Hyukjin Kwon created SPARK-61334:
------------------------------------
Summary: Make the setters of the native pyspark.ml.connect classes
set the value on the object and return it, as PySpark does
Key: SPARK-61334
URL: https://issues.apache.org/jira/browse/SPARK-61334
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
In the Python drop-in, the setters of the native pyspark.ml.connect classes
StandardScaler, MaxAbsScaler, StringIndexer, VectorAssembler,
LogisticRegression, RegressionEvaluator, BinaryClassificationEvaluator,
MulticlassClassificationEvaluator, Pipeline and CrossValidator returned a new
object and left the receiver unchanged. So a call whose result was not kept was
silently ignored: ev.setMetricName("r2"); ev.evaluate(df) returned the rmse,
MulticlassClassificationEvaluator().setMetricName("accuracy") gave f1,
Pipeline().setStages([...]) kept no stages, CrossValidator's
setNumFolds/setParallelism/setSeed were ignored, and
LogisticRegression().setLabelCol("y") then fit failed on the column label.
PySpark's setters (Params._set) set the value on the object and return the
object; the native KMeans, LinearSVC and LinearRegression already do. The fix
makes every native setter do so, with the argument name value, as PySpark's.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]