Support for `MultiOutputRegressor` estimator type about sklearn2pmml

[shlomibabluki]

RandomForestRegressor has a built-in support for multiple outputs (no need to use MultiOutputRegressor).

I think it is more important to add support for that case, before implementing the MultiOutputRegressor meta-regressor.

[NicoleLee918]

Here is my environment:

print(sklearn.version)
0.18.2
print(sklearn.externals.joblib.version)
0.10.3
print(sklearn_pandas.version)
1.5.0
print(sklearn2pmml.version)
0.20.4
Python 2.7.13

[vruusmann]

Exception in thread "main" java.lang.IllegalArgumentException
at sklearn2pmml.PMMLPipeline.encodePMML(PMMLPipeline.java:74)

This exception means that PMMLPipeline does not recognize the enclosed estimator object.

Class MultiOutputRegressor is currently not supported (for a list of supported estimator and transformer classes, see the README file of the JPMML-SkLearn project). According to the Scikit-Learn documentation, it can be implemented fairly easily, because "under the hood" it is just constructing and fitting multiple copies of the argument estimator type.

Is there any plan to support this function?

Sure. But the timeline is unclear, as I have more important things to do in the near future.