Metrics 0.3.0 poissondeviance

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.

PoissonDeviance

The deviance for a count: how far a prediction is from the truth when the target is a number of events and its variance grows with its mean. TweedieDeviance at power = 1, which is how scikit-learn defines it too.

A type of its own because the reference exposes one, and because a caller counting claims, visits or failures should not have to know that 1 is the Poisson. The refusal sentence still names the power, as scikit-learn's does.

A zero truth is legal; a zero prediction is not. Counting zero events is ordinary data, and the deviance has a finite limit there — y × log(y / ŷ) is taken as 0 at y = 0. Predicting zero events is not: the logarithm has no value, and the reference refuses it rather than returning an infinity.

Members

Member	What it does
PoissonDeviance.Score	The mean deviance for a count target.