Context See description for [unvectorized full text dataset](https://www.kaggle.com/jeffkao/proc_17_108_unique_comments_text_dupe_count). Post clustering, the same dataset. level_0 is the manually picked out clusters; level_1 is the HDBSCAN clusters. Also left in the text hash this time. Looking back through the data, I am still amazed by what a great job HDBCSAN did at breaking everything up (down?). Check out [Leland's project](https://hdbscan.readthedocs.io/en/latest/) and try it for yourself.