[19324] in s-news-athena
Re: [S] S-plus ace() function
daemon@ATHENA.MIT.EDU (Prof Brian Ripley)
Mon Sep 27 12:39:07 1999
Message-Id: <199909271633.RAA16117@toucan.stats.ox.ac.uk>
Date: Mon, 27 Sep 1999 17:33:11 +0100 (BST)
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Reply-To: Prof Brian Ripley <ripley@stats.ox.ac.uk>
To: s-news@wubios.wustl.edu, dvir@math.tau.ac.il
Mime-Version: 1.0
Content-Type: TEXT/plain; charset=us-ascii
Content-Md5: bvij64zVmRrJjJNSAAzU+w==
> Date: Mon, 27 Sep 1999 14:59:49 +0200 (GMT+0200)
> From: Yoav Dvir <dvir@math.tau.ac.il>
>
> I use Splus 4.5 and have some questions about the ace() function:
>
> 1. As part of the algorithm ace() must have something like gam().
> In gam() one can use the function s() or l() which usually must be
> supplied by smoothing parameters. Nothing like that appears in ace().
> What are the defaults?
It uses supsmu (a running lines smoother), and that chooses its own
bandwidth by cross-validation. You can read about supsmu in its man
page or in the S-PLUS manuals, or in the source code (see below) or
even in Venables & Ripley (1999, p. 285).
> 3. There are two steps in each iteration of the algorithm. If one
> wants to use weights in my view there must be two sets of weights,
> one for each step, Yet one can supply only one set of weights (wt=...),
> how this set of weights is incorporated into the algorithm?
As I understand it, w[i] = 3 means `act as if there are three copies
of (y[i], x[i, ])'.
> 4. In general I have found the help of the software for ace() far from
> being enough. Where I can find more information about the ace()
> implementation (not on the theory of ace()).
You can look at the source code, both the S functions and the Fortran code,
which is on Statlib http://lib.stat.cmu.edu/general/ace. As far as I know,
the call to "acemace" is essentially to MACE in that source file.
(This is guesswork, but I think I did once try substituting it and got
the same answers.) I found this (and the sources for avas and ppreg) very
helpful in understanding these. If you need any more specific help,
I may be able to assist, so please ask me privately.
--
Brian D. Ripley, ripley@stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
