I have a question:
How to choose the kernel in thekernel PCA method? As far as I know, in current applications of support vector machine, one of the key problems is to choose a proper kernel. It depends on the researcher's experience in a large extent. I have done some work in the application of SVM, and in some cases, the different kernel may cause different results.(Although according to Vapnik's theory, the kernel shouldn't have much effect on the result.) So I wonder if the kernel chosen in PCA would have effect on the result and if there are some methods that can help us to choose a good kernel function. Thanks.