[
https://issues.apache.org/jira/browse/ARROW-18137?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17622859#comment-17622859
]
Jacek Pliszka commented on ARROW-18137:
---------------------------------------
actually this works fine
{code:python}
t.group_by(['keys', 'values']).aggregate()
{code}
> Allow passing no aggregations to TableGroupBy.aggregate
> -------------------------------------------------------
>
> Key: ARROW-18137
> URL: https://issues.apache.org/jira/browse/ARROW-18137
> Project: Apache Arrow
> Issue Type: New Feature
> Components: C++, Python
> Affects Versions: 9.0.0
> Reporter: Jacek Pliszka
> Priority: Minor
>
> If we could allow TableGroupBy.aggregate to accept no aggregation functions
> then it would behave like pandas drop_duplicates.
> {code:python}
> t.group_by(['keys', 'values']).aggregate()
> {code}
> I did some naive benchmarks and looks like it should be 30% faster than
> converting to pandas and deduplicating. This was my naive test:
> {code:python}
> t.append_column('i', pa.array([1]*len(t),pa.int64())).group_by(['keys',
> 'values']).aggregate([("i", "max")]).drop(['i_max'])
> {code}
> And on small 5M table it took 245ms while 359ms for
> t.to_pandas().drop_duplicates()
> Actual aggregation without adding dummy column should be even faster still
> will allow drop_duplicates functionality until better implementation arrives
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
