• Do you think all of the top 500 word tokens contain valuable information? If not, can you impose another list of stop words?
  • Can you use stemming instead of lemmatization to process the newsgroups data?
  • Can you increase max_features in CountVectorizer from 500 to 5000 and see how the t-SNE visualization will be affected?
  • Try visualizing documents from six topics (similar or dissimilar) and tweak parameters so that the formed clusters look reasonable.

Get Python Machine Learning By Example - Second Edition now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.