Bryan Cutler created SPARK-13625:
------------------------------------
Summary: PySpark-ML method to get list of params for an obj should
not check property attr
Key: SPARK-13625
URL: https://issues.apache.org/jira/browse/SPARK-13625
Project: Spark
Issue Type: Improvement
Components: ML, PySpark
Reporter: Bryan Cutler
In PySpark params.__init__.py, the method {{Param.params()}} to return a list
of Params belonging to that object should not check an attribute to be an
instance of {{Param}} if it is a class property (uses the {{@property}}
decorator). This causes the property to be invoked to 'get' the attribute, and
that can lead to an error, depending on the property. If an attribute is a
property it is not an ML {{Param}}, so no need to check it.
I came across this in working on SPARK-13430 while adding
{{LinearRegressionModel.summary}} as a property to give a training summary,
similar to the Scala API. It is possible that a training summary does not
exist and will then raise an exception if the {{summary}} property is invoked.
Calling {{getattr(self, x)}} will cause the property to be invoked if {{x}} is
a property. To fix this, just need to check if it a class property before
making the call to {{getattr()}}, which is fine because no instance of a
{{Param}} should be a property.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
