Many thanks to Greg Snow who provided such helpful reply:
He has already replied via the list, but I 'm taking the liberty
of posting the following from him:
Best,
Sabit
---------------------- Forwarded by Sabit Cakmak/HC-SC/GC/CA on 09/27/2001 01:36
PM ---------------------------
Greg Snow <snow@fisher.byu.edu> on 09/27/2001 09:48:37 AM
To: Sabit Cakmak/HC-SC/GC/CA@HWC
cc: s-news@wubios.wustl.edu
Subject: Re: [S] LOESS Smoother
On Tue, 25 Sep 2001 Sabit_Cakmak@hc-sc.gc.ca wrote:
>
>
> The loess smoother first computes the lamda % nearest-neighbours to the target
> point,. A tricube kernel , centered at the target point, becomes zero at the
> furthest neighbour. The smooth at the target point is the fitted value from
the
> locally-weighted linear fit, with weights supplied by the kernel (Hastie
> &Tibshirani).
>
> Is it possible to see those weights when we employed loess.smooth in Splus and
> how does Splus deal with edge effects.
Attached is a demonstration function for loess that helps picture what is
going on (in 2 dimensions). A sample of using it is:
> source("loess.demo.ssc")
> attach(ethanol)
> loess.demo(E,NOx)
> loess.demo(E,NOx, span=.25)
> loess.demo(E,NOx, degree=2)
for each run of the function, click on the graph and it will redraw to
show the window of points being considered, their relative weights (the
area of the circles) and the line/curve being used to for the prediction
at that point. 2 loess curves are drawn, the exact loess prediction and
the smooth that S-PLUS adds by default (they are usually pretty close).
Right click to end the demonstration.
I don't know if this is what you wanted, but it at least shows what is
going on behind the scenes.
To the Mailing list: I sent a copy of this function to the entire mailing
list last spring, then I received some suggestions/modifications to make
it better. The version attached has more features than the original so if
you found the original useful, take a look at the improved version.
--
Greg Snow, PhD Office: 223A TMCB
Department of Statistics Phone: (801) 378-7049
Brigham Young University Dept.: (801) 378-4505
Provo, UT 84602 email: gls@byu.edu
loess.demo.ssc
Description: Binary data
|