[R] Start Model for POLYCLASS

[Xiaogang Su]

Dear Prof. Kooperberg,

Thanks a lot for your clarification! That is very helpful. I will check
out the polynomial
spline functions in S-Plus that you mentioned. 

Best regards,
-XG Su


================================
Xiaogang Su,  Assistant Professor
Department of Statistics and Actuarial Science
University of Central Florida
Orlando, FL 32816
(407) 823-2940 [O]
xiaosu at mail.ucf.edu
http://pegasus.cc.ucf.edu/~xsu/

>>> <clk at fhcrc.org> 07/10/06 4:01 PM >>>
That's correct. The option is not there.
You can force it in a starting model - but it can still be deleted,

At some stage Insightful implemented a commercial set of polynomial
spline functions in S-Plus that does have that option. They got it
as far as a beta-library [quoting from my own homepage
http://bear.fhcrc.org/~clk]:
============================
 Doug Clarkson at Insightful Corp. has implemented commercial versions
of hare,
heft, logspline and a variety of related methods, with lots of
additional
features and options. See
http://www.insightful.com/downloads/libraries/default.asp. Look for
S+BEST. 
============================
There are remaining bugs in this code.... Doug Clarkson is no longer
at Insightful, and the code seems to be pretty much abbandoned.

Charles

Quoting Xiaogang Su <xiaosu at mail.ucf.edu>:

> Dear Dr. Graves, 
> 
> Thanks for your information and kindness! I really appreciate it.
> According to my findings so far, POLYCLASS doesnot seem to allow for
> variables being forced into the model. Even the commercialized MARS
> doesnot have this function either. However, one ad hoc approach that
> they suggested is to fit linear (or logistic) model and form rediduals
> first and then run MARS with the residuals.  
> 
> Best regards,
> -XG Su 
> 
> 
> ================================
> Xiaogang Su,  Assistant Professor
> Department of Statistics and Actuarial Science
> University of Central Florida
> Orlando, FL 32816
> (407) 823-2940 [O]
> xiaosu at mail.ucf.edu
> http://pegasus.cc.ucf.edu/~xsu/
> 
> 
> >>> Spencer Graves <spencer.graves at pdf.com> 7/7/2006 5:49:05 AM >>>
> 	  I have not seen a reply, so I will offer a few suggestions,
> even 
> though I've never used the 'polspline' package.  I scanned several of 
> the help pages and looked in "~\library\polspline' where R is
installed
> 
> on my hard drive and found no further documentation, and I found
> nothing 
> new from RSiteSearch("polyclass").  Googling for 'polyclass' led me to
> 
> "http://bear.fhcrc.org/~clk/soft.html", the home page for Charles 
> Kooperberg, the author and maintainer.  If it were my problem, I might
> 
> try writing him directly at the email address given in 
> help(package="polyclass");  I'm including him as a "cc" on this reply.
> 
> 	  I spent time looking at this, because 'polyclass', 'polymars'
> and 
> 'polspline' seem potentially related related to one of my secondary 
> interests.  Unfortunately, I couldn't find sufficient documentation to
> 
> allow me to proceed with the time I felt I could afford to invest in 
> this right now.
> 
> 	  If it were my problem, before I wrote another email about
> this, I'd 
> first list the function 'polyclass', make a local copy, then work 
> through an example line by line using 'debug(polyclass)'.  This
'debug'
> 
> facility is remarkably powerful and easy to use.  I've solved many 
> problems like this using 'debug' in this way.
> 
> 	  If that failed to provide the necessary enlightenment, I'd
> submit 
> another post including a self-contained example based on a
modification
> 
> of the 'iris' example featured in the 'polyclass' help page.  Your 
> example is NOT self contained, so I would NOT use that.  Using the 
> 'iris' example would make it much easier to explain clearly what you 
> want.  It also makes it much easier for someone like me to experiment 
> with alternatives and describe what I did in terms of a tested
> example.
> 
> 	  Hope this helps.
> p.s.  I suggest you also review the posting guide! 
> "www.R-project.org/posting-guide.html".
> 
> Xiaogang Su wrote:
> > Dear all, 
> > 
> > I have a question on how to set up the starting model in POLYCLASS
> and
> > make sure the terms in the starting model retained in the final
> > POLYCLASS model. 
> > 
> > In the function POLYMARS, this can be done using the STARTMODEL
> option.
> > See below for example, I started with model 
> > y= b0 + b1*X1 + b2*X2 + b3*X4 + b4*X5 + b5*X2*X5 + e
> > 
> >> m00 <- matrix(c(
> >      1,  NA, 0, NA, 1,     
> >      2,  NA, 0, NA, 1,     
> >      4,  NA, 0, NA, 1,     
> >      5,  NA, 0, NA, 1,
> >      2,  NA, 5, NA, 1),nrow = 5, ncol=5, byrow=TRUE);
> > 
> >> m2 <- polymars(response=PID2$y, predictors=PID2[,1:7], 
> > startmodel=m00) 
> >> summary(m2)  
> > 
> > But I could not figure out how this works for POLYCLASS. There is an
> > option FIT in POLYCLASS, which needs to be a POLYCLASS object
though.
> 
> > 
> > Any suggestion or information is greatly appreciated. 
> > 
> > Sincerely,
> > Xiaogang Su
> > 
> > 
> > ================================
> > Xiaogang Su,  Assistant Professor
> > Department of Statistics and Actuarial Science
> > University of Central Florida
> > Orlando, FL 32816
> > (407) 823-2940 [O]
> > xiaosu at mail.ucf.edu 
> > http://pegasus.cc.ucf.edu/~xsu/ 
> > 
> > ______________________________________________
> > R-help at stat.math.ethz.ch mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help 
> > PLEASE do read the posting guide!
> http://www.R-project.org/posting-guide.html
>