divControl is a novel method to smoothly control the diversity of a cluster ensemble, thus providing an effective way to study the diversity influence on consensus performance. Given a data set, the diversity of the ensemble is smoothly changed by using the full ensemble grouping algorithm and a consensus function based on evidence accumulation. Once it finishes, it is possible to download the ensemble with groups of partitions and the ensembles with controlled diversity. Also, it will show the evolution of several diversity measures and the Average Normalized Mutual Information (ANMI) of the final consensus partition. A comparison with a traditionally used method (random) is shown at the left subfigure.
Note: depending on the parameters you choose (particularly the length of the range for h), the web-demo could take more than 3 minutes to finish.
Contact: Milton Pividori