plot_feature_importance.py

import os
import glob
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

def generate_plots_per_feature(folder_path, output_root_dir="plots_features"):
    """
    Iterates over each CSV, extracts each feature row by row,
    and generates a horizontal barplot of its sub-column metrics.
    """
    # 1. Find all CSV files in the folder
    csv_files = glob.glob(os.path.join(folder_path, "*.csv"))

    if not csv_files:
        print(f"No CSV file found in: {folder_path}")
        return

    # Create the main folder to store all the plots
    os.makedirs(output_root_dir, exist_ok=True)

    for csv_file in csv_files:
        csv_name = os.path.splitext(os.path.basename(csv_file))[0]
        print(f"\n--- Processing file: {csv_name}.csv ---")

        # Create a specific subfolder for this CSV file to keep things separate
        csv_output_dir = os.path.join(output_root_dir, csv_name)
        os.makedirs(csv_output_dir, exist_ok=True)

        # Read the CSV file
        df = pd.read_csv(csv_file)

        name_features = df["Feature_Name"]

        # 2. Loop over each row (each individual Feature)
        for col in df.columns:
            
            if col != "Feature_Name":
                metrics = df[col]
                feature_name = str(col).strip().replace("_Importance","")
                safe_feature_name = "".join([c if c.isalnum() or c in ('_', '-') else '_' for c in feature_name])
            
                metrics = pd.to_numeric(metrics, errors='coerce').dropna()
                
                # Convert into a DataFrame usable by Seaborn/Matplotlib
                df_metrics = pd.DataFrame({
                    'Metric_Name': name_features,
                    'Importance': metrics.values
                })

                df_metrics = df_metrics.drop(df_metrics[df_metrics['Importance'] == 0].index)
                # Optional: Sort the bars by value for better readability
                df_metrics = df_metrics.sort_values(by='Importance', ascending=False)
                print(df_metrics)
                # 3. Create the horizontal bar chart
                # Adjust the figure size based on the number of sub-components (columns)
                plt.figure(figsize=(12, max(5, len(df_metrics) * 0.25)))
                sns.set_theme(style="whitegrid")
                
                # Barplot horizontal
                sns.barplot(
                    x='Importance',
                    y='Metric_Name',
                    data=df_metrics,
                    palette="coolwarm"  # Gradient from blue (negative) to red (positive)
                )

                # Titles and labels
                plt.title(f"Feature Importance for {feature_name}", fontsize=14, fontweight='bold', pad=12)
                plt.xlabel("Importance Values", fontsize=11)
                plt.ylabel("Feature Names", fontsize=11)
                
                plt.tight_layout()
                
                # 4. Save the individual PNG file
                filename = f"{safe_feature_name}.png"
                full_path = os.path.join(csv_output_dir, filename)

                plt.savefig(full_path, dpi=150) # 150 DPI is more than enough and speeds up generation
                plt.close() # Very important to free up RAM at each iteration

        print(f"-> Success: {len(df)} plots generated in the folder: {csv_output_dir}")

# --- EXECUTION ---
# Replace "." with the path of your folder containing your CSV files
csv_folder_path = "/home/luciacev/Documents/SurgicalMovementPrediction-Training/feature_importance"

generate_plots_per_feature(csv_folder_path)