Expression of column names in by: clarify the function of list and parentheses

[Henrik-P]

I'm slightly confused by the use of list (or not) in by.

From ?data.table:

by accepts a list() of expressions of column names

This might be nitpicking, but as long as you don't name your expression, the list doesn't seem to be needed. Unnamed expression in by works without list:

d <- data.table(x = 1:4, y = 2:5)
d[ , sum(y), by = x %% 2]
#    x V1
# 1: 1  6
# 2: 0  8

Furthermore, wrappin a named expression in parentheses instead of list works:

d[ , sum(y), by = (grp = x %% 2)]
#    grp V1
# 1:   1  6
# 2:   0  8

I haven't found anything in the docs on the use of ( in by. The description in Reference semantics, "e) Multiple columns and :=", seems like something else.

---

While being confused by the parentheses anyway, I just tried them in j:

d[ , sum_y = sum(y), by = (grp = x %% 2)]
# Error, fine.

# wrap j in parentheses -> no error, albeit naming of result fails
d[ , (sum_y = sum(y)), by = (grp = x %% 2)]
#    grp V1
# 1:   1  6
# 2:   0  8

---

At this point, I suppose I should provide some constructive comments on the docs, but first I want to check that I haven't overlooked something obvious here. Most likely I have.

Thanks for your great work!

[eantonya]

There is nothing special about the (). The parens in R are a function (and so is the equality operator), that return a value, so it's as if you wrote d[, somefunction(y), by = someotherfunction(x)]. When you write (a = blah), that returns the result of the assignment, i.e. a - it's equivalent to writing

"("("="(a, blah))

to give you the expanded functional form.

[MichaelChirico]

I'm surprised by the name of the output in both cases of by (x then grp). compare too by = foo(x) the column will be named foo

…

On Sep 29, 2017 3:24 AM, "Henrik-P" ***@***.***> wrote: I'm slightly confused by the use of list (or not) in by. From ?data.table: by accepts a list() of expressions of column names d <- data.table(x = 1:4, y = 2:5) This might be nitpicking, but as long as you don't *name* your expression, the list doesn't seem to be needed. Unnamed expression in by works without list: d[ , sum(y), by = x %% 2] # x V1 # 1: 1 6 # 2: 0 8 Furthermore, just tried to wrap a named expression in *parentheses*, it also works: d[ , sum(y), by = (grp = x %% 2)] # grp V1 # 1: 1 6 # 2: 0 8 I haven't found anything in the docs on the use of ( in by. The description in Reference semantics <https://cran.r-project.org/web/packages/data.table/vignettes/datatable-reference-semantics.html>, "e) Multiple columns and :=", seems like something else. ------------------------------ While being confused by the parentheses anyway, I just tried them in j: d[ , sum_y = sum(y), by = (grp = x %% 2)] # Error, fine. # wrap in parentheses -> no error, albeit naming of result fails d[ , (sum_y = sum(y)), by = (grp = x %% 2)] # grp V1 # 1: 1 6 # 2: 0 8 ------------------------------ I suppose I should suggest some improvement of the docs, but first I want to check that I haven't overlooked something obvious here. Most likely I have. Thanks for your great work! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2391>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHQQdXrDuifNUn18s6k7u2k8qlZBgEhWks5sm_JdgaJpZM4Pnw3x> .