Simplify weights

[lbittarello]

This pull request:

• removes the type of the values vector as a parameter of weights;
• deprecates the values method;
• adds dims as a keyword argument of sum (in line with the unweighted version).

It addresses issues #358 and #371.

I have not dropped the tests of values yet.

[lbittarello]

@nalimilan, we've also discussed the possibility of changing indexing so it returns another weight vector. I've tried this:

@propagate_inbounds function Base.getindex(wv::W, i::Integer) where W <: AbstractWeights
    @boundscheck checkbounds(wv, i)
    v = [wv.values[i]]
    W(v, v)
end

@propagate_inbounds function Base.getindex(wv::W, i::AbstractArray{<:Int}) where W <: AbstractWeights
    @boundscheck checkbounds(wv, i)
    v = wv.values[i]
    W(v, sum(v))
end

Base.getindex(wv::AbstractWeights, ::Colon) = wv

function Base.iterate(it::AbstractWeights, (el, i) = (it.values[1], 0))
	return i >= it.len ? nothing : (el, (it.values[i], i + 1))
end

It works in creating weight vectors from subsets of weight vectors. The iterator seems fine too. But other functionality breaks (for instance, display).

[nalimilan]

It works in creating weight vectors from subsets of weight vectors. The iterator seems fine too. But other functionality breaks (for instance, display).

Thanks, I had forgotten about this part. We should only define the i::AbstractArray method I think. Indexing with i::Integer should still return a scalar as that's part of the AbstractArray interface.

[lbittarello]

I relaxed the input type of sum (from AbstractArray{<:Number} to AbstractArray) to make up for my previous removal of the fallback method for sum (line 348 of the old code). All tests should go through now.

[codecov]

Codecov Report

Merging #526 into master will increase coverage by 0.02%.
The diff coverage is 85.71%.

@@            Coverage Diff            @@
##           master    #526      +/-   ##
=========================================
+ Coverage   90.47%   90.5%   +0.02%     
=========================================
  Files          21      21              
  Lines        2100    2095       -5     
=========================================
- Hits         1900    1896       -4     
+ Misses        200     199       -1

Impacted Files	Coverage Δ
src/deprecates.jl	0% <ø> (ø)	⬆️
src/weights.jl	86.53% <85.71%> (ø)	⬆️
src/moments.jl	100% <0%> (+0.67%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 95b794a...6f974ea. Read the comment docs.

[lbittarello]

Should we harmonize the inputs of sum! and mean!? Since the dimension is a compulsory argument for both, it would make sense to make it a positional instead of a keyword argument in mean!.

[nalimilan]

Should we harmonize the inputs of sum! and mean!? Since the dimension is a compulsory argument for both, it would make sense to make it a positional instead of a keyword argument in mean!.

Since sum is defined in Base, we should follow what happens there: drop dims, and infer it from the shape of the output array. But I wouldn't spend too much time on this in StatsBase since I've already rewritten a lot of this code as part of JuliaLang/julia#31395. Any changes made here will make it harder to me to rebase that PR on the latest StatsBase.