Object: Clustering

The clustering procedure uses the $k$-means algorithm to cluster the snapshots into the number of clusters specified by NonlinearRomFileSystem.NumClusters.

under Clustering {
    UseExistingClusters = useexistingclusters-flag;
    PercentOverlap = percentoverlap-real;
    KMeansMaxIterations = kmeansmaxit-int;
    KmeansRandomSeed = kmeansrandseed-int;
    MinClusterSize = minclustersize-int;
}

with

useexistingclusters-flag [False]:
When set to True, the computation of clusters is skipped and an existing clustering from disk is used.

True
Clustering computations are skipped, using instead an existing clustering stored in the NonlinearRomFilesystem.

False
Clustering computations are performed and a new clustering is constructed and written to disk.

percentoverlap-real [10.0]:
After clustering the state snapshots, the initially disjoint clustering can be adjusted with the goal of sharing snapshots close to cluster boundaries between the two neighboring clusters. This value specifies a percentage of each neighbor's clusters to add to produce overlap.

kmeansmaxit-int [100]:
Maximum number of iterations for the $k$-means clustering algorithm.

kmeansrandseed-int [–]:
Random seed provided to select the initial cluster centroids for the $k$-means algorithm. This is useful if seeking to reproduce a previously computed clustering for testing purposes.

minclustersize-int [1]: Any clusters at the termination of $k$-means with size less than minclustersize-int will be assimilated into the nearest neighboring cluster.