def __init__(self, entity_pairs, eps=0.5, min_samples=5, metric='euclidean', algorithm='auto'):
self.vm = VectorModel()
self.cluster = DBSCAN(eps=eps, min_samples=min_samples, metric=metric, algorithm=algorithm)
self.entity_pairs = entity_pairs
self.X, self.relation_index = self.get_relation_vecs()
print self.X.shape
self.cluster.fit(self.X)
self.labels = self.cluster.labels_
class RelationCluster():
'''
def __init__(self, entity_set_list, eps=0.7, min_samples=3, metric='euclidean'):
self.vm = VectorModel()
self.cluster = DBSCAN(eps=eps, min_samples=min_samples, metric=metric)
self.entity_set_list = entity_set_list
self.X, self.relation_index = self.get_relation_vecs()
self.cluster.fit(self.X)
self.labels = self.cluster.labels_
'''
def __init__(self, entity_pairs, eps=0.5, min_samples=5, metric='euclidean', algorithm='auto'):
self.vm = VectorModel()
self.cluster = DBSCAN(eps=eps, min_samples=min_samples, metric=metric, algorithm=algorithm)
self.entity_pairs = entity_pairs
self.X, self.relation_index = self.get_relation_vecs()
print self.X.shape
self.cluster.fit(self.X)
self.labels = self.cluster.labels_
def get_relation_vec(self, pair):
# Gets the difference between the word vector for w1 and the word vector for w2
w1, w2 = pair
return self.vm.vector(w1) - self.vm.vector(w2)
'''
def get_relation_vecs(self):
# Gets the relation vectors between every pair in the set
relation_index = {}
relation_list = []
index = 0
for entity_set in self.entity_set_list:
for entity1 in entity_set:
for entity2 in entity_set:
if entity1 == entity2:
continue
if (entity1, entity2) in relation_index:
# the oredered pair has already been seen
continue
relation_index[(entity1, entity2)] = index
relation_index[index] = (entity1, entity2)
relation_list.append(self.get_relation_vec(entity1, entity2))
index += 1
X = np.array(relation_list)
return X, relation_index
'''
def get_relation_vecs(self):
# Gets the relation vectors between every pair in the set
relation_index = {}
relation_list = []
index = 0
for pair in self.entity_pairs:
relation_index[pair] = index
relation_index[index] = pair
relation_list.append(self.get_relation_vec(pair))
index += 1
X = np.array(relation_list)
return X, relation_index
def get_num_clusters(self):
# Returns the number of clusters found
# The set of relation vectors that do not belong in any cluster
# does not count as a cluster
return len(set(self.labels)) - (1 if -1 in self.labels else 0)
def get_unique_labels(self):
# Returns the set of unique labels, or the index of each cluster
return set(self.labels)
def get_label(self, pair):
# Returns the label of the cluster that a pair's relation vector belongs in
return self.labels[self.relation_index[pair]]
def is_core(self, pair):
# Returns whether a pair's relation vector is a core component
return self.relation_index[pair] in self.cluster.core_sample_indices_
def is_clustered(self, pair):
# Returns whether a pair's relation vector is part of a cluster
return self.labels[self.relation_index[pair]] != -1
def get_cluster_sizes(self):
# Returns a dictionary mapping cluster labels to the cluster sizes
counts = {}
label_list = list(self.labels)
for label in self.get_unique_labels():
counts[label] = label_list.count(label)
return counts
def get_vector_clusters(self):
# Returns a tuple of two dictionaries
# The first dictionary maps labels to lists of relation vectors
# The second dictionary maps labels to the indices of the relation vectors that are
# in the cluster of that label
clusters = collections.defaultdict(list)
index_clusters = collections.defaultdict(list)
core_indices = self.cluster.core_sample_indices_
for i in xrange(len(self.labels)):
if self.labels[i] == -1:
continue
clusters[self.labels[i]].append(self.X[i])
index_clusters[self.labels[i]].append(i)
'''
for label in self.get_unique_labels():
for index in core_indices:
if self.labels[index] == label:
index_clusters[label].append(index)
'''
print len([i for i in self.labels if i != -1])
print index_clusters
return clusters, index_clusters
def get_vector_cluster_means(self):
clusters, index_clusters = self.get_vector_clusters()
means = {}
for label in self.get_unique_labels():
cluster = np.array(clusters[label])
mu = cluster.mean(axis=0)
means[label] = mu
return means
def get_vector_cluster_variances(self):
# finds the variance from the mean
pass
def print_clusters(self):
if len(self.cluster.core_sample_indices_) == 0:
print "NO CLUSTERS :("
return
clusters, index_clusters = self.get_vector_clusters()
cluster_sizes = self.get_cluster_sizes()
#cluster_vector_means = self.get_vector_cluster_means()
for label in self.get_unique_labels():
if label == -1:
continue
print "Cluster: %d" % label
print "cluster size: %d" % cluster_sizes[label]
print "indices in cluster: ", index_clusters[label]
#mean = cluster_vector_means[label]
for index in index_clusters[label]:
print '\t' + str(self.relation_index[index])
'''