Clustering performance for different space partitionings and metrics.

To assess the quality of a tree, the distance μ between two recordings is the number of edges in-between. The shortest distance between two recordings of the same gene is thus 2. A well-balanced tree avoids long branches and should minimize the radius r = max(⌠(vi,vj)). The average closest distance between recordings of the same time is found to be d = E[min(⌠(vi,vj))], vi~ vj. To find the expected distance in a random tree, the expression becomes only D = E[min(⌠(vi,vj))]. Both of these values were calculated by bootstrapping. Finally, to compare the quality of trees, the ratio q = (d-2)/(D/2) should be small. The best values are highlighted in bold.
Clustering performance for different space partitionings and metrics.