[GitHub] [spark] EnricoMi commented on pull request #37360: [SPARK-39931][PYTHON][WIP] Improve applyInPandas performance for very small groups

[GitBox <gi...@apache.org>]

EnricoMi commented on PR #37360:
URL: https://github.com/apache/spark/pull/37360#issuecomment-1202140475

   > Hm, the general idea might be fine but I think the implementation is the problem. For example, the current design is that the user defined `function` always takes one group for `pdf`. To keep this behaviour, you should send the multiple groups into one, and apply the same function multiple times for each group.
   
   This is happening here: https://github.com/apache/spark/pull/37360/files#diff-5862151bb5e9fe7a6b2d1978301c235504dcc6c1bbbd1f9745a204a3ba93146eR218-R229
   The user function gets decorated to take the batch of groups, group by key and apply the actual user function on each group. This is the trivial part.
   
   I need guidance on how to prepend the `InternalRow` key (as a single struct column) to the `InternalRow` value: https://github.com/apache/spark/pull/37360/files#diff-d153e10db7aa6557eb995300730b7f2b2d437fa5659dfeaa611800b49a09da9dR45-R49
   The difficult part is the deduplicated unsafe projection: https://github.com/apache/spark/pull/37360/files#diff-4d4a9c23cb4c92c1f60def46451bc5666ed466922d9b56bbd1accc06aefee4e2R82-R87


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org