app.py

from datetime import datetime, timezone

import gradio as gr
import pandas as pd

from src.about import INTRODUCTION_TEXT, TITLE
from src.display.css_html_js import custom_css
from src.leaderboard.columns import DisplayColumns, RequiredInputColumns
from src.leaderboard.input import load_csv_from_github
from src.leaderboard.output import format_output_df

LEADERBOARD_GITHUB_URL = "https://github.com/upgini/mle-bench/blob/main/rankings/low/tabular/overall_ranks.csv"


def load_leaderboard() -> pd.DataFrame:
    """Download the remote leaderboard CSV from GitHub (handles Git LFS).

    Returns a processed DataFrame ready for display.
    """

    df = load_csv_from_github()
    # Process dates
    df["Date"] = pd.to_datetime(df["Date"], errors="coerce").dt.strftime("%Y-%m-%d")

    # Sort by mean_normalized_score descending before formatting
    df = df.sort_values(by="mean_normalized_score", ascending=False, ignore_index=True)

    # Format columns for display
    result_df = format_output_df(df)
    return result_df


def refresh_leaderboard():
    """Fetch the leaderboard and build the status message for the UI."""
    df = apply_styling(load_leaderboard())
    status = (
        f"Showing data from [GitHub]({LEADERBOARD_GITHUB_URL}). "
        f"Last refreshed: {datetime.now(timezone.utc):%Y-%m-%d %H:%M UTC}."
    )
    return df, status


def apply_styling(df: pd.DataFrame):
    """Apply styling to the leaderboard table."""

    if df.empty:
        return pd.DataFrame(columns=DisplayColumns.values())

    display_df = df[DisplayColumns.values()]

    style = (
        display_df.style.background_gradient(
            subset=[DisplayColumns.NORMALIZED_SCORE],
            high=0.5,
            low=0.0,
            cmap="Greens",
            gmap=df[RequiredInputColumns.MEAN_NORMALIZED_SCORE],
        )
        .background_gradient(
            subset=[DisplayColumns.ANY_MEDAL_SCORE],
            high=1.2,
            low=0.0,
            cmap="Oranges",
            gmap=df[RequiredInputColumns.MEAN_MEDAL_PCT],
        )
        .format(
            subset=(
                df[RequiredInputColumns.MEAN_NORMALIZED_SCORE] == df[RequiredInputColumns.MEAN_NORMALIZED_SCORE].max(),
                DisplayColumns.NORMALIZED_SCORE,
            ),
            formatter=lambda x: f"**{x}**",
        )
    )

    return style


def create_app():
    """Create and configure the Gradio app without launching it."""
    with gr.Blocks(title="Upgini MLE-Bench Leaderboard", css=custom_css) as demo:
        gr.HTML(TITLE)
        gr.Markdown(INTRODUCTION_TEXT)

        leaderboard_table = gr.DataFrame(
            value=pd.DataFrame(columns=DisplayColumns.values()),
            wrap=True,
            interactive=False,
            type="pandas",
            datatype="markdown",
            label="Leaderboard",
            elem_id="leaderboard-table",
            show_search="search",
        )

        status_text = gr.Markdown()

        demo.load(refresh_leaderboard, outputs=[leaderboard_table, status_text])

    return demo


if __name__ == "__main__":
    demo = create_app()
    demo.queue(default_concurrency_limit=8).launch()