Cluster Methods

CBPtools currently supports three clustering algorithms: k-means, spectral, and agglomerative clustering. The k-means clustering algorithm is the default, hence when using the cbptools example directive, an example file will be generated containing the k-means clustering default options. Below are examples of how the different clustering algorithms can be defined, each with their own set of unique options.

KMeans clustering

...

parameters:
    clustering:
        method: kmeans
        n_clusters: [2, 3, 4, 5]
        cluster_options:
            algorithm: auto
            init: k-means++
            max_iter: 10000
            n_init: 256

...

Spectral clustering

...

parameters:
    clustering:
        method: spectral
        n_clusters: [2, 3, 4, 5]
        cluster_options:
            n_init: 256
            kernel: nearest_neighbors
            assign_labels: kmeans
            eigen_solver: arpack
            eigen_tol: 1.0e-5

...

Agglomerative clustering

...

parameters:
    clustering:
        method: agglomerative
        n_clusters: [2, 3, 4, 5]
        cluster_options:
            distance_metric: euclidean
            linkage: ward

...