merged roc auc plots to summary all folds

Author: GabrielCabas

bin/roc_curve.R

@@ -36,12 +36,11 @@ compute_auc <- function(roc_df) {
   sum((x[-1] - x[-length(x)]) * (y[-1] + y[-length(y)]) / 2)
 }
 
-roc_plot <- function(df, output_path = "roc_curve.png") {
+compute_roc_and_auc <- function(df) {
   if (!all(c("y_true", "y_score") %in% colnames(df))) {
     if (!("y_pred" %in% colnames(df))) {
       stop("Expected columns y_true and y_score (or y_pred) in input CSV.", call. = FALSE)
     }
-    message("Column y_score not found, using y_pred as score (may degrade ROC).")
     df$y_score <- df$y_pred
   }
 
@@ -51,31 +50,106 @@ roc_plot <- function(df, output_path = "roc_curve.png") {
   roc_df <- compute_roc(y_true, y_score)
   auc <- compute_auc(roc_df)
 
-  p <- ggplot(roc_df, aes(x = fpr, y = tpr)) +
-    geom_line(color = "#67a9cf", size = 1) +
+  list(roc_df = roc_df, auc = auc)
+}
+
+roc_plot_combined <- function(roc_data_list, output_path = "roc_curve.png", title = "ROC Curves") {
+  # Combine all ROC data with labels
+  all_roc_data <- data.frame()
+  
+  for (i in seq_along(roc_data_list)) {
+    roc_df <- roc_data_list[[i]]$roc_df
+    auc <- roc_data_list[[i]]$auc
+    label <- roc_data_list[[i]]$label
+    
+    # Label format: "fold_num (AUC = X.XXX)"
+    roc_df$fold <- sprintf("%s (AUC = %.3f)", label, auc)
+    all_roc_data <- rbind(all_roc_data, roc_df)
+  }
+
+  # Generate color palette
+  n_folds <- length(roc_data_list)
+  colors <- rainbow(n_folds)
+  
+  p <- ggplot(all_roc_data, aes(x = fpr, y = tpr, color = fold)) +
+    geom_line(linewidth = 1) +
     geom_abline(slope = 1, intercept = 0, linetype = "dashed", color = "grey60") +
     coord_equal(xlim = c(0, 1), ylim = c(0, 1)) +
+    scale_color_manual(values = colors, name = "K-Fold") +
     labs(
       x = "False Positive Rate",
       y = "True Positive Rate",
-      title = sprintf("ROC curve (AUC = %.3f)", auc)
+      title = title
     ) +
-    theme_minimal()
-
-  ggsave(output_path, plot = p, width = 6, height = 6, dpi = 300)
+    theme_minimal() +
+    theme(
+      legend.position = "right",
+      legend.title = element_text(size = 10),
+      legend.text = element_text(size = 9)
+    )
+
+  ggsave(output_path, plot = p, width = 8, height = 6, dpi = 300)
+  message(sprintf("Generated combined ROC curve: %s", output_path))
 }
 
 main <- function() {
   args <- commandArgs(trailingOnly = TRUE)
   if (length(args) < 1) {
-    stop("Usage: Rscript roc_curve.R <test_predictions.csv> [output_path]", call. = FALSE)
+    stop("Usage: Rscript roc_curve.R <test_predictions1.csv> [test_predictions2.csv ...]", call. = FALSE)
+  }
+
+  csv_files <- args
+
+  # Extract feature_extractor and algorithm from first file
+  # Expected format: {feature_extractor}.{algorithm}.{fold}.test_predictions.csv
+  first_file <- basename(csv_files[1])
+  parts <- strsplit(first_file, "\\.")[[1]]
+  
+  if (length(parts) >= 4) {
+    feature_extractor <- parts[1]
+    algorithm <- parts[2]
+    output_path <- sprintf("%s.%s.roc_auc_curve.png", feature_extractor, algorithm)
+  } else {
+    output_path <- "roc_auc_curve.png"
   }
 
-  test_predictions <- args[1]
-  output_path <- if (length(args) >= 2) args[2] else "roc_auc_curve.png"
+  # Process all CSV files and collect ROC data
+  roc_data_list <- list()
+  
+  for (csv_file in csv_files) {
+    if (!file.exists(csv_file)) {
+      warning(sprintf("File not found: %s, skipping...", csv_file))
+      next
+    }
+
+    # Read and process
+    df <- read_csv(csv_file, show_col_types = FALSE)
+    base_name <- tools::file_path_sans_ext(basename(csv_file))
+    
+    # Extract fold number from filename (e.g., model.algorithm.0.test_predictions -> 0)
+    file_parts <- strsplit(base_name, "\\.")[[1]]
+    if (length(file_parts) >= 3) {
+      fold_num <- file_parts[3]
+      label <- fold_num
+    } else {
+      # Fallback: try to extract number from end of filename
+      fold_match <- regmatches(base_name, regexpr("\\d+", base_name))
+      label <- if (length(fold_match) > 0) fold_match else base_name
+    }
+    
+    roc_data <- compute_roc_and_auc(df)
+    roc_data$label <- label
+    roc_data_list[[length(roc_data_list) + 1]] <- roc_data
+    
+    message(sprintf("Processed %s: AUC = %.3f", csv_file, roc_data$auc))
+  }
+
+  if (length(roc_data_list) == 0) {
+    stop("No valid CSV files found to process.", call. = FALSE)
+  }
 
-  df <- read_csv(test_predictions, show_col_types = FALSE)
-  roc_plot(df, output_path)
+  # Generate combined plot
+  roc_plot_combined(roc_data_list, output_path)
 }
 
 if (identical(environment(), globalenv())) {

modules/visualization.nf

@@ -39,14 +39,18 @@ process roc_auc_curve {
     path(test_predictions)
     path(script_roc_auc_curve)
     output:
-    path("${test_predictions}.roc_auc_curve.png"), emit: roc_auc_curve
+    path("*.roc_auc_curve.png"), emit: roc_auc_curve
     script:
+    // Output filename is auto-generated by R script as {feature_extractor}.{algorithm}.roc_auc_curve.png
     """
     Rscript ${script_roc_auc_curve} $test_predictions
-    cp roc_auc_curve.png ${test_predictions}.roc_auc_curve.png
     """
     stub:
+    // Extract feature_extractor and algorithm from first file
+    def parts = test_predictions[0].name.toString().split('\\.')
+    def feature_extractor = parts[0]
+    def algorithm = parts[1]
     """
-    touch ${test_predictions}.roc_auc_curve.png
+    touch ${feature_extractor}.${algorithm}.roc_auc_curve.png
     """
 }

workflows/grid_search.nf

@@ -20,11 +20,11 @@ workflow grid_search_workflow {
     algorithms = ["elasticnet"]
     if (params.task == "classification"){
         grid_search(dataset, script_grid_search_classification, algorithms)
-        roc_auc_curve(grid_search.out.test_predictions.flatten(), script_roc_auc_curve)
+        roc_auc_curve(grid_search.out.test_predictions, script_roc_auc_curve)
     }
     else {
         grid_search(dataset, script_grid_search_regression, algorithms)
-        scatterplot(grid_search.out.test_predictions.flatten(), script_scatterplot)
+        scatterplot(grid_search.out.test_predictions, script_scatterplot)
     }
     emit:
     cv_results = grid_search.out.cv_results