Metrics 0.3.0 hammingloss

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.

HammingLoss

The fraction of labels predicted wrongly.

On single-label input that is one minus Accuracy, and it agrees with ZeroOneLoss exactly. On a label matrix the three part company, and the difference is the reason both of these exist: this counts wrong labels where ZeroOneLoss counts wrong rows, so one wrong label out of three costs a third of a sample here and a whole sample there.

Measured on two samples over three labels with two labels wrong: 0.3333… here and 1 there.

Members

Member	What it does
HammingLoss.Score	The share of labels that disagree, from labels or from a matrix.