QT_CLUST
============
This provides an efficient C implementation of the QT Clust algorithm (Heyer,
Kruglyak, and Yooseph, 1999). This is a partitional clustering algorithm,
similar to k-means, but has the advantage of not requiring a number of clusters
to be specified a priori.
The algorithm has two stages:
1. Clustering: A candidate cluster is built, starting with each vector in the
population. This is done by adding the next closest agent to the cluster
until the diameter threshold is reached. This is done in parallel for a
significant speed increase.
2. Filtering: The largest candidate cluster that does not overlap with a
previously selected cluster is selected as a final cluster, until no
viable candidates remain. After filtering, unclustered elements are then
reclustered within the remaining population until all elements are
classified.
The filtering step amortizes the time complexity of the original QT-Clust
algorithm, while maintaining the quality control advantages of the algorithm.
These implementations advantages are being prepared for a paper by Murdock and
Yaeger, with applications in the artificial life domain.
Laurie J. Heyer, Semyon Kruglyak, and Shibu Yooseph. Exploring expression data:
Identification and analysis of coexpressed genes. Genome Research, 9:1106-1115,
1999. http://dx.doi.org/10.1101/gr.9.11.1106
The R implementation of the original algorithm is part of the CRAN package
flexclust: http://cran.r-project.org/web/packages/flexclust/index.html