melt.data.table inconsistent with reshape2::melt

[MichaelChirico]

See this SO Q&A.

Here's what works in reshape2:

buyer      <- c("A", "A", "B", "A", "B", "C")
seller     <- c("C", "B", "C", "B", "C", "A")
negotiator <- c("B", "C", "D", "D", "A", "B")

df <- data.frame(buyer, seller, negotiator)
reshape2::melt(df,0)
#      variable value
#1       buyer     A
#2       buyer     A
#3       buyer     B
#4       buyer     A
#5       buyer     B
#6       buyer     C
#7      seller     C
#8      seller     B
#9      seller     C
#10     seller     B
#11     seller     C
#12     seller     A
#13 negotiator     B
#14 negotiator     C
#15 negotiator     D
#16 negotiator     D
#17 negotiator     A
#18 negotiator     B

But that's an error in melt.data.table:

setDT(df)
melt(df, 0)

Error in melt.data.table(df, 0) :
One or more values in id.vars is invalid.

This is fine, as we can simply set measure.vars = names(df), but IIUC we're aiming for melt.data.table to subsume reshape2::melt.

[arunsrinivasan]

Michael, thanks. But I'm quite happy with current behaviour of returning an error. melt(df, -1) returns something else. It's hard to keep up with undocumented behaviours.

[MichaelChirico]

Sounds good to me. Just figured I'd report. I also noticed reshape2::melt(df, -1) doing strange things.

[MichaelChirico]

Perhaps worth adding a note to ?melt.data.table??

[arunsrinivasan]

Sure. A separate Note section should be fine.

[MichaelChirico]

Was putting this off because I first wanted to dive into reshape2 to figure out what's going on. For posterity, here's the relevant snippet from reshape2:::melt_check:

varnames <- names(data)
if (!missing(id.vars) && is.numeric(id.vars)) {
    id.vars <- varnames[id.vars]
}
# ...
measure.vars <- setdiff(varnames, id.vars)

So id.vars becomes character(0) when id.vars = 0, and in turn measure.vars becomes exactly varnames. This also helps explain what's happening when id.vars = -1 -- id.vars becomes varnames[-1], i.e., all but the last column.

[mattdowle]

Closed by 6437417 in #2290.
Think the space between # and issue number in the commit message prevented auto close and link.