CODEWITHSUNDEEP
r
PirateGrunt
PirateGrunt

Reputation: 424

Warning in gam with release of R 3.6.1

I now see a warning when running gam(). MWE below:

library(gam)
data(Boston, package = 'MASS')
model.gam <- gam(formula = medv ~ crim + zn, data = Boston)

The warning I receive is:

Warning message:
In model.matrix.default(mt, mf, contrasts) :
   non-list contrasts argument ignored

I think this may relate to a change in the model.matrix() function in the latest R release: https://cran.r-project.org/doc/manuals/r-release/NEWS.html. I did not receive a warning earlier today when running R version 3.5.

> sessionInfo()
R version 3.6.1 (2019-07-05)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252 
[2] LC_CTYPE=English_United States.1252   
[3] LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C                          
[5] LC_TIME=English_United States.1252    

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

other attached packages:
[1] gam_1.16      foreach_1.4.4

loaded via a namespace (and not attached):
[1] compiler_3.6.1   tools_3.6.1      yaml_2.2.0      
[4] codetools_0.2-16 iterators_1.0.10

Upvotes: 2

Views: 2453

Answers (1)

Ben Bolker
Ben Bolker

Reputation: 226761

tl;dr this looks like a (long-standing?) bug in gam::gam that's exposed by the new warning in version 3.6.0, but seems more or less harmless.

To see what's going on, set options(warn=2) (convert warnings to errors) and then run traceback() after the error occurs.

...
4: warning("non-list contrasts argument ignored")
3: model.matrix.default(mt, mf, contrasts)
2: model.matrix(mt, mf, contrasts)
1: gam(formula = medv ~ crim + zn, data = Boston)

OK, the fact that the warning/error occurs inside model.matrix.default() is not surprising. We have to examine/debug gam::gam to see what's going on. If we examine the source code, the only place that "contrasts" occurs is:

X <- if (!is.empty.model(mt)) 
    model.matrix(mt, mf, contrasts)

contrasts is not an argument to the function, and can't have come from the ... argument. The ... argument gets passed through to gam.control(), which also has a ... argument -- which it silently ignores! Thus if you used gam(..., contrasts=contr.sum) (or something like that) in the past, thinking it would change the contrasts, you were wrong. The only reason that this use of contrasts did not throw a warning (for the use of an undefined global variable) is that there is a stats::contrasts function that would be found.

It would be public-spirited to e-mail the maintainer (maintainer("gam")) and let them know about this. Two possible fixes would be (1) remove the contrasts argument from the model.matrix call or (2) add a contrasts argument to gam() (default value NULL) that will be passed through.

Upvotes: 1

Related Questions