new feature: partition into N sets (to add to sets of length N)

[bjarthur]

currently, Base.Iterators.partition accepts an argument n which specifies the length of the disjoint subsets. i frequently find i would rather specify the number of disjoint subsets. has this been considered before? a similar interface-design issue exists with range(start, stop, length) and range(start, stop, step).

i actually already have some code that does this via a new iterator. and i was about to submit a PR to IterTools.
but i think a new method to Base.Iterators.partition might make more sense.

here is the docstring showing the API of what i currently have:

"""
split(xs, n)

Group values of xs into n disjoint sets. The sets will have identical
lengths if length(xs) is evenly divisible by n, otherwise they will be
approximately similar.

julia> collect(split(1:9,4))
4-element Array{Any,1}:
 1:2
 3:4
 5:7
 8:9

See also: partition.
"""

the existing partition API could be changed to partition(collection; n, l), where only one of the two keyword arguments would be specified, one acting as the old second positional argument, and the other specifying the desired number of disjoint sets.

should i stick with IterTools or extend Base? or neither? thanks!

[mcabbott]

I agree this would often be convenient:

Iterators.partition(itr; count) = Iterators.partition(itr, cld(length(itr), count))

But unlike the present method, this one won't work for things like itr = (i for i in 1:10 if rand()>0.5) for which haslength(itr) == false. Perhaps unlike every other method in Iterators. Does that mean it should collect first in such cases? Or not live in Iterators? Or just give an error?

[xgdgsc]

Besides a split for iterator, would including https://juliaml.github.io/MLUtils.jl/stable/api/#MLUtils.chunk for arrays also useful?

[IanButterworth]

Or not live in Iterators

Yeah, a Base.partition function that could do either ngroups and nelements while collecting seems reasonable