hclust_in_R

This is a basic implementation of hierarchical clustering written in R. It produces output structured like the output from R's built in hclust function in the stats package.

I wrote these functions for my own use to help me understand how a basic hierarchical clustering method might be implemented. It's very short (the clustering function hc is less than 50 lines, including comments) and easy to experiment with, and I think it could easily be extended to other variations on hierarchical clustering.

This implementation is less efficient than R's hclust (written in Fortan), both in terms of memory use and performance. But it's easy to read and experiment with.

The clustering method is very short and proceeds somewhat differently than R's default method, but produces output compatible with the functions associated with hclust.

The plotting methods for hclust output use an ordering vector to lay the clusters out nicely and avoid branch crossings. That code is translated verbatim from the original Fortran in the iorder function presented here. It was originally written by C F. Murtagh, ESA/ESO/STECF, Garching, June 1991.

I include an inefficient but illustrative Euclidean distance matrix function that is different than R's built in dist function. The clustering method presented here can used either a full distance matrix or a sparse lower- or upper-triangular one.

See the bottom of the source code file for an example.

Example

Try this:

source("hc.R")
h  = hclust(USArrests, "single")    # standard method
h1 = hc(USArrests, "single")        # our example method

And compare:

plot(h)

``` plot(h1) ```