scoringRules
An R package to compute scoring rules for fixed (parametric) and simulated forecast distributions. Authored by Alexander Jordan (Heidelberg Institute for Theoretical Studies (HITS)), Fabian Krüger (Karlsruhe Institute of Technology (KIT)), Sebastian Lerch (KIT and HITS) and Sam Allen (ETH Zürich), with contributions from Maximiliane Graeter (KIT).
Highlights
- Coherent, dictionary-like reference for computing scoring rules in a wide range of situations
- Previously unavailable closed-form expressions of the CRPS for many parametric distributions
- Efficient implementation thanks to R and Rcpp
- Whenever more than one implementation variant exists, we offer statistically principled default choices
Installation
CRAN version:
install.packages("scoringRules")Development version (GitHub):
# install.packages("devtools")
library(devtools)
install_github("FK83/scoringRules")Background
Scoring rules are functions S(F, y) which evaluate the accuracy of a forecast distribution F, given that an outcome y was observed. The scoringRules package contains functions to compute scoring rules, for a variety of distributions F that come up in applied work, and several choices of S. Two main classes of distributions are
- Parametric distributions like normal, t, and gamma. For example, most weather forecasts (which apply statistical postprocessing to physical models) take such a form.
- Distributions that are not known analytically, but are indirectly described through a sample of simulaton draws. For example, Bayesian forecasts produced via Markov Chain Monte Carlo (MCMC) take this form.
We cover various scoring rules, including
- the continuous ranked probability score and the logarithmic score
- the energy and variogram scores for multivariate forecast distributions given by discrete samples
- weighted scoring rules for (univariate or multivariate) forecast distributions given by discrete samples
Please refer to the package vignettes 'Evaluating Probabilistic Forecasts with scoringRules' and 'Weighted scoringRules' for details and references.
History
- May 2023: Version 1.1, including threshold and outcome weighted scoring rules (Sam Allen)
- August 2019: Version 1.0.0 on CRAN, vignette published in the Journal of Statistical Software
- November 2017: Version 0.9.4 on CRAN, including a detailed vignette
- July 2017: Vignette for closed-form expressions of the CRPS (Alexander Jordan), and functions for CRPS-based fitting of truncated/censored distributions
- July 7, 2016: Version 0.9 published on CRAN
- September 15, 2014: First commit