Hyukjin Kwon created SPARK-61705:
------------------------------------
Summary: Add PySpark's Pipeline and PipelineModel to the native ML
classes, of PySpark's persistence format
Key: SPARK-61705
URL: https://issues.apache.org/jira/browse/SPARK-61705
Project: Spark
Issue Type: Improvement
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
The native Pipeline of the Python client (pyspark.ml.connect.Pipeline) holds
stage names and does not fit. It should be PySpark's Pipeline and PipelineModel
of pyspark.ml.pipeline, natively:
- stages of any estimator or transformer;
- the fit of Pipeline._fit and the transform of PipelineModel._transform;
- PySpark's copy;
- the persistence of PipelineWriter and PipelineReader in PySpark's Python
format: the metadata of DefaultParamsWriter (a text file of one line at
<path>/metadata), and the stages at stages/<index>_<uid>. PySpark and this
client then load what each other saves.
The Rust API gets the metadata helpers of that format (save_metadata,
load_metadata, stage_path, handle_overwrite), for the validators and OneVsRest
to follow.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]