Using deploytool to create Java package

Hello!

My intention is to use a Random Forest Ensemble, trained previously in Matlab, in a Java application. My Matlab function, which I compiled, using deploytool is:

   function [Probability]=PredictingTest (Predictor)

     load (['D:/Test/Tree.mat'], 'Tree')
     [~, prob]=predict(Tree , Predictor);
     Probability=prob(:,1);

   end

Where:

• Predictor input is an integer between -10 to 10.
• Probability output is a double between 0 to 1.
• Tree is a .mat file with a CompactTreeBagger object stored in it.

Next I deploy the jar files to my Eclipse project, and try to run this Java code:

/* Necessary package imports */
import com.mathworks.toolbox.javabuilder.*;
import PredictTest.*;

public class predict_test {
  
   static MWNumericArray rhs = null;   /* Stores input value */
   static PredictTest prediction;
   static Object[] result = null;    /* Stores the result */
  
   public static void main(String[] args) {
  
  try {
       prediction = new PredictTest();
       rhs=new MWNumericArray(5,MWClassID.DOUBLE);      
       result=prediction.PredictingTest(1, rhs);
       }
      catch (MWException e) {
      e.printStackTrace();
  }}}

Sadly what I get is this exception:

{Warning: Variable 'Tree' originally saved as a CompactTreeBagger cannot be instantiated as an object and will be read in as a uint32.}

> In PredictingTest at 3

{??? Undefined function or method 'predict' for input arguments of type 'uint32'.

So, as far as I understood, this means that Java can't use Matlab objects even by the methods, compiled from Matlab functions. It will be very nice if I could get some help on how can I overcome this. It is very important to me. Thank you all in advance!

P.S I'm surely not confined to a Matlab TreeBagger algorithm. If there is a Java package you are familiar with, that can do the work, it can be a nice solution too.

[Emin BAKIR]

I know the question is quite old, but recently I face with a similar example, just in case if it helps to someone else... If you create an empty object of the not found class just before loading it, then Matlab will be able to instantiate that class.

For the above code, the Matlab code should be something like the following

   function [Probability]=PredictingTest (Predictor)
     Tree  =CompactTreeBagger.empty;
     load (['D:/Test/Tree.mat'], 'Tree')
     [~, prob]=predict(Tree , Predictor);
     Probability=prob(:,1);

   end

I did not test the code, but it should work.

Answer 1

[Titus Edelhofer]

Hi,

try to add the CompactTreeBagger class definition to your project. Then you should be able to handle objects of this class within your project ...

Titus

[Artik Crazy]

Thanks, what am I supposed to do in order to define it?

Add the TreeBagger.m class definition while building with Builder JA?

[Titus Edelhofer]

Yes, just add the file to the lower part (something like "helper files") of deploytool.

Just by looking at the code, the compiler does not know it has to add the file ...

[Artik Crazy]

Thank you, it is actually working now :)

[LiJie Yu]

I run into the same problem. From my R2011a Matlab toolbox, I see two TreeBagger.m files: one with original .m code, another one under a subfolder "ja" and is in binary form. Which one shall I add as the helper file for deployment?

[Michael McDowell]

Thanks for this, I wasn't expecting to be required to explicitly reference the toolbox files.

[Steven Lord]

There's no reference to TreeBagger in the code. That is why you need to explicitly include the files in your project. When MATLAB Compiler tries to determine what it needs to include, it can't look inside the MAT-file (what if that MAT-file name was to be specified at runtime?) to determine that the MAT-file contains a TreeBagger object and realize that it should include the object definition in case you are loading that variable to use it.

[Emin BAKIR]

How do we add the class definition to our project. is it with import statement? I was able to solve a similar problem by creating an empty object of the class first, then loaded the .mat file. But I would like to learn how do we add class definition. Thank you.

Answer 2

[Joan Puig]

For a more robust deployment I would also suggest you change this line:

load (['D:/Test/Tree.mat'], 'Tree')

The reason is that people using this tool might not even have a D: drive. You could make the file name an input to your function.

[Artik Crazy]

Yeah, Thank you :)

I used it only to simplify the example.

There are a number of such trees, stored as .mat files, all created using different training sets. So the real function gets also a set of arguments that are used to choose the right tree to load and an input folder path as well.