most_common method for counters

[ymer]

Return a list of the most common elements in a counter and their counts, ordered from the most common to the least.

most_common(counter([1, 6, 2, 2, 5, 5, 5]))

4-element Array{Pair{Int64,Int64},1}:
5=>3
2=>2
6=>1
1=>1

[rawls238]

I think fixing the indentation here should get rid of the weirdness in the rich diff

[kmsquire]

For parity with Julia Base, it might be nice to create a version of select which does this. See http://docs.julialang.org/en/release-0.4/stdlib/sort/#Base.select! (and http://docs.julialang.org/en/release-0.4/stdlib/sort/#Base.select). Or use select to define most_common, if that makes sense.

There are related functions, nsmallest and nlargest, which are implemented for heaps in this package as well.

[ymer]

Could the heap nlargest be extended to take a "by" keyword?

[kmsquire]

Well, it could. This might not be a good idea if performance is an issue (see http://docs.julialang.org/en/release-0.4/manual/performance-tips/#declare-types-of-keyword-arguments, although there isn't much explanation as to why keywords can be slow).

I'm curious what your use case would be, since nlargest would already choose the largest values according to the value, and that's what you're proposing here.

Another idea, which I like a little better, would be to implement a separate HeapCounter type, which created a counter based on heaps instead of dictionaries.

[ymer]

I think that the users would often want all the values in order, rather than the n most common. And most_common could be somewhat more suitable for this than nlargest.

Here is another suggestion:

most_common(ct) = sort(collect(ct.map), by=p->p[2], rev=true)

most_common(ct, n) = select(collect(ct.map), 1:n, by=p->p[2], rev=true)

[oxinabox]

It may also be worth special casing most_common(ct, 1) which I think is another common use-case.
select is really slow compared to other methods for finding the single most common.

Also good idea. I didn't even know this is missing. This is one of the main uses of the Counter type.

[rawls238]

@ymer any updates on this? It would be nice to merge this in once some of the comments are resolved

[kmsquire]

Bump.

[kmsquire]

This seems to have been satisfied by #375 (using nlargest and nsmallest).