rhelp.20.trevva at spamgourmet.com
2008-Aug-17 21:38 UTC
[R] Optim stripping attributes from relistable objects
Dear all, The following code is inspired by the help file for the relist() function (see?relist), which explicitly details how you can use a relistable object in conjunction with optim to pass and reconstruct complex parameter structures/groupings. The idea is that the optim() function can only work with vectors, but in many cases you would like to use a complex structure inside the objective function- relist is one way to do that. The problem is that optim appears to be stripping the attributes and therefore the example doesn't seem to run, giving the error at the bottom.> rb.banana <- function(params) {+ #Params is initially a vector + cat("Params initially has the attributes:\n") + print(names(attributes(params))) + #Relisting it turns it into a list... + params <- relist(params) + cat("---------\n") + #..which can then be called in the standard list manner + return( (1-params$x)^2 + 100*(params$y - params$x^2)^2) + }> > ipar <- as.relistable(list(x=5,y=0)) > initial.params <- unlist(ipar) > > #Test to see if rb.banana works properly in the "normal" case > rb.banana(initial.params)Params initially has the attributes: [1] "names" "skeleton" --------- [1] 62516> > #OK, that's good. How about with optim though? > optim(initial.params,rb.banana)Params initially has the attributes: [1] "names" Error in relist(params) : The flesh argument does not contain a skeleton attribute. Either ensure you unlist a relistable object, or specify the skeleton separately.>What's going on here? Has this functionality been removed and the documentation in relist() not updated? Or has the feature been broken? Or have I misinterpreted something here (it wouldn't be the first time!!) Am running R version 2.7.1 (2008-06-23) under windows. Cheers, Mark
Katharine Mullen
2008-Aug-17 22:15 UTC
[R] Optim stripping attributes from relistable objects
> > The following code is inspired by the help file for the relist() > function (see?relist), which explicitly details how you can use aThe example is in the 'Details' section and, indeed, it looks like it no longer works.> relistable object in conjunction with optim to pass and reconstruct > complex parameter structures/groupings. The idea is that the optim() > function can only work with vectors, but in many cases you would like > to use a complex structure inside the objective function- relist is > one way to do that. The problem is that optim appears to be stripping > the attributes and therefore the example doesn't seem to run, giving > the error at the bottom.You can get around this by specifying skeleton for relist: rb.banana <- function(params) { params <- relist(params, skeleton=list(x=NA,y=NA)) return( (1-params$x)^2 + 100*(params$y - params$x^2)^2) } ipar <- as.relistable(list(x=5,y=0)) initial.params <- unlist(ipar) xx <- optim(unlist(initial.params), rb.banana)> > > rb.banana <- function(params) { > + #Params is initially a vector > + cat("Params initially has the attributes:\n") > + print(names(attributes(params))) > + #Relisting it turns it into a list... > + params <- relist(params) > + cat("---------\n") > + #..which can then be called in the standard list manner > + return( (1-params$x)^2 + 100*(params$y - params$x^2)^2) > + } > > > > ipar <- as.relistable(list(x=5,y=0)) > > initial.params <- unlist(ipar) > > > > #Test to see if rb.banana works properly in the "normal" case > > rb.banana(initial.params) > Params initially has the attributes: > [1] "names" "skeleton" > --------- > [1] 62516 > > > > #OK, that's good. How about with optim though? > > optim(initial.params,rb.banana) > Params initially has the attributes: > [1] "names" > Error in relist(params) : > The flesh argument does not contain a skeleton attribute. > Either ensure you unlist a relistable object, or specify the skeleton > separately. > > > > What's going on here? Has this functionality been removed and the > documentation in relist() not updated? Or has the feature been broken? > Or have I misinterpreted something here (it wouldn't be the first > time!!) > > Am running R version 2.7.1 (2008-06-23) under windows. > > Cheers, > > Mark > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. >
Prof Brian Ripley
2008-Aug-18 12:21 UTC
[R] Optim stripping attributes from relistable objects
On Mon, 18 Aug 2008, rhelp.20.trevva at spamgourmet.com wrote:> Ok - there is an easy work-around by just to passing the skeleton to > the objective function. But would it be possible to clarify this point > in the relist() help file please (specificaly the example I would > think), implies that attributes are copied through by optim and that > this is expected behaviour?Already done. You can always check the current version on the SVN server, here https://svn.r-project.org/R/branches/R-2-7-branch/src/library/utils/man/relist.Rd> > Thanks, > > Mark > > > > 2008/8/18 Prof Brian Ripley - ripley at stats.ox.ac.uk > <>: >> On Mon, 18 Aug 2008, Katharine Mullen wrote: >> >>>> >>>> The following code is inspired by the help file for the relist() >>>> function (see?relist), which explicitly details how you can use a >>> >>> The example is in the 'Details' section and, indeed, it looks like it no >>> longer works. >> >> I don't think it ever did (after all there is a comment that dnorm does not >> have the parameters stated). >> >> The subject line here illustrates the misunderstanding. Optim *strips* >> nothing -- it creates a new object to pass to the function. The help says >> >> Any names given to 'par' will be copied to the vectors passed to >> 'fn' and 'gr'. >> >> Note the use of 'copy' here. You cannot assume that any other attributes >> will be copied, and this behaviour has not been changed (except to add names >> in March 2006, well before relist() was added), AFAICS. >> >>>> relistable object in conjunction with optim to pass and reconstruct >>>> complex parameter structures/groupings. The idea is that the optim() >>>> function can only work with vectors, but in many cases you would like >>>> to use a complex structure inside the objective function- relist is >>>> one way to do that. The problem is that optim appears to be stripping >>>> the attributes and therefore the example doesn't seem to run, giving >>>> the error at the bottom. >>> >>> You can get around this by specifying skeleton for relist: >>> >>> rb.banana <- function(params) { >>> params <- relist(params, skeleton=list(x=NA,y=NA)) >>> return( (1-params$x)^2 + 100*(params$y - params$x^2)^2) >>> } >>> >>> ipar <- as.relistable(list(x=5,y=0)) >>> >>> initial.params <- unlist(ipar) >>> >>> xx <- optim(unlist(initial.params), rb.banana) >>> >>>> >>>>> rb.banana <- function(params) { >>>> >>>> + #Params is initially a vector >>>> + cat("Params initially has the attributes:\n") >>>> + print(names(attributes(params))) >>>> + #Relisting it turns it into a list... >>>> + params <- relist(params) >>>> + cat("---------\n") >>>> + #..which can then be called in the standard list manner >>>> + return( (1-params$x)^2 + 100*(params$y - params$x^2)^2) >>>> + } >>>>> >>>>> ipar <- as.relistable(list(x=5,y=0)) >>>>> initial.params <- unlist(ipar) >>>>> >>>>> #Test to see if rb.banana works properly in the "normal" case >>>>> rb.banana(initial.params) >>>> >>>> Params initially has the attributes: >>>> [1] "names" "skeleton" >>>> --------- >>>> [1] 62516 >>>>> >>>>> #OK, that's good. How about with optim though? >>>>> optim(initial.params,rb.banana) >>>> >>>> Params initially has the attributes: >>>> [1] "names" >>>> Error in relist(params) : >>>> The flesh argument does not contain a skeleton attribute. >>>> Either ensure you unlist a relistable object, or specify the skeleton >>>> separately. >>>>> >>>> >>>> What's going on here? Has this functionality been removed and the >>>> documentation in relist() not updated? Or has the feature been broken? >>>> Or have I misinterpreted something here (it wouldn't be the first >>>> time!!) >>>> >>>> Am running R version 2.7.1 (2008-06-23) under windows. >>>> >>>> Cheers, >>>> >>>> Mark >>>> >>>> ______________________________________________ >>>> R-help at r-project.org mailing list >>>> https://stat.ethz.ch/mailman/listinfo/r-help >>>> PLEASE do read the posting guide >>>> http://www.R-project.org/posting-guide.html >>>> and provide commented, minimal, self-contained, reproducible code. >>>> >>> >>> ______________________________________________ >>> R-help at r-project.org mailing list >>> https://stat.ethz.ch/mailman/listinfo/r-help >>> PLEASE do read the posting guide >>> http://www.R-project.org/posting-guide.html >>> and provide commented, minimal, self-contained, reproducible code. >>> >> >> -- >> 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 >> >-- 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