src/structures/mat.py

import pandas as pd
from pathlib import Path
from ..styles import highlight_color

# Define the absolute path to the file
abs_path = Path(__file__).parent.parent.parent

def load_json_data(file_path):
    # Load the JSON data
    MAT_SCORES = pd.read_json(file_path)

    # Reset index so model names become a column and transpose for (year, name) pairs as rows
    MAT_SCORES = MAT_SCORES.T.reset_index()

    # Rename the first column as 'Model' to keep model names visible
    MAT_SCORES.rename(columns={'index': 'Model'}, inplace=True)

    # Filter columns that contain 'Egzaminy Gimnazjalne' in the name
    filtered_columns = ['Model'] + [col for col in MAT_SCORES.columns if "Egzaminy Maturalne" in col]
    MAT_SCORES = MAT_SCORES[filtered_columns]
    MAT_SCORES["Model"] = MAT_SCORES["Model"].apply(
        lambda name: f"[{name.replace('__','/')}](https://huggingface.co/{name.replace('__','/')})"
    )

        # Round numeric values to 2 decimal places
    numeric_columns = MAT_SCORES.columns[1:]  # Get all year columns
    MAT_SCORES[numeric_columns] = MAT_SCORES[numeric_columns].apply(pd.to_numeric, errors='coerce') * 100
    MAT_SCORES[numeric_columns] = MAT_SCORES[numeric_columns].round(2)

    # Convert year part in column names to strings for Gradio compatibility
    MAT_SCORES.columns = [col.split(',')[0][1:] if col != 'Model' else col for col in MAT_SCORES.columns]
    year_columns = MAT_SCORES.columns[1:]
    sorted_year_columns = sorted(year_columns.astype(str).tolist())  # Sort the year columns as strings
    sorted_columns = ['Model'] + sorted_year_columns
    MAT_SCORES = MAT_SCORES[sorted_columns]

    # Sort alphabetically by model name
    MAT_SCORES = MAT_SCORES.sort_values(by='Model')
    return MAT_SCORES

# Define file path
file_path = str(abs_path / "leaderboards/all_types_years.json")
MAT_SCORES = load_json_data(file_path)
MAT_SCORES = MAT_SCORES.style.highlight_max(
    color = highlight_color,
    subset=MAT_SCORES.columns[-22:]).format(precision=2)