Why only 4/8 methods?

[vsandul]

Hi! First of all, thanks a lot for the library - it seems it will very helpful for my job! But I have a question - why there are only 4 methods instead of 8 which can be founded in Mullner's package?

[cdalitz]

The problem with the remaining methods ("ward" and "centroid") is that they do not work with the matrix of distances as input, but its square (for the "ward" method this is even more complicated, that's why there are three (!) "ward" methods). This means, that an end user will most likely get incorrect results when using these methods, unless (s)he is very knowledgeable about the internal implementation of these methds.

Actually, I was even reluctant to exposing "median", because the implementation only works with Euclidean distances, which will mean that some users will scratch their heads due to the surprising/incorrect results of the "median" method. Nevertheless, it might be useful in some instances, so I included it.

Concerning the confusion about the "ward" method, see

Murtagh, Legendre: "Ward’s Hierarchical Agglomerative Clustering Method: Which Algorithms Implement Ward’s Criterion?" Journal of Classification 31:274-295 (2014)