def decompose_and_cluster(tasks, word2vec, output_file, method='KMeans', option=10):
"""
You should pass parameter 'method', 'option' as follows
method = 'DBSCAN' or 'KMeans'
option = eps or n_clusters
"""
print 'Get task vector'
whole_vector = tasks_to_vectors(tasks, word2vec)
print 'Down dimension...'
pca = PCA(5)
d_vector = pca.fit_transform(whole_vector)
# print 'PCA Log Likelihood Score : ' + str(pca.score())
if method=='KMeans':
print 'Training K-means...'
cluster = KMeans(n_clusters=option, n_jobs=3)
else:
print 'Training DBSCAN ... '
cluster = DBSCAN(eps=option)
cluster.fit(d_vector)
labels = cluster.predict(d_vector)
pipe = Pipeline(steps=[
('w2v_200_to_5_PCA', pca), ('clustering', cluster)
])
if os.path.exists(output_file):
os.remove(output_file)
joblib.dump(pipe, output_file, compress=3)
print 'Complete dumping'
return pipe, labels
def decompose_and_cluster(tasks, word2vec, output_file, n_clusters):
print "Get task vector"
whole_vector = tasks_to_vectors(tasks, word2vec)
print "Down dimension..."
pca = PCA(5)
d_vector = pca.fit_transform(whole_vector)
print "Training K-means..."
kmeans = KMeans(n_clusters=n_clusters, n_jobs=3)
kmeans.fit(d_vector)
labels = kmeans.predict(d_vector)
pipe = Pipeline(steps=[("w2v_200_to_5_PCA", pca), ("Kmeans_1000", kmeans)])
if os.path.exists(output_file):
os.remove(output_file)
joblib.dump(pipe, output_file, compress=3)
print "Complete dumping"
return pipe, labels
