Figure 2

Bootstrapping: Defining conflicts. After the clustering method is applied to a given data set, a membership vector m is inferred (3 clusters, A, D, C, in this example) from which a set of bipartitions B can be deduced. By re-sampling with replacement (bootstrapping) the data set, we obtain n replicates which give us a set of bipartitions B^′ (left-hand side). When a single bipartition from B is compared against B^′, a conflict (C) or non-conflict (NC) is inferred (right-hand side). This procedure is repeated for all clusters in the original m, giving support for to each cluster as the percentage of non-conflicts for that cluster.