STA305H1 Lecture Notes - Lecture 14: Loess, New South Wales Xpt, Scatter Plot
Document Summary
In lecture, we brie y discussed a simple approach to computing possibly computationally complex estimates when n is very large: divide and conquer . To illustrate this process, we will consider the non-parametric regression model. Yi = 2xi exp(xi) + 3 cos(2 xi) (cid:6) (cid:4)(cid:5) g(xi) (cid:3) + i for i = 1, , 100000 where we will estimate g(x) using the loess procedure. In particular, we will compare (cid:2)g(x) computed using all 100000 observations to the divide and conquer estimate g(x) computed as. 100(cid:7) j=1 (cid:2)gj(x) where (cid:2)g1(x) is computed from the rst 1000 observations, (cid:2)g2(x) is computed from the second. The r code used to generate the data and compute the estimates is given below. > y r0 fit0