def overlapping_normalized_mutual_information_MGH(self,
clustering,
normalization="max"):
"""
Overlapping Normalized Mutual Information between two clusterings.
Extension of the Normalized Mutual Information (NMI) score to cope with overlapping partitions.
This is the version proposed by McDaid et al. using a different normalization than the original LFR one. See ref.
for more details.
:param clustering: NodeClustering object
:param normalization: one of "max" or "LFK". Default "max" (corresponds to the main method described in the article)
:return: onmi score
:Example:
>>> from cdlib import evaluation, algorithms
>>> g = nx.karate_club_graph()
>>> louvain_communities = algorithms.louvain(g)
>>> leiden_communities = algorithms.leiden(g)
>>> evaluation.overlapping_normalized_mutual_information_MGH(louvain_communities,leiden_communities)
:Reference:
1. McDaid, A. F., Greene, D., & Hurley, N. (2011). Normalized mutual information to evaluate overlapping community finding algorithms. arXiv preprint arXiv:1110.2515. Chicago
"""
return evaluation.overlapping_normalized_mutual_information_MGH(
self, clustering, normalization=normalization)
def test_onmi(self):
g = nx.karate_club_graph()
lp_communities = label_propagation(g)
lp2_communities = label_propagation(g)
score = evaluation.overlapping_normalized_mutual_information_MGH(lp2_communities, lp_communities)
self.assertLessEqual(score.score, 1)
self.assertGreaterEqual(score.score, 0)
score = evaluation.overlapping_normalized_mutual_information_LFK(lp2_communities, lp_communities)
self.assertLessEqual(score.score, 1)
self.assertGreaterEqual(score.score, 0)
NMI_result[N] = dict()
Omega_result[N] = dict()
for mu in [0.1, 0.3]:
NMI_result[N][mu] = {'NMI': [], 'NMI_max': []}
Omega_result[N][mu] = {'omega': []}
for om in range(2, 9):
name = 'LFR/N-{}-mu{:1.1f}-om{}'.format(N, mu, om)
communities = get_communities(
os.path.join("../../data/", name, "result.dat"), " ", " ",
False)
gt_communities = get_communities(
os.path.join("../../data/", name, "community.dat"), "\t",
" ", True)
G = nx.read_adjlist(os.path.join("../../data/", name,
"network.dat"),
nodetype=int)
coms = NodeClustering(communities, G, "", overlap=True)
gt_coms = NodeClustering(gt_communities, G, "", overlap=True)
nmi = evaluation.overlapping_normalized_mutual_information_LFK(
coms, gt_coms)[0]
nmi_max = evaluation.overlapping_normalized_mutual_information_MGH(
coms, gt_coms)[0]
omega = evaluation.omega(coms, gt_coms)[0]
NMI_result[N][mu]['NMI'].append(nmi)
NMI_result[N][mu]['NMI_max'].append(nmi_max)
Omega_result[N][mu]['omega'].append(omega)
with open('LFR_nmi.json', 'w') as f:
json.dump(NMI_result, f)
with open('LFR_omega.json', 'w') as f:
json.dump(Omega_result, f)