R ClustOfVar cutree height error

Question

ClustOfVar is a package for clustering columns.

mtcars[] = lapply(mtcars,as.character)

fit = ClustOfVar::hclustvar(X.quali  = mtcars)

labels = cutree(fit,h=0.5)

But when I cutree for some fits, I get the following message.

Warning: Error in cutree: the 'height' component of 'tree' is not sorted (increasingly)

The data on which this problem occurs is both private and big. The fit is generated successfully. I will try my best to find a reproducible data example but if you have met this kind of problem before, please let me know how to deal with it.

Answer 1

With certain linkage methods, it can happen that there is a non-monotonicty in the tree. Then this error occurs. If I'm not mistaken, centroid linkage is particularly prone to this.

The problem looks like this: C is created from A and B. Subtree A is created at height x. Subtree B is created at height y. Usually A and B would merge at a height such that z > x and z > y. But with some linkages, this property may be violated and eg z < x. Now if we want to cut the tree at height h with z < h < x, then we have the contradiction that subtree C should be merged (becaude z < h) but A should be split (because h < x) - but A is part of C.