Ordering result of group_by

[hadley]

It would be nice to have an option to order the result of group_by so that (e.g.)

mtcars %.% group_by(cyl) %.% summarise(mpg = mean(mpg))

is ordered by increasing cyl. I think this should be the default for relatively small numbers of groups.

It doesn't make any difference to an analysis, but I find it surprisingly distracting.

[romainfrancois]

There would be some extra cost. Essentially a sort but in the order of the number of groups. So relatively small compared to the training of the original hash map.

But the upside obviously is that we get predictable results and it simplifies reading the output. So I'd vote for making this the default.

[hadley]

I think it could be the default if n_groups < 1e4 - then the cost will be minimal, and if you have more groups than that, you're less likely to care about the order.

[mnel]

In reference to closed issue #263 group_by should retain the order of appearance the grouped data by default. (Which it currently does not).

df <- data.frame(users=c(1,2,3),items=1:3) %.%group_by(users)
#Source: local data frame [3 x 2]
#Groups: users
#
#  users items
# 1     1     1
# 2     2     2
# 3     3     3
df%.%group_by(users)%.%summarise(identity(items))
# Source: local data frame [3 x 2]
# 
# users identity(items)
# 1     3               3
# 2     1               1
# 3     2               2