Lodestar.Cluster 0.1.0. This page is frozen at that release. Read the current documentation for what main says now. A link to a decision or a migration page follows main, and leaves the archive.

Assigns unseen samples to the fitted centres.

public int[] Predict(ReadOnlySpan<double> samples)

Parameters — samples is the matrix to assign, row-major, with FeatureCount values per row.

Returns — one cluster index per row.

Exceptions — ArgumentException when samples holds no row, or a partial one.

Example — rows the fit never saw.

using Lodestar.Cluster;

double[] samples = [0.0, 0.0, 0.0, 1.0, 10.0, 10.0, 10.0, 11.0, 5.0, 5.0];
KMeans model = KMeans.Fit(samples, featureCount: 2, clusterCount: 3,
    new KMeansOptions { InitialCentres = [0.0, 0.0, 10.0, 10.0, 5.0, 5.0] });

int[] unseen = model.Predict([0.5, 0.5, 9.5, 10.5]);

int low = unseen[0];   // => 0
int high = unseen[1];  // => 1

Remarks — nothing is refitted: this is the assignment step alone, over the centres already found. Passing the fitted samples back reproduces Labels exactly, which is not a coincidence but the contract — when the loop stops on the centre shift rather than on the labels, a final assignment runs so the two agree.

Applies to — net10.0, netstandard2.0.

See also — KMeans.Fit, KMeans.