app.py

import gradio as gr
import pandas as pd
import json
from constants import BANNER, INTRODUCTION_TEXT, CITATION_TEXT, METRICS_TAB_TEXT, DIR_OUTPUT_REQUESTS, LEADERBOARD_CSS
from init import is_model_on_hub, upload_file, load_all_info_from_dataset_hub
from utils_display import AutoEvalColumn, fields, make_clickable_model, styled_error, styled_message
from datetime import datetime, timezone

LAST_UPDATED = "Nov 22th 2024"

column_names = {
    "model": "Model",
    "Average WER ⬇️": "Average WER ⬇️",    
    "RTFx": "RTFx ⬆️️",
    "Bulgarian_female": "Bulgarian female",
    "Bulgarian_male": "Bulgarian male",
    "Catalan_female": "Catalan female",
    "Chinese_female": "Chinese female",
    "Chinese_male": "Chinese male",
    "Eastern_European_male": "Eastern European male",
    "European_male": "European male",
    "French_female": "French female",
    "Ghanain_English_female": "Ghanain English female",
    "Indian_English_female": "Indian English female",
    "Indian_English_male": "Indian English male",
    "Indonesian_female": "Indonesian female",
    "Irish_English_female": "Irish English female",
    "Irish_English_male": "Irish English male",
    "Israeli_male": "Israeli male",
    "Italian_female": "Italian female",
    "Jamaican_English_female": "Jamaican English female",
    "Jamaican_English_male": "Jamaican English male",
    "Kenyan_English_female": "Kenyan English female",
    "Kenyan_English_male": "Kenyan English male",
    "Latin_American_female": "Latin American female",
    "Latin_American_male": "Latin American male",
    "Lithuanian_male": "Lithuanian male",
    "Mainstream_US_English_female": "Mainstream US English female",
    "Mainstream_US_English_male": "Mainstream US English male",
    "Nigerian_English_female": "Nigerian English female",
    "Nigerian_English_male": "Nigerian English male",
    "Romanian_female": "Romanian female",
    "Scottish_English_male": "Scottish English male",
    "Southern_British_English_male": "Southern British English male",
    "Spanish_female": "Spanish female",
    "Spanish_male": "Spanish male",
    "Vietnamese_female": "Vietnamese female",
    "Vietnamese_male": "Vietnamese male",
}

eval_queue_repo, requested_models, csv_results = load_all_info_from_dataset_hub()

if not csv_results.exists():
    raise Exception(f"CSV file {csv_results} does not exist locally")
    
# Get csv with data and parse columns
original_df = pd.read_csv(csv_results)

# Formats the columns
def formatter(x):
    if type(x) is str:
        x = x
    else: 
        x = round(x, 2)
    return x

for col in original_df.columns:
    if col == "model":
        original_df[col] = original_df[col].apply(lambda x: x.replace(x, make_clickable_model(x)))
    else:
        original_df[col] = original_df[col].apply(formatter) # For numerical values
 
original_df.rename(columns=column_names, inplace=True)
original_df.sort_values(by='Average WER ⬇️', inplace=True)

COLS = [c.name for c in fields(AutoEvalColumn)]
TYPES = [c.type for c in fields(AutoEvalColumn)]


with gr.Blocks(css=LEADERBOARD_CSS) as demo:
    # gr.HTML(BANNER, elem_id="banner")
    # Write a header with the title
    gr.Markdown("<h1>🤗 Open Automatic Speech Recognition on EdAcc Dataset</h1>", elem_classes="markdown-text")
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")

    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem("🏅 Leaderboard", elem_id="od-benchmark-tab-table", id=0):
            # Add column filter dropdown
            column_filter = gr.Dropdown(
                choices=["All"] + [v for k,v in column_names.items() if k != "model"],
                label="Filter by column", 
                multiselect=True,
                value=["All"],
                elem_id="column-filter"
            )
            
            leaderboard_table = gr.components.Dataframe(
                value=original_df,
                datatype=TYPES,
                elem_id="leaderboard-table",
                interactive=False,
                visible=True,
            )
            
            # Update table columns when filter changes
            def update_table(cols):
                if "All" in cols:
                    return gr.Dataframe(value=original_df)
                selected_cols = ["Model"] + cols  # Always include the Model column
                return gr.Dataframe(value=original_df[selected_cols])
            
            column_filter.change(
                fn=update_table,
                inputs=[column_filter],
                outputs=[leaderboard_table]
            )

demo.launch(ssr_mode=False)