Hyukjin Kwon created SPARK-61752:
------------------------------------

             Summary: Keep the params of the native ML classes in PySpark's own 
param maps, of PySpark's own Params API
                 Key: SPARK-61752
                 URL: https://issues.apache.org/jira/browse/SPARK-61752
             Project: Spark
          Issue Type: Improvement
          Components: Connect
    Affects Versions: connect-rust-4.3.0
            Reporter: Hyukjin Kwon


The public classes of the native ML classes (pyspark.ml.connect.*, subclasses 
of PySpark's classes) keep their params in the native object, and their Params 
API is native code. PySpark's own Python code over a Params object's param maps 
does not apply to them: _paramMap and _defaultParamMap, _set, _setDefault, 
_copyValues, _resetUid, _input_kwargs, the order of extractParamMap, the 
TypeConverters' conversions, the messages of a call that does not bind, the 
Params of an old uid after an assignment of uid, and PySpark's own getters, 
setters, setParams and keyword_only constructors.

This makes the param maps PySpark's, in the instance's attributes: a public 
class takes PySpark's own functions of the Params API (and its getters, 
setters, constructors and copy) as they are, and makes the native object's 
params those of the maps (by name) before each native member that reads them (a 
fit, a transform, an evaluation, a save, a model member). The maps of an object 
that PySpark's constructor did not make (a fit's model, a load, a native copy) 
are made of the native object's values at their first read.




--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to