tails_ev: Extreme value distirbutions for distribution tails
In jbrowell/ProbCast: Flexible Probabilistic Forecasting

tails_ev

R Documentation

Extreme value distirbutions for distribution tails

Description

This function fits semi-parametric models for extreme value distributions using evgam with facilities for cross-validation.

Usage

tails_ev(
  data,
  mqr_data,
  tail_starts = range(as.numeric(gsub("q", "", names(mqr_data))), na.rm = T),
  formula,
  formula_r = formula,
  CVfolds = NULL,
  BadData_col = NULL,
  evgam_family = "gpd",
  print_summary = F,
  return_models = F
)

Arguments

`data`	The modelling table as a `data.frame` or `data.table`
`mqr_data`	A `MultiQR` object corresponding to `data`
`tail_starts`	The upper and lower probability levels beyond which the tail distribution is used. E.g. `c(5,95)` to use tail distribution for probabilities below quantile 0.05 and above 0.95. Defaults to min and max from `mqr_data`
`formula_r`	as `formula` but for the upper tail only.
`CVfolds`	Control for cross-validation if not supplied in `data`.
`BadData_col`	Name of a boolean column in `data` indicating bad data which should be excluded from parameter estimation.
`evgam_family`	specifies distribution, see `?evgam`.
`print_summary`	If `TRUE`, summary is printed for each evgam fit.
`return_models`	If `TRUE`, the function returns a list including both the data and the fitted data models.
`formala`	A `formula` object with the response on the left of an ~ operator, and the terms, separated by + operators, on the right. This formula is used for both upper and lower tails unless `formula_r` is specified in which case `formula` is used for the lower tail only.

Details

The returned predictive quantiles are those produced out-of-sample for each cross-validation fold (using models trained on the remaining folds but not "Test" data). Predictive quantiles corresponding to "Test" data are produced using models trained on all non-test data.

Value

If 'return_models' is TRUE, then returns a list containing a list of tail models, and data with additional columns containing the predicted parameters of the specified tail distribution. Otherwise, just returns data.