def test_partition_distance():
q = load_sample_group_qball()
d = load_sample_group_dsi()
q = np.mean(q, axis=2)
d = np.mean(d, axis=2)
qi, _ = bct.modularity_und(q)
di, _ = bct.modularity_und(d)
vi, mi = bct.partition_distance(qi, di)
print(vi, mi)
assert np.allclose(vi, 0.1964, atol=0.01)
assert np.allclose(mi, 0.6394, atol=0.01)
def entropic_similarity(a1, a2):
ma, _ = bct.modularity_und(a1)
mb, _ = bct.modularity_und(a2)
vi, _ = bct.partition_distance(ma, mb)
return 1 - vi
#**********************************
# plot number of communities as a function of threshold
thresholds = np.arange(.15, .35, .01)
partition_distances = []
cluster_size = []
for t in thresholds:
# simulate threhsold 100 times
clusters = simulate(100,
fun=lambda: Graph_Analysis(
corr_data, threshold=t, weight=True, display=False)
[0].vs['community'])
distances = []
for i, c1 in enumerate(clusters):
for c2 in clusters[i:]:
distances.append(bct.partition_distance(c1, c2)[0])
partition_distances += zip([t] * len(distances), distances)
cluster_size += (zip([t] * len(clusters), [max(c) for c in clusters]))
plt.figure(figsize=(16, 12))
sns.stripplot(x=list(zip(*cluster_size))[0],
y=list(zip(*cluster_size))[1],
jitter=.4)
plt.ylabel('Number of detected communities', size=20)
plt.xlabel('Threshold', size=20)
plt.figure(figsize=(16, 12))
#sns.stripplot(x = zip(*partition_distances)[0], y = zip(*partition_distances)[1], jitter = .2)
sns.boxplot(x=list(zip(*partition_distances))[0],
y=list(zip(*partition_distances))[1])
plt.ylabel('Partition distance over 100 repetitions', size=20)
def entropic_similarity(a1, a2):
ma, _ = bct.modularity_und(a1)
mb, _ = bct.modularity_und(a2)
vi, _ = bct.partition_distance(ma, mb)
return 1 - vi