Formula for mpe()

[FelicienLL]

Hello,

I'm not sure the implementation of Mean Percentage Error is correct there:

yardstick/R/num-mpe.R

Line 113 in 8d41dc5

error <- (truth - estimate) / truth

In my opinion it should be calculated by doing estimate - truth instead of truth - estimate. See below, the estimator clearly over-estimates the observations, yet a negative bias is calculated. Is this running as expected ?

df <- data.frame(
  actual = c(1, 2, 3, 4), 
  prediction = c(1.1, 2.2, 3.3, 4.4)
)

yardstick::mpe(df, truth = actual, estimate = prediction)
#> # A tibble: 1 x 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 mpe     standard       -10.0

In addition, I would divide the results with abs(truth) instead of truth to preserve the sign of the difference if the actual value is negative.

Thanks in advance.

Félicien

[EmilHvitfeldt]

The formula follows the equation from wikipedia https://en.wikipedia.org/wiki/Mean_percentage_error. This isn't well reflected in the documentation, so I'm going to keep this issue open to deal with that