Infrastructure for Forecasting and Assessment of Probabilistic Models

The R package topmodels provides unified infrastructure for probabilistic models and distributional regressions:

• procast: Probabilistic forecasting (in-sample and out-of-sample) of probabilities, densities, quantiles, and moments.
• proresiduals: Probabilistic residuals, e.g., (randomized) quantile residuals, PIT or Pearson residuals.
• proscore: Probabilistic scoring rules, e.g., log-score (or log-likelihood), (continuous) ranked probability score, etc.
• rootogram, pithist, qqrplot, reliagram: Diagnostic graphics like rootograms, PIT histograms, (randomized) quantile residual Q-Q plots, and reliagrams (reliability diagrams).

Modular object-oriented implementation with support for many model objects, including lm, glm, glm.nb, gamlss, bamlss, hurdle, zeroinfl, zerotrunc, nbreg, crch, betareg, and more to come.