scores_mixnorm: Calculating scores for a mixture of normal distributions.

In FK83/scoringRules: Scoring Rules for Parametric and Simulated Distribution Forecasts

Calculating scores for a mixture of normal distributions.

Description

Calculating scores for a mixture of normal distributions.

Usage

crps_mixnorm(y, m, s, w = NULL)

crps_mixnorm_int(y, m, s, w = NULL, rel_tol = 1e-06)

logs_mixnorm(y, m, s, w = NULL)

dss_mixnorm(y, m, s, w = NULL)

Arguments

y	vector of observations.
m	matrix of mean parameters; rows represent observations, columns represent mixture components.
s	matrix of scale parameters; same structure as m.
w	optional; matrix of non-negative weights; same structure as m.
rel_tol	relative accuracy for numerical integration.

Details

logs_mixnorm and crps_mixnorm calculate scores via analytical formulas. crps_mixnorm_int uses numerical integration for the CRPS; this can be faster if there are many mixture components (i.e., if m, s and w have many columns). See examples below.

Value

A vector of score values.

Examples

# Example 1: 100 observations, 15 mixture components
mval <- matrix(rnorm(100*15), nrow = 100)
sdval <- matrix(rgamma(100*15, shape = 2), nrow = 100)
weights <- matrix(rep(1/15, 100*15), nrow = 100)
y <- rnorm(100)
crps1 <- crps_mixnorm(y = y, m = mval, s = sdval, w = weights)
crps2 <- crps_mixnorm_int(y = y, m = mval, s = sdval, w = weights)

## Not run: 
# Example 2: 2 observations, 10000 mixture components
mval <- matrix(rnorm(2*10000), nrow = 2)
sdval <- matrix(rgamma(2*10000, shape = 2), nrow = 2)
weights <- matrix(rep(1/10000, 2*10000), nrow = 2)
y <- rnorm(2)
# With many mixture components, numerical integration is much faster
system.time(crps1 <- crps_mixnorm(y = y, m = mval, s = sdval, w = weights))
system.time(crps2 <- crps_mixnorm_int(y = y, m = mval, s = sdval, w = weights))

## End(Not run)

FK83/scoringRules documentation built on Feb. 20, 2024, 8:01 p.m.