Compute statistics along dimensions

[lindahua]

It is useful to compute certain statistics along specific dimensions. I open this thread to derive the development along this line.

• [ ] skewness
• [ ] kurtosis
• [ ] moment
• [ ] entropy
• [ ] geomean
• [ ] harmmean
• [ ] variation
• [ ] sem

Some other functions like mode and quantile require more sophisticated data structure to compute, and thus not included in this list. We may look at those in future.

[bjarthur]

+1 if you mean something like this:

StatsBase.trimmean{T}(v::AbstractArray{T}, p, region) = mapslices(x->trimmean(x,p), v, region)

[jamblejoe]

Any progress on this?

[nalimilan]

Now that Julia provides these, it is probably be better to use e.g. skewness.(eachrow(x)) and skewness.(eachcol(x)). The dims=1/dims=2 approach requires adjusting lots of methods and is less explicit.

[cossio]

It would be nice to have functions like skewness(A; dims) taking a dims keyword to reduce over, where A can be a general N-dimensional Array. Any plans to implement this?

[kleinschmidt]

@cossio There's a PR to add a generic eachslice iterator to Base: https://github.com/JuliaLang/julia/pull/32310; it's been under review for a while but there's recent activity...

[nilshg]

This is now merged, time to close this issue?

[jamblejoe]

skewness(rand(10,10), dims=1) doesn't work on 1.9rc1 and latest StatsBase

[bjarthur]

@jamblejoe but you can use eachslice on julia 1.8 to achieve the same:

julia> skewness.(eachslice(rand(10,10), dims=1))
10-element Vector{Float64}:
 -0.357654736525527
 -0.06636393783605732
 -0.20782459433865907
  1.25151036890323
  0.13408296667729347
 -0.4302398373636257
 -0.14363707095203634
 -0.00889806440269636
 -0.6370034546305223
  0.5832296811450159

[jamblejoe]

@bjarthur works thanks! Can be closed from my side