Classification or regression using loo errors

摘要

Approach to implement the algorithm for determining the regularization parameter λ of the vast number LOO of a matter of (RLS) regularized least squares (leave-one-out) error for the regularization parameter λ [challenge] a vast number of I will be disclosed. Algorithm implemented in [SOLUTION] The method uses time and space approximately the same as when training regularized least squares classifier / regression algorithm one. Based on the eigenvalue decomposition of the non-regularization kernel matrix, wherein the method comprises a suitable classification / regression process to the data set of medium. This process, accurate classification / regression is made possible by applying standard large sets of data (benchmark datasets), experimentally, using a Gaussian kernel with a value slightly greater than the bandwidth parameter σ. We also show a method of using a method of using such large and σ, as performed in the entire range of λ operations LOO value determines a linear time algorithm suitable for large data sets. [Selection Figure Figure 2