Institute of Mathematics and University of Stuttgart
print

liquidSVM for Matlab

liquidSVM for MATLAB

Welcome to the MATLAB bindings for liquidSVM.

This is a preview version of the new MATLAB bindings to liquidSVM, stay tuned for updates. On Windows there is a heavy Bug at the moment that renders it unusable.

Both liquidSVM and these bindings are provided under the AGPL 3.0 license.

Installation

Usage

% load some data sets with train/test split from http://www.isa.uni-stuttgart.de/liquidData/
banana = liquidData('banana-bc');  % binary labels
banana_mc = liquidData('banana-mc');  % labels with four unique values
reg = liquidData('reg-1d');  % real labels

%% Least Squares Regression
model = svm_ls(reg.train,'DISPLAY','1');
[result, err] = model.test(reg.test);
result = model.predict(reg.testFeatures);

%% Mutli-Class classification
model = svm_mc(banana_mc.train,'DISPLAY','1','folds','3');
[result, err] = model.test(banana_mc.test);

%% Quantile Regression here for the 20%, 50%, and 80% quantiles
model = svm_qt(reg.trainFeatures, reg.trainLabel,[0.2,0.5,0.8],'DISPLAY','1');
[quantiles, err] = model.test(reg.testFeatures,reg.testLabel);
plot(reg.testFeatures, reg.testLabel, '.', reg.testFeatures, quantiles(:,1),'.',...
    reg.testFeatures, quantiles(:,2),'.',reg.testFeatures, quantiles(:,3),'.')

% now quantiles has three columns corresponding to the three requested quantiles

%% Expectile Regression here for the 20% and 50% expectiles
model = svm_ex(reg.trainFeatures, reg.trainLabel,[.05,.5],'DISPLAY','1');
[expectiles, err] = model.test(reg.testFeatures,reg.testLabel);
plot(reg.testFeatures, reg.testLabel, '.', reg.testFeatures, expectiles(:,1),'.',...
    reg.testFeatures, expectiles(:,2),'.')

%% Receiver Operating Characteristic curve
model = svm_roc(banana.trainFeatures, banana.trainLabel,6,'DISPLAY','1');
[result, err] = model.test(banana.test);

%% Neyman-Pearson lemma
model = svm_npl(banana.trainFeatures, banana.trainLabel, 1,'DISPLAY','1');
[result, err] = model.test(banana.test);

%% Write a solution (after train and select have been performed)
model = svm_ls(reg.train,'DISPLAY','1');
save myModelFile model
clear model

%% read a solution from file
load myModelFile model
[result, err] = model.test(reg.test);

The meaning of the configurations in the constructor is described in the next chapter.

NOTE: MATLAB does not respect flushing of print methods, hence setting display to 1 does not help in monitoring progress during execution because the output only shows at the end of the computation.

NOTE: On macOS if you use MATLAB 2016a and Xcode 8 you have to make the new version available to MATLAB by changing /Applications/MATLAB_R2015b.app/bin/maci64/mexopts/clang_maci64.xml to also include MacOSX10.12.sdk on two occasions - similar details (for other versions) can be found int https://de.mathworks.com/matlabcentral/answers/243868-mex-can-t-find-compiler-after-xcode-7-update-r2015b. Remark that this change needs admin privileges.

Octave

Since Octave 4.0.x the classdef type of object-orientation is (experimentally) implemented so liquidSVM can be used there as well. Unzip the file http://www.isa.uni-stuttgart.de/software/matlab/liquidSVM-octave.zip change into a directory, start octave and issue:

makeliquidSVM native

If this works you can use demo_svm etc. as above.