further data.frame-like consistency in j: logical vector

[MichaelChirico]

This is simply an extension of #1188 asking for data.frame-like consistency with respect to another common subsetting operation:

set.seed(2030)
DF = data.frame(a = LETTERS, b = letters, c = rnorm(26), d = rnorm(26), e = as.factor(1:26))
DF[ , sapply(DF, is.numeric)]
#              c            d
# 1   1.22587150  0.654058019
# 2  -1.02641266  1.442031659
# 3  -1.33873014 -0.442854108
# --TRUNCATED --
# 24  0.73060958  0.513225722
# 25  0.33502260  0.543853894
# 26  0.53034743 -1.786070335

setDT(DF)
DF[ , sapply(DF, is.numeric)]
#     a     b     c     d     e 
# FALSE FALSE  TRUE  TRUE FALSE 

For now, the workaround is to wrap sapply in which. But here is another case where the data.frame code is more readable and I think unambiguous.

[mattdowle]

Good thought but one ambiguity could be if "DF" is a column name of DF. Say it's a list column where user may well want to sapply is.numeric through the items of the DF$DF list column.

DF[ , sapply(.SD, is.numeric)] ?

[MichaelChirico]

I guess that's indeed possible. Seems like poor practice to keep a column within a table that shares the name, no? Unless you have a use case in mind?

Anyone that does that already has to walk on eggshells to be careful about scoping rules...

[mattdowle]

x and y are very common for both data.table names and column names.
What about making it work for .SD?