Metrics 0.3.0 silhouette

Lodestar.Metrics 0.3.0. 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.

Silhouette

The only metric in this namespace that needs the samples themselves rather than a reference partition — which is what makes it the one you can use when no truth exists, to choose how many clusters to ask for.

Each sample is scored on how much closer it sits to its own cluster than to the nearest other one: 1 means comfortably inside, 0 means on the boundary, and a negative value means it would be better off elsewhere. The score of a clustering is the mean of those.

Two ways in, one computation. Hand it the samples and it uses the euclidean distance; hand it a distance matrix you already computed and it uses that. The frozen corpus checks the two agree at 1e-9 on every case — not bit for bit, because the distances themselves are computed differently on each side.

Members

Member	What it does
Silhouette.PerSample	The score of each sample, from the samples themselves.
Silhouette.PerSampleFromDistances	The score of each sample, from a distance matrix.
Silhouette.Score	The mean over every sample, from the samples themselves.
Silhouette.ScoreFromDistances	The mean over every sample, from a distance matrix.