def DBSCAN_then_BERT_agglomerative(corpus, k=5):
cluster_assignment = cosine_with_DBSCAN(corpus, output="list")
clusters = cluster_by_index(cluster_assignment)
outliers = clusters[-1]
cluster_sentences = get_cluster_dict(cluster_assignment, corpus)
outlier_sentences = cluster_sentences[-1]
outliers_embeddings = embed_corpus(outlier_sentences)
clustering_model = AgglomerativeClustering(n_clusters=k)
clustering_model.fit(outliers_embeddings)
outlier_assignment = clustering_model.labels_
outlier_clusters = {}
for i in range(k):
outlier_clusters[i] = []
for i in range(len(outlier_assignment)):
cluster = outlier_assignment[i]
index = outliers[i]
if index == 1:
print(i)
outlier_clusters[cluster].append(index)
outlier_clusters_re_indexed = re_index_clusters(outlier_clusters,
max(cluster_assignment))
return [clusters, outlier_clusters_re_indexed]
def DBSCAN_then_BERT_SVM(corpus):
cluster_assignment = cosine_with_DBSCAN(corpus, output="list")
corpus_embeddings = embed_corpus(corpus)
original_clusters = cluster_by_index(cluster_assignment)
non_outlier_assignments = []
non_outlier_embeddings = []
for i in range(len(cluster_assignment)):
if cluster_assignment[i] != -1:
non_outlier_assignments.append(cluster_assignment[i])
non_outlier_embeddings.append(corpus_embeddings[i])
clf = svm.SVC()
clf.fit(non_outlier_embeddings, non_outlier_assignments)
max_cluster = max(cluster_assignment)
outlier_classifications = {}
for i in range(-1, max_cluster + 1):
outlier_classifications[i] = []
for i in range(len(cluster_assignment)):
if cluster_assignment[i] == -1:
cluster = clf.predict([corpus_embeddings[i]])[0]
outlier_classifications[cluster].append(i)
return [original_clusters, outlier_classifications]
def BERT_with_agglomerative(corpus, k=10):
corpus_embeddings = embed_corpus(corpus)
clustering_model = AgglomerativeClustering(n_clusters=k)
clustering_model.fit(corpus_embeddings)
cluster_assignment = clustering_model.labels_
return [cluster_by_index(cluster_assignment)]
def BERT_with_kmeans(corpus, k=10):
corpus_embeddings = embed_corpus(corpus)
clustering_model = KMeans(n_clusters=k)
clustering_model.fit(corpus_embeddings)
cluster_assignment = clustering_model.labels_
return [cluster_by_index(cluster_assignment)]
def DBSCAN_then_BERT_KNN(corpus, k=5):
cluster_assignment = cosine_with_DBSCAN(corpus, output="list")
clusters = cluster_by_index(cluster_assignment)
outliers = clusters[-1]
corpus_embeddings = embed_corpus(corpus)
outlier_classifications = classify_outliers(cluster_assignment,
corpus_embeddings, outliers, k)
return [clusters, outlier_classifications]
