svmtrain compared to fitcsvm

In the previous depreciated "svmtrain" function, The function will stop and return "No convergence error" if it's couldn't train the svm successfully after certain number of iteration.

However, I notice in the newer "fitcsvm", it returns without any error message, even if no convergence reached. Is there is away to force "fitcsvm" to only return if svm successfully trained, otherwise through an error. Just like "svmtrain".

Answer 1

[Ilya]

Check the ConvergenceInfo.Converged property of the returned object to see if optimization converged.

This behavior of fitcsvm is an improvement over svmtrain. First, although optimization does not converge, the model may be still accurate and you can use it. Second, you can use the resume method of the returned object to continue optimization for more iterations. Throwing an error would strip you of both possibilities.