import gradio as gr
import pandas as pd
from datasets import load_dataset
from jiwer import wer, cer, transforms
import os
from datetime import datetime


transform = transforms.Compose([
    transforms.RemovePunctuation(),
    transforms.ToLowerCase(),
    transforms.RemoveWhiteSpace(replace_by_space=True),
])


dataset = load_dataset("sudoping01/bambara-asr-benchmark", name="default")["train"]
references = {row["id"]: row["text"] for row in dataset}


leaderboard_file = "leaderboard.csv"
if not os.path.exists(leaderboard_file):
    pd.DataFrame(columns=["submitter", "WER", "CER", "timestamp"]).to_csv(leaderboard_file, index=False)

def process_submission(submitter_name, csv_file):

    try:
        # Read and validate the uploaded CSV
        df = pd.read_csv(csv_file)
        if set(df.columns) != {"id", "prediction"}:
            return "Error: CSV must contain exactly 'id' and 'prediction' columns.", None
        if df["id"].duplicated().any():
            return "Error: Duplicate 'id's found in the CSV.", None
        if set(df["id"]) != set(references.keys()):
            return "Error: CSV 'id's must match the dataset 'id's.", None
        
        # Calculate WER and CER for each prediction
        wers, cers = [], []
        for _, row in df.iterrows():
            ref = references[row["id"]]
            pred = row["prediction"]
            wers.append(wer(ref, pred, standardize=transform))
            cers.append(cer(ref, pred, standardize=transform))
        
        # Compute average WER and CER
        avg_wer = sum(wers) / len(wers)
        avg_cer = sum(cers) / len(cers)
        
        # Update the leaderboard
        leaderboard = pd.read_csv(leaderboard_file)
        timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
        new_entry = pd.DataFrame(
            [[submitter_name, avg_wer, avg_cer, timestamp]],
            columns=["submitter", "WER", "CER", "timestamp"]
        )
        leaderboard = pd.concat([leaderboard, new_entry]).sort_values("WER")
        leaderboard.to_csv(leaderboard_file, index=False)
        
        return "Submission processed successfully!", leaderboard
    except Exception as e:
        return f"Error processing submission: {str(e)}", None

# Create the Gradio interface
with gr.Blocks(title="Bambara ASR Leaderboard") as demo:
    gr.Markdown(
        """
        # Bambara ASR Leaderboard
        Upload a CSV file with 'id' and 'text' columns to evaluate your ASR predictions. 
        The 'id's must match those in the dataset. 
        [View the dataset here](https://huggingface.co/datasets/MALIBA-AI/bambara_general_leaderboard_dataset).
        
        - **WER**: Word Error Rate (lower is better).
        - **CER**: Character Error Rate (lower is better).
        """
    )
    with gr.Row():
        submitter = gr.Textbox(label="Submitter Name or Model Name", placeholder="e.g.,  MALIBA-AI/asr")
        csv_upload = gr.File(label="Upload CSV File", file_types=[".csv"])
    submit_btn = gr.Button("Submit")
    output_msg = gr.Textbox(label="Status", interactive=False)
    leaderboard_display = gr.DataFrame(
        label="Leaderboard",
        value=pd.read_csv(leaderboard_file),
        interactive=False
    )
    
    submit_btn.click(
        fn=process_submission,
        inputs=[submitter, csv_upload],
        outputs=[output_msg, leaderboard_display]
    )

demo.launch()