Confusing warning message

[MichaelChirico]

I'm trying to split data according to 15-minute timestamps; my approach is:

library(data.table)
set.seed(210349)
DT = data.table(t = sample(10000, 100), V = rnorm(100))
# t is in seconds, so 900 seconds is 15 minutes
DT[ , ID := .GRP, keyby = .(t %/% 900L)]

Warning message:
In [.data.table(DT, , :=(ID, .GRP), keyby = .(t%/%900L)) :
:= keyby not straightforward character column names or list() of column names, treating as a by:t%/%900

It's weird that this gives me a warning since it's purely related to the use of keyby:

# (no warning)
DT[order(t), ID := .GRP, by = .(t %/% 900L)]

(i in this case doesn't need to match that in by exactly)

For sanity:

DT[ , ID := .GRP, keyby = .(t %/% 900L)]
DT[order(t), ID2 := .GRP, by = .(t %/% 900L)]
DT[ , identical(ID, ID2)]
# [1] TRUE

What's the reason for this warning?

[franknarf1]

I'm guessing it's that when you use := with keyby, it ...

1. does the column edit with :=
2. sets the key to the keyby= columns

If your keyby= columns do not exist in the table, it cannot set them as the key. Dunno if there's some clearer way to write the warning.

[MichaelChirico]

I see, yes, good point. maybe a better warning would be along the lines of "Cannot set key to an ad-hoc column, so skipping to setkey() following computation of j"