import pandas as pd
from sklearn.decomposition import NMF, LatentDirichletAllocation, TruncatedSVD
from sklearn.feature_extraction.text import CountVectorizer
data = pd.read_csv('dataset.tsv.txt', sep='\t')
data.columns = ['a','b']
data = data[data.b==1]
data = data.a.values
vectorizer = CountVectorizer(min_df=5, max_df=0.9,
                             stop_words='english', lowercase=True,
                             token_pattern='[a-zA-Z\-][a-zA-Z\-]{2,}')

data_vectorized = vectorizer.fit_transform(data)
nmf_model = NMF(n_components=5000)
nmf_Z = nmf_model.fit_transform(data_vectorized)
print(nmf_Z.shape)  # (NO_DOCUMENTS, NO_TOPICS)