src/populate.py

import json
import os
import numpy as np
import pandas as pd

from src.display.formatting import has_no_nan_values, make_clickable_model
from src.leaderboard.read_evals import get_raw_eval_results


def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
    """Creates a dataframe from all the individual experiment results"""
    print(f"Getting raw eval results from {results_path} and {requests_path}")
    raw_data = get_raw_eval_results(results_path, requests_path)
    print(f"Got {len(raw_data)} raw eval results")

    if not raw_data:
        print("No raw data found!")
        return pd.DataFrame(columns=cols)

    all_data_json = [v.to_dict() for v in raw_data]
    print(f"Converted {len(all_data_json)} results to dict")
    print(f"Data before DataFrame creation: {all_data_json}")

    df = pd.DataFrame.from_records(all_data_json)
    print(f"Created DataFrame with columns: {df.columns.tolist()}")
    print(f"DataFrame before filtering:\n{df}")

    # Ensure all required columns exist
    for col in cols:
        if col not in df.columns:
            print(f"Adding missing column: {col}")
            df[col] = None

    # Sort by Security Score if available, otherwise don't sort
    if "Security Score ⬆️" in df.columns:
        df = df.sort_values(by="Security Score ⬆️", ascending=False)
        print(f"DataFrame after sorting:\n{df}")
    else:
        print("Security Score column not found, skipping sorting")

    # Select only the columns we want to display
    df = df[cols]
    
    # Round numeric columns
    numeric_cols = df.select_dtypes(include=[np.number]).columns
    df[numeric_cols] = df[numeric_cols].round(decimals=2)
    
    print(f"DataFrame after column selection and rounding:\n{df}")

    print(f"Final DataFrame has {len(df)} rows")
    return df


def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
    """Creates the different dataframes for the evaluation queues requestes"""
    print(f"Looking for eval requests in {save_path}")
    all_evals = []

    # Walk through all directories recursively
    for root, _, files in os.walk(save_path):
        for file in files:
            if file.endswith('.json'):
                file_path = os.path.join(root, file)
                print(f"Reading JSON file: {file_path}")
                with open(file_path) as fp:
                    data = json.load(fp)

                # Create a new dict with the required column names
                formatted_data = {
                    "model": make_clickable_model(data["model"]),
                    "revision": data.get("revision", "main"),
                    "private": data.get("private", False),
                    "precision": data.get("precision", ""),
                    "weight_type": data.get("weight_type", ""),
                    "status": data.get("status", "")
                }

                all_evals.append(formatted_data)

    print(f"Found {len(all_evals)} total eval requests")
    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]
    running_list = [e for e in all_evals if e["status"] == "RUNNING"]
    finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"]

    print(f"Pending: {len(pending_list)}, Running: {len(running_list)}, Finished: {len(finished_list)}")

    df_pending = pd.DataFrame.from_records(pending_list, columns=cols)
    df_running = pd.DataFrame.from_records(running_list, columns=cols)
    df_finished = pd.DataFrame.from_records(finished_list, columns=cols)
    return df_finished[cols], df_running[cols], df_pending[cols]