % f)
# print "Motifs (v0):"
# f_v0 = motifs_v0(A)
# print "\tf_v0 =", f_v0
# print "Comparison: ", zip(f, f_v0)
# npt.assert_equal(f[3:], f_v0[3:], "Motif frequencies of different versions are not compatible")
print("Motifs (naive):")
f_naive = nsm.motifs(A, algorithm='brute-force')
print("\tf_naive = %s" % f_naive)
print("Comparison: %s" % list(zip(f, f_naive)))
npt.assert_equal(
f[3:], f_naive[3:],
"Motif frequencies from optimized algorithm are not correct")
C3n = comb(n, 3)
npt.assert_equal(sum(f_naive), C3n)
def test_a_single_pattern_permuations():
tid = 6
M = nsm.triad_patterns()[tid]
# plot_a_triad(M, r=.5)
A = np.zeros(shape=(5, 5), dtype=np.int64)
A[1:4, 1:4] = M
expected_f = np.eye(16, dtype=np.int64)[tid]
for permutation in itertools.permutations(range(5), 5):
print("# permutation:", permutation)
perm_B = permute_array(A, permutation)
print(perm_B)
f = nsm.motifs(perm_B)
f[0:3] = 0
npt.assert_equal(f, expected_f)
f, participation = nsm.motifs(perm_B, participation=True)
f[0:3] = 0
print(participation)
npt.assert_equal(f, expected_f)
npt.assert_equal(
[len(p) for p in participation][3:], f[3:],
"Frequency and participation lists are not consistent")
r, g, b = [
np.where(np.array(permutation) == i)[0][0] for i in [1, 2, 3]
]
print("(R=%d, G=%d, B=%d) vs. participation=%s\n" %
(r, g, b, participation[tid][0]))
npt.assert_array_equal(participation[tid][0], [r, g, b])
def create_graph_motifs(self):
"""Creates the graph that contains only motifs of the type A->B<-C,
using the results of the detected motifs.
"""
motifs = nsm.motifs(
self.binary_adjacency_matrix.astype(int),
algorithm='louzoun', participation=True)
G_only_motif = nx.DiGraph()
for triplet in motifs[1][3]:
G_only_motif.add_node(triplet[0])
G_only_motif.add_node(triplet[1])
G_only_motif.add_node(triplet[2])
G_only_motif.add_edge(triplet[0], triplet[1])
G_only_motif.add_edge(triplet[2], triplet[1])
new_labels = {}
for node in G_only_motif.nodes:
new_labels[node] = self.channels[node]
nx.relabel.relabel_nodes(G_only_motif, new_labels, copy=False)
nodes = set(self.G.nodes)
m_nodes = set(G_only_motif)
nodes_diff = nodes.difference(m_nodes)
G_only_motif.add_nodes_from(nodes_diff)
print("Density:", nx.density(G_only_motif))
plt.figure(figsize=(12, 10))
nx.draw_networkx(G_only_motif, pos=self.channel_locations, arrowsize=1,
node_color='lightcyan', edge_color='silver')
def test_basic_patterns():
print("Testing basic patterns..")
M = nsm.triad_patterns()
for i in np.arange(len(M)):
motif = M[i].copy()
print("Motif #%d\n%s" % (i, motif))
f_naive = nsm.motifs(motif, algorithm='brute-force')
npt.assert_equal(motif, M[i],
"Argument array was altered during execution")
print("\t%s %s" % (f_naive[0:3], f_naive[3:]))
npt.assert_equal(np.where(f_naive == 1)[0], np.array([i]))
if i >= 3:
f = nsm.motifs(motif)
npt.assert_equal(motif, M[i],
"Argument array was altered during execution")
print("\t%s %s" % (f[0:3], f[3:]))
npt.assert_equal(np.where(f == 1)[0], np.array([i]))
def compute_motifs(self, algorithm):
"""Compute motifs of a given graph with netsci package,
using the specified algorithm.
Args:
----------
algorithm : string
2 options:
- 'brute-force' : the naive implementation using brute force algorithm.
The complexity is high $(O(|V|^3))$, counts all 16 triplets.
- 'louzoun' : an efficient algorithm for sparse networks.
The complexity is low $(O(|E|))$, but it counts only the 13 connected
triplets (the first 3 entries will be -1).
"""
motifs = nsm.motifs(
self.binary_adjacency_matrix.astype(int), algorithm=algorithm)
print(motifs)
nsv.bar_motifs(motifs)
def test_geometrical_network():
n = 10
Ys = np.random.rand(n)
Is, Js = np.mgrid[0:n, 0:n]
P = (Ys[Is] < Ys[Js]).astype(np.int8)
A = np.random.binomial(1, P)
f, participation = nsm.motifs(A, participation=True)
npt.assert_equal([len(p) for p in participation][3:], f[3:],
"Frequency and participation lists are not consistent")
part6 = participation[6]
assert np.all([(Ys[r] < Ys[g] < Ys[b]) for (r, g, b) in part6])
fy = Ys[np.array(part6)]
plt.plot([1, 2, 3], fy.T, '-o', zorder=1, alpha=.25, lw=1)
plt.errorbar([1, 2, 3],
fy.mean(axis=0),
fy.std(axis=0),
lw=2,
color='k',
zorder=2)
print("Comparison: %s" % list(zip(f, f_naive)))
npt.assert_equal(
f[3:], f_naive[3:],
"Motif frequencies from optimized algorithm are not correct")
C3n = comb(n, 3)
npt.assert_equal(sum(f_naive), C3n)
if False:
import netsci.visualization.motifs as nsv
n, p = 250, 0.2
A = nsr.er(n, p)
C3n = comb(n, 3)
f_analytic = C3n * triads_mean(p)
print("Counting motifs..")
f = nsm.motifs(A)
print("Plotting..")
nsv.bar_motifs(f,
f_analytic,
triad_order=triad_order_nn4576,
title="nn4576 triad order")
nsv.bar_motifs(f,
f_analytic,
triad_order=triad_order_bct,
title="BCT triad order")
import logging
logging.basicConfig()
logging.getLogger('netsci').setLevel(logging.DEBUG)
import numpy as np
import numpy.testing as npt
from netsci.models.random import erdos_renyi
import netsci.metrics.motifs as nsm
A = erdos_renyi(100, p=0.01)
f = nsm.motifs(A, algorithm='matmul')
print(f"f (matmul) = {f}")
f_expected = nsm.motifs(A)
print(f"f_expected = {f_expected}")
npt.assert_equal(f[3:], f_expected[3:])
f_expected_bf = nsm.motifs(A, algorithm='brute-force')
print(f"f_expected_bf = {f_expected_bf}")
npt.assert_equal(f, f_expected_bf)
from tensorflow.python.client import device_lib
print(
'devices:',
{d.name: d.physical_device_desc
for d in device_lib.list_local_devices()})
A = erdos_renyi(n=2000, p=0.1, random_state=71070)
print(f'A: {A.shape}, p = {nsm.sparsity(A):.4f}')
configurations = [('matmul', 'GPU'), ('matmul', 'CPU'), ('louzoun', 'CPU')]
for algorithm, device in configurations:
with tf.device(f'/{device}:0'):
with Timer(logger=print, desc=f'{algorithm}:{device}'):
print(nsm.motifs(A, algorithm=algorithm))
"""Colab
devices = {
'/device:CPU:0': '',
'/device:GPU:0': 'device: 0, name: Tesla K80, pci bus id: 0000:00:04.0, compute capability: 3.7'
}
A: (2000, 2000), p = 0.1001
matmul:GPU took 0.680 sec.
matmul:CPU took 18.909 sec.
louzoun:CPU took 149.726 sec.
"""
"""MBP M1 Max
devices: {
'/device:CPU:0': '',