Software

MEX interface for SVMperf 3

Tue, 17 Nov 2009 10:54:17 +0100

This is the matlab interface of version 3.1 of Thorsten Joachims’ SVMperf. It was barely tested so be careful with it. Please let me know if you find any problem with this software.

Windows users: I’m sorry but I was unable to compile this stuff for you. If anyone is able to compile it, please, send me the binaries.

(obsolete) MEX interface for SVMperf

Sun, 25 Oct 2009 01:00:19 +0200

This version of Thorsten Joachims’ SVMperf can be used from Matlab. It was compiled and tested with several versions of Matlab. Two files are provided, svmperflearn.m and svmperfclassify.m, which, in turn, call respectively two library functions obtained after the compilation of the source code.

If you prefer the pre-compiled functions you can get versions for different architectures/op. systems.

Note for 64bit compilation: It seems that there was a problem with a cast when compiling in 64bit linux. The solution was to include a prototype of a function that was missing in the source code, so that there is no need for the cast anymore (source code is now properly updated); also, the file ‘compile.m’ should be modified to include the option ‘-largeArrayDims’ (thanks to Alexander Sorokin for these clues).

Note: On some (older) versions of Matlab it could be necessary to modify the mex_interface.h file, adding this two lines just after the #include <mex.h> statement

#define mwSize int
#define mwIndex int

since those data types are not defined.

@svmperf

Tue, 11 Mar 2008 10:30:48 +0100

This is a Spider object that integrates the Matlab version of SVM_perf into the Spider toolbox. In order to use this object you need to install previously the MEX version of SVM_perf.

@cv - multiclass cross-validation

Fri, 7 Mar 2008 00:53:55 +0100

This Spider object lets you make balanced cross-validations with multiclass data sets. Classes should be specified as a column of different numbers in the Y field of a data object. You can replace the original @cv object with this one or you can rename one of them (in that case, remember to also rename the constructor and the object name inside the constructor)

input vars (data.X) target (data.Y)
0.9878 0.9282 0.2791 1
0.7944 0.2175 0.0778 3
0.4969 0.5824 0.8084 2
0.7775 0.7169 0.3956 3
0.7964 0.6408 0.1047 1