Create an ROC curve from smoothed densities
Usage
compute_smooth_density_roc(
data,
controls,
cases,
along = NULL,
best_weights = c(1, 0.5),
direction = "auto",
...
)Arguments
- data
a dataframe containing densities
- controls, cases
bare column name for the densities of the control group
- along
optional bare column name for the response values
- best_weights
weights for computing the best ROC curve points. Defaults to
c(1, .5), which are the defaults used bypROC::coords().- direction
directionto set for the forpROC::roc(). Defaults to"auto".- ...
additional arguments. Not used currently.
Value
the dataframe is updated with new columns for the .sensitivities,
.specificities, .auc, .roc_row, .is_best_youden and
.is_best_closest_topleft.
Examples
set.seed(100)
x1 <- rnorm(100, 4, 1)
x2 <- rnorm(100, 2, .5)
both <- c(x1, x2)
steps <- seq(min(both), max(both), length.out = 200)
d1 <- dnorm(steps, mean(x1), sd(x1))
d2 <- dnorm(steps, mean(x2), sd(x2))
data <- tibble::tibble(
y = steps,
d1 = d1,
d2 = d2,
outcome = rbinom(200, 1, prob = 1 - (d1 / (d1 + d2))),
group = ifelse(outcome, "case", "control")
)
compute_smooth_density_roc(data, d1, d2)
#> # A tibble: 202 × 12
#> y d1 d2 outcome group .sensitivities .specificities .auc
#> <dbl> <dbl> <dbl> <int> <chr> <dbl> <dbl> <dbl>
#> 1 0.932 0.00423 0.0264 1 case 0.000751 1.00 0.967
#> 2 0.960 0.00460 0.0319 1 case 0.00166 1.00 0.967
#> 3 0.989 0.00499 0.0383 1 case 0.00275 1.00 0.967
#> 4 1.02 0.00542 0.0459 1 case 0.00406 0.999 0.967
#> 5 1.05 0.00587 0.0546 1 case 0.00561 0.999 0.967
#> 6 1.07 0.00636 0.0647 1 case 0.00746 0.999 0.967
#> 7 1.10 0.00689 0.0763 1 case 0.00963 0.999 0.967
#> 8 1.13 0.00745 0.0895 1 case 0.0122 0.999 0.967
#> 9 1.16 0.00806 0.104 1 case 0.0152 0.998 0.967
#> 10 1.19 0.00870 0.121 1 case 0.0186 0.998 0.967
#> # ℹ 192 more rows
#> # ℹ 4 more variables: .roc_row <int>, .direction <chr>, .is_best_youden <lgl>,
#> # .is_best_closest_topleft <lgl>
compute_smooth_density_roc(data, d1, d2, along = y)
#> # A tibble: 202 × 12
#> y d1 d2 outcome group .sensitivities .specificities .auc
#> <dbl> <dbl> <dbl> <int> <chr> <dbl> <dbl> <dbl>
#> 1 0.932 0.00423 0.0264 1 case 0.000751 1.00 0.967
#> 2 0.960 0.00460 0.0319 1 case 0.00166 1.00 0.967
#> 3 0.989 0.00499 0.0383 1 case 0.00275 1.00 0.967
#> 4 1.02 0.00542 0.0459 1 case 0.00406 0.999 0.967
#> 5 1.05 0.00587 0.0546 1 case 0.00561 0.999 0.967
#> 6 1.07 0.00636 0.0647 1 case 0.00746 0.999 0.967
#> 7 1.10 0.00689 0.0763 1 case 0.00963 0.999 0.967
#> 8 1.13 0.00745 0.0895 1 case 0.0122 0.999 0.967
#> 9 1.16 0.00806 0.104 1 case 0.0152 0.998 0.967
#> 10 1.19 0.00870 0.121 1 case 0.0186 0.998 0.967
#> # ℹ 192 more rows
#> # ℹ 4 more variables: .roc_row <int>, .direction <chr>, .is_best_youden <lgl>,
#> # .is_best_closest_topleft <lgl>
# terrible ROC because the response is not present (just the densities)
data_shuffled <- data[sample(seq_len(nrow(data))), ]
compute_smooth_density_roc(data_shuffled, d1, d2)
#> # A tibble: 202 × 12
#> y d1 d2 outcome group .sensitivities .specificities .auc
#> <dbl> <dbl> <dbl> <int> <chr> <dbl> <dbl> <dbl>
#> 1 2.29 0.0963 7.70e- 1 1 case 0.0219 0.997 0.551
#> 2 3.43 0.334 1.66e- 3 0 control 0.0220 0.988 0.551
#> 3 4.40 0.363 1.49e- 8 0 control 0.0220 0.977 0.551
#> 4 6.33 0.0293 2.63e-26 0 control 0.0220 0.976 0.551
#> 5 6.35 0.0275 1.21e-26 0 control 0.0220 0.976 0.551
#> 6 5.59 0.117 2.59e-18 0 control 0.0220 0.972 0.551
#> 7 3.00 0.242 4.30e- 2 0 control 0.0232 0.965 0.551
#> 8 3.26 0.300 7.00e- 3 1 case 0.0234 0.957 0.551
#> 9 3.20 0.288 1.09e- 2 0 control 0.0237 0.949 0.551
#> 10 4.20 0.384 2.64e- 7 0 control 0.0237 0.938 0.551
#> # ℹ 192 more rows
#> # ℹ 4 more variables: .roc_row <int>, .direction <chr>, .is_best_youden <lgl>,
#> # .is_best_closest_topleft <lgl>
# sorted along response first: correct AUC
compute_smooth_density_roc(data_shuffled, d1, d2, along = y)
#> # A tibble: 202 × 12
#> y d1 d2 outcome group .sensitivities .specificities .auc
#> <dbl> <dbl> <dbl> <int> <chr> <dbl> <dbl> <dbl>
#> 1 2.29 0.0963 7.70e- 1 1 case 0.776 0.952 0.967
#> 2 3.43 0.334 1.66e- 3 0 control 1.00 0.707 0.967
#> 3 4.40 0.363 1.49e- 8 0 control 1.00 0.342 0.967
#> 4 6.33 0.0293 2.63e-26 0 control 1 0.00551 0.967
#> 5 6.35 0.0275 1.21e-26 0 control 1 0.00472 0.967
#> 6 5.59 0.117 2.59e-18 0 control 1 0.0534 0.967
#> 7 3.00 0.242 4.30e- 2 0 control 0.995 0.833 0.967
#> 8 3.26 0.300 7.00e- 3 1 case 0.999 0.762 0.967
#> 9 3.20 0.288 1.09e- 2 0 control 0.999 0.779 0.967
#> 10 4.20 0.384 2.64e- 7 0 control 1.00 0.416 0.967
#> # ℹ 192 more rows
#> # ℹ 4 more variables: .roc_row <int>, .direction <chr>, .is_best_youden <lgl>,
#> # .is_best_closest_topleft <lgl>