Github user Yunni commented on the issue:
https://github.com/apache/spark/pull/15148
Hi @MLnick @jkbradley
Thanks for the code review. I made some changes based on your comments.
- I agree it's better to align the input types to vector in internal
implementation. Sparse vectors can also better represent a set than
Array[Double] does. I made a new commit for this change.
- For NN search, my concern is if we do `approxNearestNeighbors` over a
DataFrame, the results will be very similar to `approxSimilarityJoin` while the
performance is worse (need an extra groupby and sort within each group). For
multiple-probing NN search, I think the performance will be even worse.
Our use case is mainly using similarity join to find fraud trips. I think I
can change the NN-search to only single-probing NN search of dataframe if you
think it's fine. What do you think?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
