R Package conflict between gam and mgcv?

Question

Detaching packages in R isnt good practice (see ?detach ), but for some reasons I have to switch between the packages gam and mgcv . Once mgcv was attached and detached (and all the dependencies in the namespace unloaded!), functions of gam produce some strange errors (please forgive the terminology). It seems that - even though unloaded one step before - mgcv and friends are back in the namespace and function dispatching goes wrong. Does anyone had the same problem before?

# fresh session
t.s1 <- search()
t.lN1 <- loadedNamespaces()

# some dummy data
data <-data.frame(is.exc=sample(x=c(0,1),size=100,replace=T),
year=1:100,doy=rep(1:5,times=20))
t.dof <- 2

# everything works fine
library(gam)
t.gam1 <- gam::gam(is.exc~s(year,df=t.dof)+s(doy,df=t.dof),data=data,family=poisson)
t.pred1 <- gam::predict.gam(t.gam1,newdata=data,type='terms')
detach('package:gam',unload=T,character.only=T)
detach('package:splines',unload=T,character.only=T)

# compare attached packages and namespaces with fresh session
t.s2 <- search()
t.lN2 <- loadedNamespaces()
identical(t.s1,t.s2)
identical(t.lN1,t.lN2)

# attach and detach mgcv
library(mgcv)
detach('package:mgcv',unload=T,character.only=T)
unloadNamespace('nlme')
unloadNamespace('Matrix')
unloadNamespace('lattice')
unloadNamespace('grid')

# compare again attached packages and namespaces with fresh session
t.s2 <- search()
t.lN2 <- loadedNamespaces()
identical(t.s1,t.s2)
identical(t.lN1,t.lN2)

# use package gam again and produce errors
library(gam)
t.gam2 <- gam::gam(is.exc~s(year,df=t.dof)+s(doy,df=t.dof),data=data,family=poisson)
gam::summary.gam(t.gam2)
t.pred2 <- gam::predict.gam(t.gam2,newdata=data,type='terms')

# why do we have mgcv and friends in the namespace?
loadedNamespaces()

My session infos are (fresh session):

R version 3.0.2 (2013-09-25)
Platform: x86_64-pc-linux-gnu (64-bit)

locale:
[1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=en_US.UTF-8         LC_COLLATE=en_US.UTF-8    
[5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8    LC_PAPER=en_US.UTF-8       LC_NAME=C                 
[9] LC_ADDRESS=C               LC_TELEPHONE=C             LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

loaded via a namespace (and not attached):
[1] tools_3.0.2

and I use the latest versions of gam (1.09) and mgcv (1.7-28). Any hints are appreciated!

Answer 1

The problem, as you ANTICIPATED, is that both gam and mgcv packages install S3 methods for "gam" objects. But, as stated in the ?detach documentation:

registered S3 methods from the namespace will not be removed.

So in your case it is easy to see that is the cause of your problems:

library(gam)
# installs gam::print.summary.gam
identical(getS3method('print', 'summary.gam'), gam:::print.summary.gam)
[1] TRUE


library(mgcv)
# installs mgcv::print.summary.gam
identical(getS3method('print', 'summary.gam'), mgcv:::print.summary.gam)
[1] TRUE

# save a pointer before unloading namespaces
mgcv_psgam <- mgcv:::print.summary.gam

detach('package:mgcv',unload = TRUE, character.only = TRUE)
# after the detach, the method from mgcv is still installed !!!
identical(getS3method('print', 'summary.gam'), mgcv_psgam)
[1] TRUE

Conclusion: make sure you never load mgcv