Support Vector Machine: SPEED-UP and make the computational of SVM FITCSVM & PREDICT more efficient

Dear all

I am using the build-in MATLAB svm training and classification to classify a binary class (i.e., class 0 OR 1). To avoid the classification by chance, the training and classification process was repeated for 1000 times.

However, when I run the program, it take a very long time to complete both the training and classification. To be exact, more than a days. In addition, at every iteration, MATLAB produce the following warning

Warning: Classes are perfectly separated. The optimal score-to-posterior
transformation is a step function. 
> In fitSVMPosterior (line 175)
  In classreg.learning.classif.CompactClassificationSVM/fitPosterior (line 378)
  In matlab_ask_helpSVM>EvalSVM (line 87)
  In matlab_ask_helpSVM>EvalEachCol (line 15)
  In matlab_ask_helpSVM (line 5)

With regard to the speed and warning above, I wonder if the code that I used can be optimized and make more compact. For readability of this thread I did not post the complete code here. However, the complete code can be found attached together in this thread.

I really appreciate if someone can advice me what changes I should make to make my code in par with good MATLAB coding practice.