Providing additional intrinsic evaluation metrices

[jaksle]

This is more of a question than an issue. I wanted to time series clustering using Clustering.jl and I needed intrinsic evaluation measures in order to determine optimal number of clusters. Clustering.jl offers only silhouettes in this regard so I implemented myself popular ones: Calinski-Harabasz, Davies-Boulden, Xie-Beni indices. Would it be useful if I add them to Clustering.jl?

The formulas are easy so it would not require a lot of work, it's just that first I would like to be sure there is an interest in that.

[nomadbl]

Not a maintainer here or anything, but sounds interesting to me 😄

[alyst]

Sure, this sounds very useful! One constraint, though, is that we don't want Clustering.jl to be a super-package for anything related to unsupervised learning as that would complicate maintenance and unnecessary bloat user projects.
So new features that would require extra dependencies, esp. outside of github.com/JuliaXXX ecosystem, or features that add a lot of new relatively complex code, should rather be implemented as separate packages.

If these new metrics fit the above criteria, the PR is welcome! :)

[jaksle]

No new dependencies needed, no complex code, these quality indices have definitions simpler than silhouettes. Bloat is not my goal, but the few most popular indices for hard and soft clustering I hope would come in handy.

So great, now I am busy but next month I will try check and adjust my code :)

[alyst]

Implemented in #257