how can i solve this errore in svmtrain?

i have some problem with svmtrain. i extract some pixel form different frame and base on their value in 3 channels train with svm i have about 20000 pixel and when i use svmtrain sometimes i get this error but when my pixel is low for example 100 pixel i dont get this error

??? Error using ==> svmtrain at 197
TRAINING data must not contain missing values

i use matlabR2009 7.8

do you have any suggestion ?please help me?

options = optimset('maxiter',1000);
svmstruct_linear=svmtrain(train_y,target_y);
svmstruct_quadratic=svmtrain(train_y,target_y,'Kernel_Function','quadratic','Method','QP',...
            'quadprog_opts',options);

[Walter Roberson]

Stray thought: could it be that when you use a small number of pixels, you do not happen to have any pixels with value 0, but that you get a 0 when you use the full set of pixels?

I do not know if 0 is special to svmtrain, but I could imagine it being special in the class information (i.e., target_y in your case)

Answer 1

[Alan Weiss]

Did you check whether there are any NaN values in train_y or target_y? For example, examine

 sum(isnan(train_y) + isinf(train_y))

Alan Weiss

MATLAB mathematical toolbox documentation