MetricsWeighted (version 0.3.0)

deviance_tweedie: Tweedie Deviance

Description

Weighted average of (unscaled) unit Tweedie deviance with parameter p. This includes the normal deviance (p = 0), the Poisson deviance (p = 1), as well as the Gamma deviance (p = 2), see [1] for a reference and https://en.wikipedia.org/wiki/Tweedie_distribution for the specific deviance formula. For 0 < p < 1, the distribution is not defined. The smaller the deviance, the better.

Usage

deviance_tweedie(actual, predicted, w = NULL, tweedie_p = 0, ...)

Arguments

actual

Observed values.

predicted

Predicted values.

w

Optional case weights.

tweedie_p

Tweedie power.

...

Further arguments passed to weighted_mean.

Value

A numeric vector of length one.

References

[1] Jorgensen, B. (1997). The Theory of Dispersion Models. Chapman & Hall/CRC. ISBN 978-0412997112.

See Also

deviance_normal, deviance_poisson, deviance_gamma.

Examples

# NOT RUN {
deviance_tweedie(1:10, c(1:9, 12), tweedie_p = 0)
deviance_tweedie(1:10, c(1:9, 12), tweedie_p = 1)
deviance_tweedie(1:10, c(1:9, 12), tweedie_p = 2)
deviance_tweedie(1:10, c(1:9, 12), tweedie_p = 1.5)
deviance_tweedie(1:10, c(1:9, 12), tweedie_p = 1.5, w = rep(1, 10))
deviance_tweedie(1:10, c(1:9, 12), tweedie_p = 1.5, w = 1:10)
# }