Metrics gammadeviance score

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

GammaDeviance.Score

The mean gamma deviance — sklearn.metrics.mean_gamma_deviance.

public static double Score(ReadOnlySpan<double> yTrue, ReadOnlySpan<double> yPred, ReadOnlySpan<double> sampleWeight = default)

Parameters — yTrue and yPred are the true and predicted values, of the same length, and both must be strictly positive. sampleWeight is one weight per sample, or empty — the default — to weight every sample by 1.

Returns — double, 0 for a perfect prediction and larger the worse it is. Unbounded above, and unchanged by rescaling both arguments together.

Exceptions — ArgumentException when the lengths disagree, the input is empty or holds a non-finite value, or either operand is not strictly positive — including a zero truth, which PoissonDeviance.Score accepts and this does not. The message is scikit-learn's, naming power=2.

Example — the same four samples the Poisson page scores, read as a positive quantity instead.

using Lodestar.Metrics;

double[] truth = [1.0, 2.0, 3.0, 4.0];
double[] predicted = [1.5, 2.5, 2.0, 4.5];

double deviance = GammaDeviance.Score(truth, predicted);  // => 0.0982…

Scaling both by ten leaves it alone, which is the property the type exists for:

using Lodestar.Metrics;

double[] truth = [10.0, 20.0, 30.0, 40.0];
double[] predicted = [15.0, 25.0, 20.0, 45.0];

double scaled = GammaDeviance.Score(truth, predicted);  // => 0.0982…

Remarks — identical to TweedieDeviance.Score at power: 2.0, asserted across the whole frozen corpus.

Applies to — net10.0, netstandard2.0.

See also — PoissonDeviance.Score, TweedieDeviance.Score, D2Tweedie.Score, the Python equivalence table.