[
https://issues.apache.org/jira/browse/SPARK-57294?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
tonghuaroot updated SPARK-57294:
--------------------------------
Description:
{{DataFrame.combine}} is currently declared via {{_unsupported_function}}, so
calling it on pandas-on-Spark always raises {{PandasNotImplementedError}}, even
when {{compute.pandas_fallback}} is enabled.
This proposes supporting it through the existing pandas-fallback mechanism
({{compute.pandas_fallback}}), consistent with sibling DataFrame methods such
as {{asof}} and {{set_axis}}: when the option is enabled, the operation is
computed via pandas and returned as a pandas-on-Spark DataFrame. The result
preserves Spark-mappable dtypes (e.g. int64 -> bigint), so it round-trips
through the Spark type system.
was:pandas-on-Spark lists DataFrame.to_period as an unsupported function.
Like asfreq/asof/convert_dtypes/infer_objects/set_axis (SPARK-46926,
apache/spark#44965), it can be supported through the existing
compute.pandas_fallback path without a native distributed implementation. This
adds to_period to the fallback-enabled set with classic + Spark Connect parity
tests.
Summary: Support DataFrame.combine in fallback mode (was: Support
DataFrame.to_period in fallback mode)
> Support DataFrame.combine in fallback mode
> ------------------------------------------
>
> Key: SPARK-57294
> URL: https://issues.apache.org/jira/browse/SPARK-57294
> Project: Spark
> Issue Type: Improvement
> Components: Pandas API on Spark
> Affects Versions: 4.3.0
> Reporter: tonghuaroot
> Priority: Minor
>
> {{DataFrame.combine}} is currently declared via {{_unsupported_function}}, so
> calling it on pandas-on-Spark always raises {{PandasNotImplementedError}},
> even when {{compute.pandas_fallback}} is enabled.
> This proposes supporting it through the existing pandas-fallback mechanism
> ({{compute.pandas_fallback}}), consistent with sibling DataFrame methods such
> as {{asof}} and {{set_axis}}: when the option is enabled, the operation is
> computed via pandas and returned as a pandas-on-Spark DataFrame. The result
> preserves Spark-mappable dtypes (e.g. int64 -> bigint), so it round-trips
> through the Spark type system.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
