I want to do cross validation after running nonparametric regression on my data. Unlike parametric regression where I can first find my parameters and then easily handle the CV set with these parameters to find my MSE, in the non-parametric version this should be done differently.

Given the training and CV set, I decided to do it as follows, take one data point at a time from CV set, look at its neighborhood in the training set using a window function, average the values in the window which gives the new value of the dependent variable.

My question is whether the data point itself has to be included when averaging the neighborhood or it is just the neighbors which has to be considered. Your comments are appreciated since I am relatively new to this stuff.

  • $\begingroup$ I plotted both model error and cv error, this approach worked fine. I considered average of neighbors only and used a gaussian window function. $\endgroup$
    – Zoran
    Commented Dec 1, 2012 at 20:50


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.