bwaldvogel / liblinear-weka Goto Github PK

Thanks for putting together the code for liblinear in weka. Is there any benchmarks (performance and speed) to compare different classifiers in weka, including liblinear-weka?

Thanks!

LibLINEAR 1.9.8

weka.classifiers.functions.LibLINEAR -S 1 -C 1.0 -E 0.001 -B 1.0 -L 0.1 -I 1000

In the Weka Explorer when evaluating a liblinear classifier with default parameters, and using as data an arff file containing sparse instances I received the following exception in the console:

java.lang.ArrayIndexOutOfBoundsException: 44458
weka.classifiers.functions.LibLINEAR.toString(LibLINEAR.java:1041)
weka.gui.explorer.ClassifierPanel$18.run(ClassifierPanel.java:1406)

Note that I have evaluated the same classifier with the same arff file using the JAVA api and everthing is working fine.

I'm attaching the arff file, it has 1416 Instances, 44457 attributes.

data.txt

The Weka interface to LibLINEAR does not permit the user to choose any of the regression SVMs.

The problem appears to be that the drop down menu for selecting the SVM type is returning the wrong value for the regression SVM types. It should be returning 11, 12, and 13. Instead it is returning 8, 9, and 10. It appears that the code is returning the ordinal value of the menu item rather than the SolverType id.

I suspect there is another problem. Currently the Weka code appears to always return the default capabilities for libLinear, i.e., NOMINAL_CLASS. I believe this should change when one of the regression SVMs is chosen. Otherwise Weka will not allow LibLINEAR to run, thinking that it doesn't have the capability to handle a regression problem.