I'm using "Kernlab" to apply the "Weighted Nadaraya Watson" by Kato (2012)
and Hall, Wolff, and Yao (1999).
I need to find this Gaussian Kernel in weights'calculation , where u=
(x-x0):
Kh(u) = h^(?1)*K(u/h).
I used:
rbf1 <- rbfdot(sigma = NULL)
but I have to find out "sigma" as the inverse width. I used ?sigest?
function but it is different at each run, and hyperparameter value seem to
be too high..
I have a couple of questions:
1. I must find lambda which maximize: f: sum(log(1+lambda*(x-x0)*Kh(u/h)
But with rbf1 function I obtain a very small number and the log becames
0 (log of 1+ e^-230 etc)
2. Why I find different hyperparameter for each run? I should impose
set.seed? But why hyperparameter are so high?
3. Which formula I should use for sigma in rbf1?
Thank's in advance.