matlab - How to do regression using svmtrain() function in libsvm (what settings)? what is the label matrix form?

Question

Libsvm is a small library for support vector machines written in C. here is the link to the source of libsvm I've used ( https://github.com/cjlin1/libsvm ). Can someone correct the options in the svmtrain() function or label matrix in the following code to perform regression successfully, since the output says optimization finished with #0 iteration.

The goal of the code before svmtrain() function is to arrange water pixels of an underwater color image (some rect in image) into an (m x 3) matrix as data. The label matrix is an ( m x 1 ) vector preset to 1;{ here is the link to the underwater image, equation of red channel compensation }

the output:

optimization finished, #iter = 0
nu = 0.000000
obj = 0.000000, rho = -1.000000
nSV = 0, nBSV = 0

here is the code:

    ##to load image package
    pkg load image
    ## add the path to the matlab folder inside the LIBSVM package source
    addpath('./libsvm-master/libsvm-master/matlab')
    ##load libsvm 
    pkg load statistics
    ##load an under water rgb image
    im = imread('9554.png');
    im=double(im);
    startrow=1;
    startcol=1;
    endrow=100;
    endcol=100;
    ## save rgb pixels in m x 3 form
    data=zeros(endrow*endcol,3,'double');
    for i=startrow:endrow
      for j=startcol:endcol
        data((i-1)*endrow+j,1:3)=im(i,j,1:3);
      end
    end
    ## m x 1 label matrix, all preset to 1
    label_data=ones(size(data,1),1,'double');
    model=svmtrain(label_data,data,'-s 3 -c 1');

question from:https://stackoverflow.com/questions/65868286/how-to-do-regression-using-svmtrain-function-in-libsvm-what-settings-what-i

Answer 1

Thank you for providing the image and extra details.

I think the problem is simply that you're passing a 'labels' matrix which consists only of a single label. The algorithm detects that there is no data to train with, and stops.

If you make some of the labels into zeros, then the algorithm runs as expected.