I'm trying to add arguments to the AIC method for some classes -- things like weights=TRUE to calculate AIC weights corr=TRUE, nobs to calculate AICc delta=TRUE to put a delta-AIC column in the output. The problem is that AIC is defined as AIC(object, ..., k=2) where k is the constant associated with the penalty term and ... is a list of objects that will have their AICs calculated and listed. Thus I'm not allowed (I think) to extend the definition to AIC(object, ..., nobs, corr=FALSE, delta=FALSE, weights=FALSE, k=2) An ugly solution to this is to go through the ... list and pull out the values of any of the other arguments I want, and then set them to NULL, leaving the list composed only of additional objects to calculate the AIC for. But it seems really ugly ... Any ideas for a better way to do this would be appreciated. (Please be gentle if possible, I'm a bit out of my depth with S4 ...) cheers Ben Bolker -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 254 bytes Desc: OpenPGP digital signature Url : https://stat.ethz.ch/pipermail/r-devel/attachments/20061216/4cd5425d/attachment.bin
Prof Brian Ripley
2006-Dec-16 20:31 UTC
[Rd] question about trailing arguments in an S4 method
I don't see how this can work, in S3 or S4. Callers of AIC are entitied to expect it to behave as described on the help page, and so to be able to pass objects called (e.g.) 'delta' and get back exactly the value mentioned. For what you seem to want to do, I think you need your own generic. On Sat, 16 Dec 2006, Ben Bolker wrote:> > I'm trying to add arguments to the AIC method > for some classes -- things like > weights=TRUE to calculate AIC weights > corr=TRUE, nobs to calculate AICc > delta=TRUE to put a delta-AIC column in the output. > > The problem is that AIC is defined as > AIC(object, ..., k=2) where k is the constant associated > with the penalty term and ... is a list of objects > that will have their AICs calculated and listed. > Thus I'm not allowed (I think) to extend the definition to > > AIC(object, ..., nobs, corr=FALSE, delta=FALSE, weights=FALSE, k=2) > > An ugly solution to this is to go through the ... list and > pull out the values of any of the other arguments I want, and > then set them to NULL, leaving the list composed only of > additional objects to calculate the AIC for. But it seems > really ugly ... > > Any ideas for a better way to do this would be > appreciated. (Please be gentle if possible, I'm a bit out > of my depth with S4 ...) > > cheers > Ben Bolker > > >-- Brian D. Ripley, ripley at 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 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595
Ben Bolker <bolker at zoo.ufl.edu> writes:> I'm trying to add arguments to the AIC method > for some classes -- things like > weights=TRUE to calculate AIC weights > corr=TRUE, nobs to calculate AICc > delta=TRUE to put a delta-AIC column in the output. > > The problem is that AIC is defined as > AIC(object, ..., k=2) where k is the constant associated > with the penalty term and ... is a list of objects > that will have their AICs calculated and listed. > Thus I'm not allowed (I think) to extend the definition to > > AIC(object, ..., nobs, corr=FALSE, delta=FALSE, weights=FALSE, k=2)I think you can add args before the '...'. However, you will need to provide values for them in all calls or else they will steal a fit from the '...' args. I don't understand really what you are trying to achieve, so this may not be helpful. Nevertheless, here is a dummy example that you may find useful. setClass("Foo", representation=representation( fit="lm")) setMethod("AIC", "Foo", function(object, weights=TRUE, delta=TRUE, ..., k=2) { ans <- AIC(object=object at fit, k=k, ...) args1 <- list(weights=weights, delta=delta, k=k) cat("args\n") print(args1) cat("answer:\n") print(ans) }) example("AIC") foo <- new("Foo", fit=lm1) AIC(foo) ## AIC(foo, lm1) # this doesn't work, lm1 gets picked up as the weights arg AIC(foo, weights=FALSE, delta=TRUE, lm1, lm1, lm1) + seth