I am testing the performance of a prediction model (binary scenario: 0 or 1) using tidymodels in R. I have created importance weights before the fitting process for all individuals in my dataset. I have done the data splitting part and the full workflow, but I would like my performance metrics (roc_auc and brier_class) to be weighted as well, based on the importance weights.

Right now I have:

set.seed(234)

data_folds <- group_vfold_cv(data,group = Hospital)


# Specify log reg

model <- logistic_reg(mode = "classification",engine = "glm")


# Workflows now

current_wf <- workflow() |> 
  add_case_weights(imp_weights) |>
  add_formula(Status30D ~ Max_NEWS) |>
  add_model(model)

current_wf

# Set up parallel processing
doParallel::registerDoParallel(cores = 6)

cntrl <- control_resamples(save_pred = T)

# Internal-External validation of the current EWS (checking demographic parity also)

current_fit <- fit_resamples(current_wf,resamples = data_folds,
                             metrics = metric_set(roc_auc,brier_class)
                             ,control = cntrl)

I know I can do something like that:

current_fit |> 
collect_predictions() |> 
arrange(.row) |> 
mutate(weights = data$imp_weights) |> 
roc_auc(Status30D,.pred_Deceased,case_weights = weights)

But that will not give me standard errors.

Is there a way to specify my need for weighted performance metrics in the fit_resamples function or somewhere else?