如何使用自定义getROC_AUC函数结合ggplot绘制ROC曲线并实现多线展示？

Hey there! Let's walk through how to plot multiple ROC curves using your custom getROC_AUC function and ggplot2. Here's a step-by-step solution tailored to your needs:

1. Confirm Your Function & Prepare Sample Data #

First, let's include your function (so we can work with it directly) and generate sample data for 3 different models—this lets you test the code immediately with reproducible results:

# Your custom ROC/AUC calculation function
getROC_AUC <- function(probs, true_Y){
  probsSort = sort(probs, decreasing = TRUE, index.return = TRUE)
  val = unlist(probsSort$x)
  idx = unlist(probsSort$ix)
  roc_y = true_Y[idx]; 
  stack_x = cumsum(roc_y == 0)/sum(roc_y == 0) # False Positive Rate (FPR)
  stack_y = cumsum(roc_y == 1)/sum(roc_y == 1) # True Positive Rate (TPR)
  auc = sum((stack_x[2:length(roc_y)]-stack_x[1:length(roc_y)-1])*stack_y[2:length(roc_y)])
  return(list(stack_x=stack_x, stack_y=stack_y, auc=auc))
}

# Simulate sample data (replace this with your actual model predictions!)
set.seed(123) # Ensure results are reproducible
true_Y <- sample(c(0,1), size = 100, replace = TRUE)
# Predictions from 3 hypothetical models
prob_model1 <- runif(100, 0, 1) # Random guess-like performance
prob_model2 <- ifelse(true_Y == 1, runif(50, 0.4, 1), runif(50, 0, 0.6)) # Moderate performance
prob_model3 <- ifelse(true_Y == 1, runif(50, 0.6, 1), runif(50, 0, 0.4)) # Strong performance

2. Calculate ROC Metrics for Each Model #

Run your function on each model's predictions to generate FPR, TPR, and AUC values:

# Compute ROC data for each model
roc_model1 <- getROC_AUC(prob_model1, true_Y)
roc_model2 <- getROC_AUC(prob_model2, true_Y)
roc_model3 <- getROC_AUC(prob_model3, true_Y)

3. Format Data for ggplot2 #

ggplot works best with tidy data (one row per observation). We'll combine all ROC data into a single data frame, adding a Model column to distinguish curves—we'll even include the AUC in the model name for quick performance comparison:

# Combine into a tidy data frame
roc_tidy_data <- rbind(
  data.frame(
    FPR = roc_model1$stack_x,
    TPR = roc_model1$stack_y,
    Model = paste0("Model 1 (AUC = ", round(roc_model1$auc, 3), ")")
  ),
  data.frame(
    FPR = roc_model2$stack_x,
    TPR = roc_model2$stack_y,
    Model = paste0("Model 2 (AUC = ", round(roc_model2$auc, 3), ")")
  ),
  data.frame(
    FPR = roc_model3$stack_x,
    TPR = roc_model3$stack_y,
    Model = paste0("Model 3 (AUC = ", round(roc_model3$auc, 3), ")")
  )
)

4. Plot Multiple ROC Curves #

Now we can plot all curves in one visualization, plus add a dashed gray line to represent a random guess (AUC = 0.5) for baseline comparison:

library(ggplot2)

ggplot(roc_tidy_data, aes(x = FPR, y = TPR, color = Model)) +
  # Draw ROC curves
  geom_line(linewidth = 1) +
  # Add random guess baseline
  geom_abline(intercept = 0, slope = 1, linetype = "dashed", color = "gray50") +
  # Customize labels and theme
  labs(
    x = "False Positive Rate (FPR)",
    y = "True Positive Rate (TPR)",
    title = "ROC Curves: Comparing Multiple Models",
    color = "Model"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(hjust = 0.5, size = 14, face = "bold"),
    axis.title = element_text(size = 12),
    legend.position = "bottom"
  )

Quick Tips: #

• Swap out the sample prob_model1/2/3 data with your actual model predictions.
• The Model column is essential—it tells ggplot which points belong to which curve.
• Adjust linewidth, colors, or theme elements to match your preferred styling.

内容的提问来源于stack exchange，提问作者Basel.D