Method home for knn-cv-1

This method uses an average of the k nearest neighbors in the training set for predictions. For the squared error loss function, the predictions are the mean of the targets of the nearest neighbors. For absolute error loss the median of the targets of the nearest neighbors is used. For negative log predictive loss a Gaussian predictive distribution is fit to the nearest neighbors. For each of the three loss types, the value of k (the neighborhood size) is chosen leave-one-out cross validation, repeated exhastively for all possible values of k. Detailed description of the method is available in postscript format.


Source code for an implementation in C is available as a compressed tar file.


Directory listing of the results available for the knn-cv-1 method. Put the desired files in the appropriate methods directory in your delve hierarchy and uncompress them with using the "gunzip *.gz" command and untar them using "tar -xvf *.tar".

Related References

Last Updated by Carl Edward Rasmussen, September 27, 1996