library(risk3r)
N <- 10000
predicted <- runif(N)
actual <- rbinom(N, size = 1, prob = predicted)
metrics(actual, predicted)
#> ℹ Creating woe binning ...
#> # A tibble: 1 × 4
#> ks auc iv gini
#> <dbl> <dbl> <dbl> <dbl>
#> 1 0.509 0.835 1.77 0.671Remove NA if are present.
predicted2 <- predicted
predicted2[sample(c(TRUE, FALSE), size = N, prob = c(1, 99), replace = TRUE)] <- NA
metrics(actual, predicted2)
#> 92 of 10000 'predicted' values are NAs, they will be ignorated
#> ℹ Creating woe binning ...
#> # A tibble: 1 × 4
#> ks auc iv gini
#> <dbl> <dbl> <dbl> <dbl>
#> 1 0.509 0.828 1.75 0.655Using a glm model
daux <- data.frame(actual = actual, predicted = predicted)
m <- glm(actual ~ predicted, family = binomial, data = daux)
model_metrics(m)
#> ℹ Creating woe binning ...
#> # A tibble: 1 × 4
#> ks auc iv gini
#> <dbl> <dbl> <dbl> <dbl>
#> 1 0.509 0.835 1.76 0.671Works with new data.
new_daux <- head(daux, 1000)
model_metrics(m, newdata = new_daux)
#> ℹ Creating woe binning ...
#> # A tibble: 1 × 4
#> ks auc iv gini
#> <dbl> <dbl> <dbl> <dbl>
#> 1 0.490 0.820 1.76 0.641