def test_graph_is_subgraph_with_labels(self):
G = nx.Graph()
G.add_edges_from(
[
(1, 2, {'label': 'a'}),
(2, 3, {'label': 'c'}),
(3, 4, {'label': 'd'}),
(2, 4, {'label': 'b'})
]
)
valid_subgraph_a = nx.Graph()
valid_subgraph_a.add_edge(1, 2, label='a')
valid_subgraph_b = nx.Graph()
valid_subgraph_b.add_edge(3, 4, label='d')
valid_subgraph_c = nx.Graph()
valid_subgraph_c.add_edge(2, 1, label='a')
invalid_subgraph_a = nx.Graph()
invalid_subgraph_a.add_edge(1, 2, label='b')
invalid_subgraph_b = nx.Graph()
invalid_subgraph_b.add_edge(1, 4, label='a')
self.assertTrue(ParsemisMiner.is_subgraph(G, valid_subgraph_a))
self.assertTrue(ParsemisMiner.is_subgraph(G, valid_subgraph_b))
self.assertTrue(ParsemisMiner.is_subgraph(G, valid_subgraph_c))
self.assertFalse(ParsemisMiner.is_subgraph(G, invalid_subgraph_a))
self.assertFalse(ParsemisMiner.is_subgraph(G, invalid_subgraph_b))
def test_graph_is_subgraph_without_labels(self):
G = nx.Graph()
G.add_edges_from(
[
(1, 2),
(2, 3),
(3, 4),
(2, 4)
]
)
valid_subgraph_a = nx.Graph()
valid_subgraph_a.add_edge(1, 2)
valid_subgraph_b = nx.Graph()
valid_subgraph_b.add_edge(3, 4)
valid_subgraph_c = nx.Graph()
valid_subgraph_c.add_edge(4, 3)
invalid_subgraph_b = nx.Graph()
invalid_subgraph_b.add_edge(1, 4)
self.assertTrue(ParsemisMiner.is_subgraph(G, valid_subgraph_a))
self.assertTrue(ParsemisMiner.is_subgraph(G, valid_subgraph_b))
self.assertTrue(ParsemisMiner.is_subgraph(G, valid_subgraph_c))
self.assertFalse(ParsemisMiner.is_subgraph(G, invalid_subgraph_b))
def test_similarity_when_same(self):
super_graph = nx.Graph()
super_graph.add_edge(1, 2, label='a')
super_graph.add_edge(2, 3, label='b')
sub_graph = super_graph.copy()
self.assertEqual(ParsemisMiner.calculate_dot_product_similarity(super_graph, sub_graph), 1, 'Dot Product')
self.assertEqual(ParsemisMiner.calculate_jaccard_similarity(super_graph, sub_graph), 1, 'Jaccard Similarity')
def test_similarity_when_different(self):
super_graph = nx.Graph()
super_graph.add_edge(1, 2, label='a')
super_graph.add_edge(2, 3, label='b')
sub_graph = nx.Graph()
sub_graph.add_edge(4, 5, label='c')
sub_graph.add_edge(5, 6, label='d')
self.assertEqual(ParsemisMiner.calculate_dot_product_similarity(super_graph, sub_graph), 0, 'Dot Product')
self.assertEqual(ParsemisMiner.calculate_jaccard_similarity(super_graph, sub_graph), 0, 'Jaccard Similarity')
This short example shows how you can mine frequent graphs using the wrapper
"""
from parsemis.parsemis_wrapper import ParsemisMiner
import networkx as nx
import os
# Load our graphs
graph_folder = "example_dataset"
graphs = []
for f in os.listdir(graph_folder):
path = "%s/%s" % (graph_folder, f)
graphs.append(nx.read_gml(path))
frequent_graphs = ParsemisMiner("data",
minimum_frequency="1%",
close_graph=True,
store_embeddings=False,
debug=True,
mine_undirected=True).mine_graphs(graphs)
# Count our subgraphs
frequent_graph_counts = []
for frequent_graph in frequent_graphs:
count = 0
for graph in graphs:
if graph.graph['id'] in frequent_graph.appears_in:
count += 1
frequent_graph_counts.append((count, frequent_graph))
for frequent_graph in sorted(frequent_graph_counts,
key=lambda subgraph: subgraph[0],
reverse=True):