svmtrain - unable to solve the optimization problem

Hi,

I am using svmtrain to discriminate between several pairs of data. Although svmtrain works as desired in one case (outputting a classifier object with ~70 % accuracy as verified by svmclassify), all other cases seem to fail. My feature vectors are 134 dimensions and I am using between 300 and 800 data points for each class. (Each class does not necessarily have the same number of data points). I have tried using the default kernel for svmtrain using the method

SVM = svmtrain(double(train{k}), group_train{k},'showplot',true);

In this case I get the error:

Unable to solve the optimization problem: Maximum number of iterations exceeded; increase options.MaxIter. To continue solving the problem with the current solution as the starting point, set x0 = x before calling quadprog.

I have also tried extending the number of iterations and specifying a kernel using the call:

options = optimset('maxiter',1000,'largescale','on');
SVM = svmtrain(double(train{k}),group_train{k},'Kernel_Function','mlp','Method','QP',...
   'quadprog_opts',options);

In this case, I get the error:

Unable to solve the optimization problem: Exiting: the solution is unbounded and at infinity; the constraints are not restrictive enough.

In the case that did work, I have 338 data points from the first class and 476 data points from the second class. As examples, in three of the cases that don't work, I have 828, 573, and 333 data points in the second class, while the first class remains the same and has 338 data points. Neither method call seems to work.

Could you please help me? Any advice you could give me would be useful. I have been trying to solve this problem for a week and have had no luck. I am using MATLAB 7.9.0 R2009B on a virtual machine Windows XP with a 1 GHz processor and 2 GB RAM.

Thank you so much! -Vivek