Getting error after compile classifiers using decision tree and treeBagger to Java

I build two classifiers one use classificationTree and one uses treeBagger. Runing the .m file using both models in Matlab environment, I'm able to predict test data without problem. But after compile to Java .JAR library, I got error for both models. See below for my code to apply the model on test data and followed by error messages.

I'm running on Matlab R2011a. Here attached is the Matlab version info and a picture showing the toolbox in the JAR file created, which shows component of the stat toolbox is available. Also attach two models I created.

Please help troubleshoot why the compiled version doesn't work. Thanks.

 %#function ClassificationTree
 %#function TreeBagger
 model=modelInfo.model;
 switch lower(modelInfo.modelType)
           case 'decisiontree'  
                    pred = predict(model,dn);                    
                    modelFound=2;
           case 'treeBagger'
                    pred = predict(model,dn);
                    pred = str2num(cell2mat(pred));
  end

Decision Tree model gives the same error:

{??? Undefined function or method 'classregtreeEval' for input arguments of type 'cell'. Error in ==> classregtree.eval at 90

Error in ==> CompactClassificationTree>CompactClassificationTree.predict at 485

The treeBagger model gives similar error: {??? Undefined function or method 'classregtreeEval' for input arguments of type 'cell'.

Error in ==> classregtree.eval at 90

Error in ==> CompactTreeBagger.CompactTreeBagger>CompactTreeBagger.treeEval at 1153

Error in ==> CompactTreeBagger.CompactTreeBagger>CompactTreeBagger.predictAccum at 1399

Error in ==> CompactTreeBagger.CompactTreeBagger>CompactTreeBagger.predict at 264

Error in ==> TreeBagger.TreeBagger>TreeBagger.predict at 1461

Answer 1

[Ilya]

classregtreeEval is a mex function. I don't know what you need to do to use it in your jar library. But you can find a compiled mex for your platform in matlab/toolbox/stats/stats/@classregtree/private/

[LiJie Yu]

Thanks for the reply. I build the JAR in Windows platform, but need to deploy it to run in Hadoop Linux environment. So with this platform dependent .mex function it wouldn't work then. Is there a way to build classificationTree/bagged tree ensemble models without using the .mex function? Right now I'm using the ClassificationTree and TreeBagger from stats toolbox.

[Ilya]

I don't know if you are right about the cross-platform issue, but I am not an expert on jar's. No, you have to use this mex function to make predictions.