I am performing kernel ridge regression on some really large datasets on my home computer which already is quite time consuming.. basically I have to regularize the solution by some parameter controlling the model complexity and I am adviced to find the optimal parameter by some form of efficient cross validation..I found that there is an analytical solution for computing the leave-one-out error residuals but either way I need to choose some regularization candidates to perform the CV on..And I have no idea how to choose them..I have limited computational power and it would be great to know how to choose the potential values, how to space them for a certain kernelmatrix (gaussian or polynomial)..Any advice?
Cheers
[link][4 comments]