MATLAB R2018B documents the commands oobPermute​dPredictor​Importance and predictorImportance but these commands are not recognized by MATLAB, neither at the command nor in m-code.

MATLAB R2018B documents the commands oobPermutedPredictorImportance and predictorImportance but these commands are not recognized by MATLAB. This is true for both the command line and when running m-code.

I have developed an tree ensemble model and I need to determine the predictor variables importances.

Answer 1

[Steven Lord]

From the documentation page for the oobPermutedPredictorImportance function:

Mdl must be a RegressionBaggedEnsemble model object.

It is a method of that class, not a plain old function. It will only work when its first input is one of those objects. You can't, for example, say oobPermutedPredictorImportance(magic(4)) because magic(4) returns a double array, not a RegressionBaggedEnsemble object.

The same general answer holds for predictorImportance, though it may be a different class or a different set of classes that define this method.

[Christopher Stokely]

Do I have to use fitrensemble to generate the regression mode?

I am interested in classification though. Does fitcensemble support those methods?

Are there no predictor importance methods for classification ensemble models or decision trees? If so, is the classificationLearner app sufficient to build a model that can be explored with predictor importance methods?