Hyukjin Kwon created SPARK-61591:
------------------------------------
Summary: Output PySpark's columns from the native
LogisticRegression and take pandas DataFrames in the native ML fit and
transform, in the Python client
Key: SPARK-61591
URL: https://issues.apache.org/jira/browse/SPARK-61591
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
The native classes of pyspark.ml.connect in the Python client of
spark-connect-rust differ from PySpark 4.3.0's:
- The LogisticRegression model output the JVM's rawPrediction and probability
vectors and a double prediction, where PySpark's outputs a bigint prediction
and an array<double> probability (nullable, as its pandas UDF outputs), and no
rawPrediction. Its constructor lacked probabilityCol and tol.
- fit and transform did not take pandas DataFrames: PySpark's transform of a
pandas DataFrame returns a pandas DataFrame with the output columns, its
scalers fit pandas DataFrames, and its LogisticRegression raises
NotImplementedError for one.
- A transform whose output column exists raised the server's
IllegalArgumentException, where PySpark raises ValueError "Transformers does
not allow output column having the same name with existing columns."
- The scalers' output array<double> was not nullable and did not contain null,
where PySpark's is nullable and contains null.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]