Learning Data Mining with Python - Second Edition

Implementation

Putting all this all together, we can now create a clustering algorithm fitting the scikit-learn interface that performs all of the steps in EAC. First, we create the basic structure of the class using scikit-learn's ClusterMixin.

Our parameters are the number of k-means clusterings to perform in the first step (to create the co-association matrix), the threshold to cut off at, and the number of clusters to find in each k-means clustering. We set a range of n_clusters in order to get lots of variance in our k-means iterations. Generally, in ensemble terms, variance is a good thing; without it, the solution can be no better than the individual clusterings (that said, high variance is not an indicator that the ensemble will be ...