The best candidate in a selection table

Looks the best-supported candidate up in the metric column, not by row position, and by default considers only rows the table marks as comparable.

Usage

best_candidate(x, n = 1, by = NULL, comparable_only = TRUE)

Arguments

x

An sdmTMB_selection object, or its table.

n

Number of candidates to return. Default 1.

by

Metric to rank by: "cv" (per-observation hold-out log-likelihood), "aic", or NULL (default) to use CV when it is present and AIC otherwise.

comparable_only

Consider only rows with comparable == TRUE? Default TRUE.

Value

A character vector of candidate names, best first.

Details

Reading the winner off the first row couples downstream code to how the table happens to be sorted; a later change of sort key then silently changes which model the document claims to have selected. Use this instead — including to build a cross-validation shortlist from an AIC ladder, via n.

See also

sdmTMB_selection()

Author

Mikko Vihtakari

Examples

tab <- data.frame(
  model = c("a", "b", "c"), AIC = c(1002, 1000, 1010),
  comparable = c(TRUE, TRUE, FALSE)
)
best_candidate(tab, n = 2, by = "aic")
#> [1] "b" "a"