import gradio as gr
import pandas as pd
import json

def process_df(df):
    # Set the current date and time
    df['date'] = pd.Timestamp.now().strftime('%Y-%m-%d %H:%M')
    
    # Create clickable links in the "Model" column
    df['Model'] = df.apply(lambda x: f'<a href="{x["Link"]}" target="_blank">{x["Model"]}</a>', axis=1)
    
    # Drop the "Link" column
    df = df.drop(['Link'], axis=1)
    
    # Convert columns to float
    cols = df.columns.drop(['Model', 'date'])
    df[cols] = df[cols].astype(float)
    
    # Highlight the minimum values in the columns
    styled_df = df.style.highlight_min(color='lightgreen', axis=0, subset=cols)
    
    return styled_df

def get_eval(jsonfile):
    global leaderboard_df
    
    # Load JSON data
    data = json.load(open(jsonfile))
    
    # Convert JSON data to DataFrame
    df = pd.DataFrame.from_dict(data, orient='columns')
    
    # Concatenate with the existing leaderboard
    leaderboard_df = pd.concat([leaderboard_df, df], ignore_index=True)
    
    # Process the dataframe
    df = process_df(leaderboard_df)
    
    return df

def refresh_leaderboard():
    global leaderboard_df
    if len(leaderboard_df) == 0:
        return leaderboard_df
    else:
        return process_df(leaderboard_df)

# Usage markdown
usage = """

## Using the web interface

Run your model on the evaluation datasets and create a json file in this format:
```
[
    {
        "Model": "whisperbase",
        "date": "",
        "ucfd_eng": 102.27,
        "ucfd_lug": "nan",
        "sema_eng" : 42.86,
        "sema_lug": "nan",
        "trac_fm_lug": "nan",
        "Link": "https://colab.research.google.com/drive/1MGGOotrjSXWkR3ljNEVvfEUResEx6OLV#scrollTo=Elwx1K3fEDjf"
    },
    {
        "Model": "facebookmms",
        "date": "2024-06-26",
        "ucfd_eng": 92.91,
        "ucfd_lug": "nan",
        "sema_eng" : 63.39,
        "sema_lug": "nan",
        "trac_fm_lug": 69.29,
        "Link": "https://colab.research.google.com/drive/1MGGOotrjSXWkR3ljNEVvfEUResEx6OLV#scrollTo=Elwx1K3fEDjf"
    }
]
```
>Upload this to the leaderboard and it should be updated with your results.

## Using the API

You can use the following code to programmatically upload your results from a colab or python script.
```
from gradio_client import Client, handle_file

client = Client("sunbird/SB_ASR_Leaderboard")
result = client.predict(
		jsonfile=handle_file([/path/to/eval_results.json]),
		api_name="/get_eval"
)
```
>Refresh the leaderboard and your results should be visible.
"""

# Initialize an empty dataframe for the leaderboard
columns = ['Model', 'date', 'ucfd_eng', 'ucfd_lug',
           'sema_eng', 'sema_lug', 'trac_fm_lug']
leaderboard_df = pd.DataFrame(columns=columns)

# Create Gradio interface
block = gr.Blocks()

with block:
    gr.Markdown("## Leaderboard for Sunbird AI ASR")
    
    with gr.Row():
        input_json = gr.UploadButton("Upload a JSON file", file_count="single", scale=0)
    
    with gr.Tabs():
        with gr.TabItem("ASR Leaderboard"):
            with gr.Row():
                data = gr.DataFrame(datatype="markdown")
        with gr.TabItem("Usage"):
            with gr.Row():
                gr.Markdown(usage)
                
    input_json.upload(get_eval, inputs=input_json, outputs=data)
    
    # Running the function on page load in addition to when the button is clicked
    block.load(refresh_leaderboard, inputs=None, outputs=data)

block.launch(share=True)