pinball: Pinball Score for 'MultiQR' Objects

In jbrowell/ProbCast: Flexible Probabilistic Forecasting

Pinball Score for MultiQR Objects

Description

This function calculates the pinball score for each quantile in a MultiQR object. Optionally, results are produced by cross-validation fold or covariate, 95% confidence intervals are estimated via bootstrap, and results are plotted.

Usage

pinball(
  qrdata,
  realisations,
  kfolds = NULL,
  subsets = NULL,
  breaks = 4,
  bootstrap = NULL,
  plot.it = T,
  ...
)

Arguments

qrdata	MultiQR object.
realisations	Vector of realisations corresponding to rows of qrdata. NA accepted where realisations are missing.
kfolds	Optional vector of fold/test labels corresponding to rows of qrdata. Cannot be used with subsets.
subsets	Optional vector of covariates to bin data by. Breaks between bins are the empirical quantiles of subsets by default or all unique factors or charater strings. Custom breaks may be specifed, see breaks. Cannot be used with kfolds.
breaks	Either the number of quantiles to use to bin subsets by (resulting in breaks+1 bins, defaults to breaks=4), or, if length(breaks) > 1, a vector of spcific break points. Only used if subsets provided.
bootstrap	Number of boostrap samples used to generate 95% confidence intervals.
plot.it	boolean. Make a plot?
...	Additional arguments passed to plot().

Details

Missing values in realisations are handled by na.rm=T when calculating average exceedence of a given quantile.

Value

Quantile Score data and, if plot.it=T, a plot.

Author(s)

Jethro Browell, [e-mail]

jbrowell/ProbCast documentation built on July 20, 2024, 1:53 p.m.