crps_sample: (Continuous) ranked probability score

Description

Wrapper around the crps_sample() function from the scoringRules package. Can be used for continuous as well as integer valued forecasts

The Continuous ranked probability score (CRPS) can be interpreted as the sum of three components: overprediction, underprediction and dispersion. "Dispersion" is defined as the CRPS of the median forecast $m$. If an observation $y$ is greater than $m$ then overpredictoin is defined as the CRPS of the forecast for $y$ minus the dispersion component, and underprediction is zero. If, on the other hand, $y<m$ then underprediction is defined as the CRPS of the forecast for $y$ minus the dispersion component, and overprediction is zero.

The overprediction, underprediction and dispersion components correspond to those of the wis().

Usage

crps_sample(observed, predicted, separate_results = FALSE, ...)
dispersion_sample(observed, predicted, ...)
overprediction_sample(observed, predicted, ...)
underprediction_sample(observed, predicted, ...)

Value

Vector with scores.

dispersion_sample(): a numeric vector with dispersion values (one per observation).

overprediction_quantile(): a numeric vector with overprediction values (one per observation).

underprediction_quantile(): a numeric vector with underprediction values (one per observation).

Arguments

observed: A vector with observed values of size n
predicted: nxN matrix of predictive samples, n (number of rows) being the number of data points and N (number of columns) the number of Monte Carlo samples. Alternatively, predicted can just be a vector of size n.
separate_results: Logical. If TRUE (default is FALSE), then the separate parts of the CRPS (dispersion penalty, penalties for over- and under-prediction) get returned as separate elements of a list. If you want a data.frame instead, simply call as.data.frame() on the output.
...: Additional arguments passed on to crps_sample() from functions overprediction_sample(), underprediction_sample() and dispersion_sample().

Input format

Overview of required input format for sample-based forecasts

References

Alexander Jordan, Fabian Krüger, Sebastian Lerch, Evaluating Probabilistic Forecasts with scoringRules, https://www.jstatsoft.org/article/view/v090i12

Examples

Run this code

observed <- rpois(30, lambda = 1:30)
predicted <- replicate(200, rpois(n = 30, lambda = 1:30))
crps_sample(observed, predicted)

Run the code above in your browser using DataLab