fast_accuracies: Fast accuracy measure computation

In JSzitas/soothsayer: Tidy Time Series Forecast Model Selection

Fast accuracy measure computation

Description

A pretty fast way to compute commong accuracy measures

Usage

rmsse(resid_sqr, scale, na.rm = TRUE)

mase(resid, scale, na.rm = TRUE)

mpe(scl_prc_resid, na.rm = TRUE)

mape(scl_prc_resid, na.rm = TRUE)

maape(scl_prc_resid, na.rm = TRUE)

me(resid, na.rm = TRUE)

rmse(resid_sqr, na.rm = TRUE)

mae(resid, na.rm = TRUE)

acf1(resid, na.rm = TRUE, demean = TRUE)

fast_measure(resid, actual, scale = 1, na.rm = TRUE)

fast_accuracy(
  fable,
  test,
  train = NULL,
  measures = fast_measure,
  across = c(".model", "h"),
  ...
)

Arguments

resid_sqr	Squared residuals
scale	Scaling factor - typically mean of training data
na.rm	Whether to remove NA values - defaults to **TRUE**
resid	Residuals
scl_prc_resid	Scaled percentage residuals
demean	Whether to demean before calculating autocorrelations - defaults to **TRUE**
actual	Actuals
fable	The model fable - we do not work with mables
test	The test data corresponding to forecasts in the fable
train	The original data used in fitting the models - if available, scaled accuracy measures will be computed by default.
measures	Currently only for future extensions, though **fast_measure** does everything that was required so far
across	The grouping variables across which to compute measures - by default just the ".model" column, but you can easily pass something like "h" and it will work.
...	For future extensions :)

Value

A data.frame with computed accuracy measures, grouped by whatever was supplied in **across** and the keys present in **test**.

JSzitas/soothsayer documentation built on April 18, 2023, 12:59 a.m.