Skip to contents

Define a custom predictive measure

Source: R/pred_measure.R
custom_measure.Rd

Attaches the name, the orientation, and the standard error of the difference to a measure function. Pass the result to the measures argument of the *_pred_measure() functions.

Usage

custom_measure(fun, name, se_diff_fun = NULL, loss = FALSE)

Arguments

fun

A function that returns estimate, se, and pointwise. Only the arguments y, ypred, mupred, ylp, and log_weights that are in its signature are supplied.

name

A character string. The measure is reported under this name.

se_diff_fun

How model_compare() computes the standard error of a difference: a function (ref, cmp), "sum", "mean", or NULL. With NULL (the default), the standard error of the difference is NA.

loss

TRUE if lower values are better. The default FALSE treats the measure as a utility.

Value

fun with the attributes measure_name, measure_loss, and measure_se_diff.

See also

pred_measure(), model_compare()

Examples

my_abs_err <- custom_measure(
  function(y, mupred) {
    pw <- abs(y - colMeans(mupred))
    list(estimate = mean(pw), se = sd(pw) / sqrt(length(pw)), pointwise = pw)
  },
  name = "my_abs_err", se_diff_fun = "mean", loss = TRUE
)