Metrics jaccardscore

Development build. This page describes main, not a released package. The latest published Lodestar.Metrics is 0.3.0 — read its documentation.

JaccardScore

Intersection over union: of every sample that is in a class or was predicted into it, what share is in both.

It is Precision's numerator over a larger denominator — precision divides the true positives by what was predicted, Recall divides them by what was true, and this divides them by the two together. So it can never read above either, which is what makes it the strictest of the three and a test asserts.

It takes the same four Averaging modes and the same ZeroDivision as precision and recall, because it is the same shape with a different ratio — the reason issue #211 called it the cheapest of its six.

Labels only, where Precision and its two siblings also read a ConfusionMatrix directly. Those overloads exist because ClassificationReport reads them; nothing reads a Jaccard coefficient from a report, and jaccard_score has no matrix form of its own.

A class neither side carries

Nothing is in the union, so the ratio has no value. ZeroDivision.Zero — the default — answers 0, and One answers 1; both are what zero_division=0 and zero_division=1 give.

NaN and Throw have no counterpart here. jaccard_score admits only 0, 1 and 'warn', and refuses nan outright with an InvalidParameterError. The two extra members are this package's, and the equivalence table says so.

Reaching that case needs an explicit labels set: a class that occurs in neither input is not in the sorted union the label set otherwise defaults to.

Members

Member	What it does
JaccardScore.Score	The coefficient, reduced by an Averaging mode.
JaccardScore.PerClass	One coefficient per class, in label order.