Error using segminopt>segminoptmpl No convergence achieved within maximum number of iterations

Hi, I have problem to using SVM Classifier when i extracted the features from train data (13393 of images) and from test data (5740 of images) i use these code:

model_svm = svmtrain(Feature_Train,trainingLabels,'ShowPlot',true);
predictedLabels = svmclassify(model_svm, Feature_Test,'showplot',true);

Where the Feature_Train is a matrix of 13393x4096 and Feature_Test is a matrix of 5740x4096 But this error show:

Error using segminopt>segminoptmpl (line 198)
No convergence achieved within maximum number of iterations 
Error in segminopt (line 81)
[alphas offset] = segminoptImpl(data, targetLabels,....
Error in svmtrain (line 503)
[alpha bias] = segminopt(training, groupIndex,....

What can I do to fix the error? Please advise me

Answer 1

[Walter Roberson]

Unfortunately there is no option with svmtrain() to alter the number of iterations it will try.

You could experiment with using the other "method" choices, or the ktt related options.

You could also experiment with using the newer https://www.mathworks.com/help/stats/fitcsvm.html that is replacing svmtrain()

[Tripoli Settou]

You mean that there is no solution with svmtrain and must to replace it with newer methods?? When i search i found people said the solution is must increase the maximum number of iterations

options.MaxIter = 100000;
my_svm_struct = svmtrain((Training, Group, 'Options', options);

what's your opinion with this solution?

[Walter Roberson]

Ah, it looks like that should work too.

[Tripoli Settou]

Ok thanks, but how can i know the amount of increase this number i mean it's 15000 by default, So can i know how increase it?

[Walter Roberson]

I do not think I have ever seen any formula to estimate the number of required iterations. So just increase it and if the problem keeps happening, increase it again.

[Tripoli Settou]

Ok thank you