def _create_feature_extractor(self):
bigraph = Bigraph(self.graph, source_cls=User, target_cls=Repository)
features = []
if 'behavior' in self.feature_types:
features.extend([
(bigraph.matrix.copy(), 'source', 'contributed repository'),
(bigraph.matrix.T.tocsr().copy(), 'target', 'contributor'),
])
if 'content' in self.feature_types:
languages = load_language_features(self.data_path, bigraph.targets)
descriptions = load_description_features(self.data_path, bigraph.targets)
features.extend([
(languages, 'target', 'language'),
(descriptions, 'target', 'description'),
])
if 'relation' in self.feature_types:
followers, followees = load_follow_features(self.data_path, bigraph.sources)
features.extend([
(followers, 'source', 'follower'),
(followees, 'source', 'followee'),
])
gc.collect() # force garbage collection
extractors = [
ConstantFeature(self.adj),
EdgeAttributeFeature(self.adj, keys=self.weight_key),
]
type_map = {'source': 'user', 'target': 'repository'}
for feature, type_, name in features:
print 'Creating node similarity matrix from %s feature (%s count: %d, ' \
'dimension: %d)...' % (name, type_map[type_], feature.shape[0], feature.shape[1])
similarity = BigraphSimilarity(bigraph, feature, type_)
extractor = SimilarityFeature(self.adj, similarity)
extractors.append(extractor)
return CombinedFeature(self.adj, *extractors)
def _calculate_similarities(self):
_, followees = load_follow_features(self.data_path,
self.bigraph.sources)
self.similarity = BigraphSimilarity(self.bigraph, followees, 'source')
def _calculate_similarities(self):
features = self.bigraph.matrix.copy()
self.similarity = BigraphSimilarity(self.bigraph, features, 'source')
def _calculate_similarities(self):
features = self.bigraph.matrix.T.tocsr().copy()
self.similarity = BigraphSimilarity(self.bigraph, features, 'target')
def _calculate_similarities(self):
features = load_description_features(self.data_path,
self.bigraph.targets)
self.similarity = BigraphSimilarity(self.bigraph, features, 'target')