Random Segmentation

This is an implementation of the algorithm discussed in Section 4.12. It accepts the following parameters:

• numTrials: The number of random trials to run. Default value is 1000.
• minCent: The minimum number of centroids to consider. Default value is 2.
• maxCent: The maximum number of centroids to consider. Default value is 8.
• clustBias: The algorithm currently selects a random number between minCent and maxCent based on a linear probability distribution. In other words, the probability of choosing $n$ centroids is $(n-\mathit{minCent})$ times the probability of choosing $minCent$ centroids. Strictly speaking, this is incorrect, since there are far more subsets with, 8 instances than there are with 2 instances. In fact, it follows a binomial distribution - and for cases where there are far more instances than centroids, it approximates a factorial distribution. To compensate for this, one can set a cluster bias $b$, so that the probability of choosing a subset with $n$ instances is proportional $(n-\mathit{minCent})^{b}$. Default value is 1 (i.e. linear).
• dispMeasure: The disparity measure to use. Possible values are gainratio, gain and chisquare.