Why u use Gram matrix in desiging svm

[robosina]

why we need gram matrix for training data in following line

 M = M + 1/c*eye(ntrain);

and what is your reference for writing this code? can you share with us

[lbugnon]

Thanks for your comment. Well, the Gram matrix was in the original code. It is an old way of implementing kernel operation, I just implemented it in a matricial way instead of using a kernel function at every operation.

M=gaussKernel(xdata,xdata,gamma);
...
M = M + 1/c*eye(ntrain);

M has to be positive semidefinite for this optimization problem, and 1/c is the regularization term. However, this is a home-made code, at that time I couldn't find any standard way.
I highly recommend using standard C based libs to implement real-life classifiers (and preferably python ;) ) .

[robosina]

Thanks:))