Using Java inside Matlab

John Kitchin

Matlab is pretty good, but it doesn't do everything, and you might not always have the toolbox that contains the function you want. There might even be an excellent open-source software library that can't be included in Matlab, but can be used within Matlab.

Matlab has excellent support for using Java with minimal effort. A repository of numerical libraries can be found at http://math.nist.gov/javanumerics/. In this example we install the commons-math Java library, and show an example of using it.

Contents

Here are a few things the commons libary can do:

  Computing means, variances and other summary statistics for a list of numbers
  Fitting a line to a set of data points using linear regression
  Finding a smooth curve that passes through a collection of points (interpolation)
  Fitting a parametric model to a set of measurements using least-squares methods
  Solving equations involving real-valued functions (i.e. root-finding)
  Solving systems of linear equations
  Solving Ordinary Differential Equations
  Minimizing multi-dimensional functions
  Generating random numbers with more restrictions (e.g distribution, range) than what is possible using the JDK
  Generating random samples and/or datasets that are "like" the data in an input file
  Performing statistical significance tests
  Miscellaneous mathematical functions such as factorials, binomial coefficients and "special functions" (e.g. gamma, beta functions)

Many of them duplicate features in Matlab, but they may implement different algorithms, or more convenient syntax.

clear all; clc; close all

download and install

These two commands only need to be run once, so comment them out after you run them.

%urlwrite('http://mirror.metrocast.net/apache/commons/math/binaries/commons-math-2.2.zip','commons-math-2.2.zip')
%unzip('commons-math-2.2.zip')

We have to add the jar file to the javaclasspath so we can use it.

javaaddpath([pwd '/commons-math-2.2/commons-math-2.2.jar'])

linear regression with Matlab

Note that the regress command is part of the Statistics toolbox, which is great if you have it! We show here how to do it, so we also have something to compare the Java result to.

x = [1 2 3 4 5]';
y = [3 5 7 14 11]';

[b bint]= regress(y,[x.^0 x])

b =

    0.5000
    2.5000


bint =

   -7.5615    8.5615
    0.0694    4.9306

Now the Java way

Documentation Examples

The first thing we do is import the Java library. the * at the end indicates that all classes should be imported.

import org.apache.commons.math.stat.regression.*

Create a class instance.

The SimpleRegression class has several methods that we can call later to perform analysis

regression = SimpleRegression();
methods(regression)

Methods for class org.apache.commons.math.stat.regression.SimpleRegression:

SimpleRegression            getSlope                    
addData                     getSlopeConfidenceInterval  
clear                       getSlopeStdErr              
equals                      getSumSquaredErrors         
getClass                    getTotalSumSquares          
getIntercept                hashCode                    
getInterceptStdErr          notify                      
getMeanSquareError          notifyAll                   
getN                        predict                     
getR                        toString                    
getRSquare                  wait                        
getRegressionSumSquares     
getSignificance             

add the data

regression.addData([x y]);

get the slope and some statistics of the slope

m = regression.getSlope()
mstd = regression.getSlopeStdErr()
mint = regression.getSlopeConfidenceInterval(); % 95% by default
[m - mint  m + mint] % this is the 95% interval

m =

    2.5000


mstd =

    0.7638


ans =

    0.0694    4.9306

Note the confidence interval is greater than +- 2*sigma. This is because of the student-t table multiplier on the standard error due to the small number

get the intercept and some statistics

note there is not getInterceptConfidenceInterval method for some reason!

b = regression.getIntercept()
bstd = regression.getInterceptStdErr()

sprintf('R^2 = %1.3f', regression.getRSquare())

b =

    0.5000


bstd =

    2.5331


ans =

R^2 = 0.781

Summary

It takes more lines of code to get the Java version of linear regression, but you have finer control over what data you get, and what order you get it in. You still have to read the Java documentation to learn the command syntax, and debugging can be somewhat more difficult than straight Matlab. However, the integration of Java into Matlab opens some interesting possibilities to extend Matlab's functionality.

% categories: Miscellaneous, data analysis

Published with MATLAB® 7.12