在R语言中绘制多条ROC曲线及批量计算AUC分数的替代方法咨询

Great question! Let's split this into two clear parts: alternative ways to calculate those 150 AUC scores (beyond the Map/sapply/lapply you already know), and how to plot multiple ROC curves cleanly in R.

While apply-style functions work fine, these approaches give you more structured, easy-to-use results that fit better into modern R workflows:

1. Tidyverse + pROC (Long-Format Approach) #

Reshape your wide data into long format, then group by each score column to compute AUC. The output is a structured data frame with both column names and their AUC values—much easier to analyze or visualize later:

library(tidyverse)
library(pROC)

# Assume your data frame is named `df`, with first column as `label` and 150 score columns after
auc_results <- df %>%
  pivot_longer(cols = -label, names_to = "score_column", values_to = "score_value") %>%
  group_by(score_column) %>%
  summarize(auc_value = auc(roc(label, score_value)))

# Preview the results
head(auc_results)

2. Data.table Approach #

If you're working with large datasets, this method is faster and memory-efficient. It uses data.table's melt + group-by logic:

library(data.table)
library(pROC)

setDT(df)
auc_results <- melt(df, id.vars = "label")[, 
  .(auc_value = auc(roc(label, value))), 
  by = variable
]

Here are two reliable methods to visualize all your ROC curves, depending on how much customization you need:

1. Using ggroc from pROC (Quick & Minimal Code) #

The pROC package has a built-in ggroc function that accepts a list of ROC objects and automatically generates a multi-curve plot with legends:

library(pROC)
library(ggplot2)

# Create a list of ROC objects for each score column
roc_objects <- lapply(df[, -1], function(score_col) roc(df$label, score_col))

# Plot with automatic color coding and a reference line
ggroc(roc_objects, aes(color = name)) +
  geom_abline(slope = 1, intercept = 0, linetype = "dashed", color = "gray50") +
  labs(
    title = "ROC Curves for 150 Score Columns",
    x = "False Positive Rate",
    y = "True Positive Rate",
    color = "Score Column"
  ) +
  theme_minimal()

2. Tidy Data + ggplot2 (Full Customization) #

If you want full control over line styles, legend placement, or annotations, first extract all ROC coordinates into a tidy data frame, then plot with ggplot2:

library(tidyverse)
library(pROC)

# Extract ROC coordinates for every score column
roc_tidy_data <- map_df(names(df)[-1], function(col_name) {
  roc_obj <- roc(df$label, df[[col_name]])
  tibble(
    false_positive_rate = 1 - roc_obj$specificities,
    true_positive_rate = roc_obj$sensitivities,
    score_column = col_name
  )
})

# Plot with custom styling
ggplot(roc_tidy_data, aes(x = false_positive_rate, y = true_positive_rate, color = score_column)) +
  geom_line(linewidth = 0.8) +
  geom_abline(slope = 1, intercept = 0, linetype = "dashed", color = "gray") +
  labs(
    title = "Customized Multiple ROC Curves",
    x = "False Positive Rate",
    y = "True Positive Rate",
    color = "Score Column"
  ) +
  theme_bw() +
  theme(legend.position = "bottom") # Move legend to avoid cluttering plots

Bonus: Interactive Plotting #

With 150 curves, legends can get crowded. Use plotly to make an interactive plot where you hover to see column names:

library(plotly)

ggplotly(
  ggplot(roc_tidy_data, aes(x = false_positive_rate, y = true_positive_rate, color = score_column, text = score_column)) +
    geom_line(linewidth = 0.8) +
    geom_abline(slope = 1, intercept = 0, linetype = "dashed", color = "gray") +
    labs(x = "False Positive Rate", y = "True Positive Rate") +
    theme_bw()
)

内容的提问来源于stack exchange，提问作者steves