RE: [R] cv.glm {boot}

From: Liaw, Andy <andy_liaw_at_merck.com>
Date: Wed 16 Mar 2005 - 03:08:05 EST


> From: Trevor Wiens
>
> On Tue, 15 Mar 2005 07:05:49 +0000 (GMT)
> Prof Brian Ripley <ripley@stats.ox.ac.uk> wrote:
>
> >
> > Cross-validation assumes exchangeability of units. You can
> easily write
> > your own code (lots of examples in MASS), but first you
> would need to
> > prove the validity of what you are attempting. For
> example, dropping
> > chunks in the middle of a time series is not valid unless
> your prediction
> > somehow takes the temporal structure into account (and glm
> does not).
> >
>
> Yes, I'm aware of that and I do have a number of predictors
> which vary with time (from year to year such as precipitation
> or properly timed vegetation indices from each year....) so
> that isn't my problem. Also my spatial blocking is also valid
> (distinct partitions of the study area). I'm also aware of
> the problems of spatial autocorrelation and have taken some
> measures to deal with that. I am however rather new at R and
> not a statistician, so I am heavily reliant on books such as
> Hosmer and Lemeshow or Manley(Resource selection by Animals)
> on procedure. Unforunately, they are not S-plus or R oriented
> so I have some difficulty translating those ideas to R.
>
> You mention lots of examples in MASS regarding
> cross-validation, but I can't find them. Perhaps I'm looking
> in the wrong spot. I've done help.search('validation'), ....
> and found nothing that seemed obviously applicable to my
> problem. I suppose I should pick up a copy of your books
> which would probably be very helpful. However, if it isn't
> too much trouble. I would really appreciate a bit more direct help.

`MASS' _is_ a book, the supporting software of which contains a `scripts' subdirectory that has R verion of codes used in the book, including code for CV.

Andy  

> This is what I assumed I would do somethink like this (in
> this example basp = Baird's Sparrow presence or absence)
>
> train <- birddata[birddata$recordyear != 2000]
> test <- birddata[birddata$recordyear == 2000]
> train.glm <- glm(basp ~ elev + slope + precip + precip_1 ...,
> data=birddata, family=binomial)
> pred <- predict(train.glm, newdata=test, type='response')
> actual <- test$basp
> what happens next??
>
> Thanks in advance.
>
> T
> --
> Trevor Wiens
> twiens@interbaun.com
>
> The significant problems that we face cannot be solved at the same
> level of thinking we were at when we created them.
> (Albert Einstein)
>
> ______________________________________________
> R-help@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
>
>
>



R-help@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 Received on Wed Mar 16 03:13:27 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:30:47 EST