scoringutils (version 2.1.0)

get_metrics.forecast_point: Get default metrics for point forecasts

Description

For point forecasts, the default scoring rules are:

• "ae_point" = ae()

• "se_point" = se()

• "ape" = ape()

A note of caution: Every scoring rule for a point forecast is implicitly minimised by a specific aspect of the predictive distribution (see Gneiting, 2011).

The mean squared error, for example, is only a meaningful scoring rule if the forecaster actually reported the mean of their predictive distribution as a point forecast. If the forecaster reported the median, then the mean absolute error would be the appropriate scoring rule. If the scoring rule and the predictive task do not align, the results will be misleading.

Failure to respect this correspondence can lead to grossly misleading results! Consider the example in the section below.

Usage

# S3 method for forecast_point
get_metrics(x, select = NULL, exclude = NULL, ...)

Arguments

x

A forecast object (a validated data.table with predicted and observed values, see as_forecast_binary()).

select

A character vector of scoring rules to select from the list. If select is NULL (the default), all possible scoring rules are returned.

exclude

A character vector of scoring rules to exclude from the list. If select is not NULL, this argument is ignored.

...

unused

Input format

Overview of required input format for binary and point forecasts

References

Making and Evaluating Point Forecasts, Gneiting, Tilmann, 2011, Journal of the American Statistical Association.

See Also

Other get_metrics functions: get_metrics(), get_metrics.forecast_binary(), get_metrics.forecast_nominal(), get_metrics.forecast_ordinal(), get_metrics.forecast_quantile(), get_metrics.forecast_sample(), get_metrics.scores()

Examples

get_metrics(example_point, select = "ape")

library(magrittr)
set.seed(123)
n <- 500
observed <- rnorm(n, 5, 4)^2

predicted_mu <- mean(observed)
predicted_not_mu <- predicted_mu - rnorm(n, 10, 2)

df <- data.frame(
  model = rep(c("perfect", "bad"), each = n),
  predicted = c(rep(predicted_mu, n), predicted_not_mu),
  observed = rep(observed, 2),
  id = rep(1:n, 2)
) %>%
  as_forecast_point()
score(df) %>%
  summarise_scores()