[
https://issues.apache.org/jira/browse/SPARK-61708?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyukjin Kwon updated SPARK-61708:
---------------------------------
Fix Version/s: connect-rust-4.3.0
> Add PySpark's CrossValidator, TrainValidationSplit and their models to the
> native ML classes
> --------------------------------------------------------------------------------------------
>
> Key: SPARK-61708
> URL: https://issues.apache.org/jira/browse/SPARK-61708
> Project: Spark
> Issue Type: Improvement
> Components: Connect
> Affects Versions: connect-rust-4.3.0
> Reporter: Hyukjin Kwon
> Priority: Major
> Fix For: connect-rust-4.3.0
>
>
> Part of completing the native ML classes of pyspark.ml.connect to the full
> pyspark.ml 4.3.0 surface, after the Pipeline (SPARK-61705).
> The native CrossValidator held its numeric params only, and its fit raised
> NOT_IMPLEMENTED; there was no TrainValidationSplit. Add PySpark's validators
> of pyspark.ml.tuning, natively, as PySpark 4.3.0 has them over Spark Connect:
> CrossValidator, CrossValidatorModel, TrainValidationSplit and
> TrainValidationSplitModel.
> - fit is PySpark's _fit, on the client: the splits of PySpark's (rand(seed)
> folds, a fold column, a train ratio), cached as _cache_spark_dataset caches
> them; the fits of fitMultiple in a pool of `parallelism` threads;
> collectSubModels; numpy's metrics; and the best param map fitted on the whole
> dataset.
> - copy, and the models' attributes (bestModel, avgMetrics, stdMetrics,
> validationMetrics, subModels).
> - the persistence of PySpark's Python format (_ValidatorSharedReadWrite),
> which each of PySpark and this client loads of the other, with
> persistSubModels.
> - ParamGridBuilder is PySpark's own class.
> The core gets the splits, k_fold and train_validation_split.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]