Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
Kernel PCA unexpected results #270
Reported by marcus on 20 Apr 43350065 14:46 UTC
I've tested the Kernel PCA code and got some unexpected results.
there is no covariance on which to perform eigendecomposition explicitly as we would in linear PCA.
I've written some code which provides the expected results. Maybe someone can explain if I made a wrong assumption.
I've attached the data set and the plots of the results.
Thanks and regards,
Commented by rcurtin on 4 Jan 43354293 18:37 UTC
That's not how you transpose data...
Intuitively it seems like your implementation is correct, but I need to take a look at the test in kernel_pca_test.cpp and see if it is also flawed. If it is, I will devise a new test, and I suspect that your implementation will pass the test and the existing one will not -- in which case, I'll put your implementation in place.
Thanks for reporting this.
Commented by rcurtin on 23 Dec 43359018 22:35 UTC
I took your implementation and adapted it into kernel_pca_impl.hpp, so unless you ask me to do otherwise I've added your name to the authors list in that file and core.hpp and on the website.
Thanks again for pointing this out.