K tree 0.4.2

Lance De Vine, Chris De Vries, Shlomo Geva, Ulf Grossekathofer — Mon, 04 Jul 2011 06:01:59 -0000

K-tree is a tree structured clustering algorithm. It is also refered to as a Tree Structured Vector Quantizer (TSVQ). The goal of cluster analysis is to group objects based on similarity. Each object in a K-tree is represented by an n-dimensional vector. All vectors in the tree must have the same number of dimensions. At the K-tree 0.1 release the only similarity measure for vectors is Euclidean distance.

The algorithm is a hybrid of the B+-tree and k-means algorithms. It uses a similar tree structure to the B+-tree and uses k-means to perform splits. The tree forms a nearest neighbour search tree. Unlike k-means the number of clusters does not need to be specified upfront. However, a tree order must be specified that restricts how many vectors can be stored in any node. Each level of the tree produces a different number of clusters.

mloss.org K tree

K tree 0.4.2