a bug with group_by + mutate_each ?

[GodefroyClair]

Hi,
With dplyr Version: 0.4.3 and R :
version 3.3.0 (2016-05-03)
Platform: x86_64-apple-darwin14.5.0 (64-bit)
Running under: OS X 10.11.6 (El Capitan)
this code :

set.seed(7)
df <- data.frame(x=10*runif(9), y=20*rnorm(9), group=rep(c("a","b","c"),3))
df2 <- df %>% group_by(group) %>% mutate_each(funs(scale))

passes ok and if I call df2, I can see its content. But if I do View(df2) or str(df2) I got an error

Error in FUN(X[[i]], ...) :
dims [product 3] do not match the length of object [9]

[GodefroyClair]

Can someone answer me on this please ? :)

[joranE]

The issue seems to be that scale returns a matrix not a vector, and somehow (in dplyr?) the dimension attribute is being set to be 3x1 for the whole column. Note that if you type df2[[1]] you only see three values, because the dimension attribute is sort of "lying".

A workaround might be to use my_scale <- function(x) c(scale(x)) instead to force a dimensionless vector to be returned.

[GodefroyClair]

Thank you for the answer JoranE. That was what I suspected but your proof and your solution are better than mine. Is there something else to do to report this now that we suspect quite strongly that it's a bug in dplyr ?

[jrosen48]

I think this SO question and answer may be relevant: http://stackoverflow.com/questions/35775696/trying-to-use-dplyr-to-group-by-and-apply-scale

[krlmlr]

I think this is a duplicate of #1918.

[krlmlr]

Can you please try devtools::install_github("krlmlr/dplyr@b-1918-grouped-dims")?

[vnijs]

Possibly related issue:

> dat <- mtcars[1:5,1:3]
> df <- mutate_each(dat, funs(scale(.))) %>% mutate(somevar = cyl)
Error: Each variable must be a 1d atomic vector or list.
Problem variables: 'mpg', 'cyl', 'disp'

Using as.vector(scale(.)), suggested by @krlmlr, works