app,py

import gradio as gr
import pandas as pd
import os
import re
from datetime import datetime

# Leaderboard Data (example CSV file for leaderboard)
LEADERBOARD_FILE = "leaderboard.csv"

def clean_answer(answer):
    if pd.isna(answer):
        return None
    answer = str(answer)
    clean = re.sub(r'[^A-Da-d]', '', answer)
    if clean:
        first_letter = clean[0].upper()
        if first_letter in ['A', 'B', 'C', 'D']:
            return first_letter
    return None

def update_leaderboard(results):
    # Append results to leaderboard file
    new_entry = {
        "Model Name": results['model_name'],
        "Overall Accuracy": f"{results['overall_accuracy']:.2%}",
        "Valid Accuracy": f"{results['valid_accuracy']:.2%}",
        "Correct Predictions": results['correct_predictions'],
        "Total Questions": results['total_questions'],
        "Timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
    }
    leaderboard_df = pd.DataFrame([new_entry])
    if os.path.exists(LEADERBOARD_FILE):
        existing_df = pd.read_csv(LEADERBOARD_FILE)
        leaderboard_df = pd.concat([existing_df, leaderboard_df], ignore_index=True)
    leaderboard_df.to_csv(LEADERBOARD_FILE, index=False)

def evaluate_predictions(prediction_file):
    ground_truth_file = "ground_truth.csv"  # Specify the path to the ground truth file
    if not prediction_file:
        return "Prediction file not uploaded", None

    if not os.path.exists(ground_truth_file):
        return "Ground truth file not found", None

    try:
        predictions_df = pd.read_csv(prediction_file.name)
        ground_truth_df = pd.read_csv(ground_truth_file)
        filename = os.path.basename(prediction_file.name)
        model_name = filename.split('_')[1].split('.')[0] if "_" in filename else "unknown_model"

        merged_df = pd.merge(predictions_df, ground_truth_df, on='question_id', how='inner')
        merged_df['pred_answer'] = merged_df['predicted_answer'].apply(clean_answer)
        correct_predictions = (merged_df['pred_answer'] == merged_df['Answer']).sum()
        total_predictions = len(merged_df)
        overall_accuracy = correct_predictions / total_predictions

        results = {
            'model_name': model_name,
            'overall_accuracy': overall_accuracy,
            'correct_predictions': correct_predictions,
            'total_questions': total_predictions,
        }

        update_leaderboard(results)

        return "Evaluation completed successfully! Leaderboard updated.", LEADERBOARD_FILE
    except Exception as e:
        return f"Error: {str(e)}", None

# Gradio Interface with Leaderboard
def display_leaderboard():
    if not os.path.exists(LEADERBOARD_FILE):
        return "Leaderboard is empty."
    leaderboard_df = pd.read_csv(LEADERBOARD_FILE)
    return leaderboard_df.to_markdown(index=False)

demo = gr.Blocks()

with demo:
    gr.Markdown("# Prediction Evaluation Tool with Leaderboard")
    with gr.Tab("Evaluate"):
        file_input = gr.File(label="Upload Prediction CSV")
        eval_status = gr.Textbox(label="Evaluation Status")
        eval_results_file = gr.File(label="Download Evaluation Results")
        eval_button = gr.Button("Evaluate")
        eval_button.click(
            evaluate_predictions, inputs=file_input, outputs=[eval_status, eval_results_file]
        )
    with gr.Tab("Leaderboard"):
        leaderboard_text = gr.Textbox(label="Leaderboard", interactive=False)
        refresh_button = gr.Button("Refresh Leaderboard")
        refresh_button.click(display_leaderboard, outputs=leaderboard_text)

if __name__ == "__main__":
    demo.launch()