Github user icexelloss commented on a diff in the pull request:
https://github.com/apache/spark/pull/20211#discussion_r160864143
--- Diff: python/pyspark/sql/group.py ---
@@ -233,6 +233,27 @@ def apply(self, udf):
| 2| 1.1094003924504583|
+---+-------------------+
+ Notes on grouping column:
--- End diff --
Yeah. To be honest I don't think there is behavior that is both simple and
works well with all use cases. It's probably a matter of leaning towards
simpler behavior that doesn't work well in some cases or towards somewhat
"magic" behavior. I don't think there is an obvious answer here.
Another option is to always prepend grouping columns, if users want to
return grouping columns in the UDF output, they can do a `drop` after `groupby
apply`
```
pandas_udf('id int, v double', GROUP_MAP)
def foo(pdf):
return pdf.assign(v=pdf.v+1)
df.groupby('id').apply(foo).drop(df.id)
```
I don't think it's too annoying to add a drop after apply and it works well
with the linear regression case. This is also a pretty straight forward non
magical behavior.
What do you all think?
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
