Convert Decision Tree with default condition to PMML using sklearn2pmml

[NAYAN GUPTA]

Hi Villu, 

Hope you are safe, healthy and doing well .
I am using scikit-learn to train a decision tree  and am able to convert it into a PMML file using the sklearn2pmml library . 

However, in some of the requirements . I need to introduce a default condition in the PMML  along with the trained decision tree. 

For example I have explained here
https://stackoverflow.com/questions/63511829/convert-decision-tree-with-default-condition-to-pmml-using-sklearn2pmml

A high level logic is as follows

if  some_condition == True:

     score = x

     return(score)

else:

  score = Decision_Tree(x)

  return(score)

Is is possible to create such a node in PMML ? 
Has anyone in the group tried it out. 

Thanks
Nayan Gupta

[Villu Ruusmann]

Hi Nayan,

>
> For example I have explained here
> https://stackoverflow.com/questions/63511829/convert-decision-tree-with-default-condition-to-pmml-using-sklearn2pmml
>

It's easy - see the sklearn2pmml.ensemble.SelectFirstClassifier estimator type:
https://github.com/jpmml/sklearn2pmml/blob/0.61.0/sklearn2pmml/ensemble/__init__.py#L181-L199

>
> if some_condition == True:
> score = x
> return(score)
> else:
> score = Decision_Tree(x)
> return(score)
>

As the name suggests, SelectFirstClassifier returns the prediction of
the first child estimator whose "controlling predicate" evaluates to
True. For constant prediction you could use a
sklearn.dummy.DummyClassifier estimator.

Something like this:

classifier = SelectFirstClassifier([
("condition", "X['A1'] == 'Beta'", DummyClassifier(..)),
("default", "True", DecisionTreeClassifier(..))
])


Villu

[NAYAN GUPTA]

Hi Villu, Thanks for the prompt reply. Highly appreciate your support. 

I tried what you are suggesting and I am getting the following error in the last step of the code. 

Here is the error message : 

java.lang.IllegalArgumentException: The transformer object (Python class sklearn2pmml.ensemble.SelectFirstEstimator) is not a supported Transformer

Attaching the python code to reproduce the example 

Python version 3.5.x
sklearn2pmml version 0.61

Best Regards
Nayan Gupta,
Data Science , Robert Bosch Gmbh

[Villu Ruusmann]

Hi Nayan,

>
> java.lang.IllegalArgumentException: The transformer object
> (Python class sklearn2pmml.ensemble.SelectFirstEstimator)
> is not a supported Transformer
>

Class SelectFirstEstimator is an abstract base class that has two
concrete subclasses - SelectFirstClassifier and SelectFirstRegressor.

As you're dealing with a classification-type problem, then you should
be using SelectFirstClassifier.


Villu