Compute a receiver-operating curve (ROC) for a two-class or multiclass dataset

Usage

tof_make_roc_curve(input_data, truth_col, prob_cols)

Arguments

input_data: A tof_tbl, tbl_df, or data.frame in which each row is an observation.
truth_col: An unquoted column name indicating which column in `input_data` contains the true class labels for each observation. Must be a factor.
prob_cols: Unquoted column names indicating which columns in `input_data` contain the probability estimates for each class in `truth_col`. These columns must be specified in the same order as the factor levels in `truth_col`.

Value

A tibble that can be used to plot the ROC for a classification task. For each candidate probability threshold, the following are reported: specificity, sensitivity, true-positive rate (tpr), and false-positive rate (fpr).

Examples

feature_tibble <-
    dplyr::tibble(
        sample = as.character(1:100),
        cd45 = runif(n = 100),
        pstat5 = runif(n = 100),
        cd34 = runif(n = 100),
        outcome = (3 * cd45) + (4 * pstat5) + rnorm(100),
        class =
            as.factor(
                dplyr::if_else(outcome > median(outcome), "class1", "class2")
            )
    )

split_data <- tof_split_data(feature_tibble, split_method = "simple")

# train a logistic regression classifier
log_model <-
    tof_train_model(
        split_data = split_data,
        predictor_cols = c(cd45, pstat5, cd34),
        response_col = class,
        model_type = "two-class"
    )

# make predictions
predictions <-
    tof_predict(
        log_model,
        new_data = feature_tibble,
        prediction_type = "response"
    )
prediction_tibble <-
    dplyr::tibble(
        truth = feature_tibble$class,
        prediction = predictions$.pred
    )

# make ROC curve
tof_make_roc_curve(
    input_data = prediction_tibble,
    truth_col = truth,
    prob_cols = prediction
)