Support Vector Machine: SPEED-UP and make the computational of SVM FITCSVM & PREDICT more efficient
2 次查看(过去 30 天)
显示 更早的评论
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.
0 个评论
回答(0 个)
另请参阅
产品
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!