MLRS/MELABench

1import json2from pathlib import Path3 4import pandas as pd5 6from src.display.formatting import has_no_nan_values, make_clickable_model7from src.display.utils import AutoEvalColumn, EvalQueueColumn8from src.leaderboard.read_evals import get_raw_eval_results9 10 11def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:12    """Creates a dataframe from all the individual experiment results"""13    raw_data = get_raw_eval_results(results_path)14    all_data_json = [v.to_dict() for v in raw_data]15 16    df = pd.DataFrame.from_records(all_data_json)17    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)18    df = df[cols].round(decimals=2)19    df.dropna(how="all", axis=1, inplace=True)20 21    return df22 23 24def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:25    """Creates the different dataframes for the evaluation queues requestes"""26    all_evals = []27 28    for file_path in Path(save_path).rglob("requests_*.json"):29        with open(file_path) as fp:30            data = json.load(fp)["leaderboard"]31 32        data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])33        data[EvalQueueColumn.revision.name] = data.get("revision", "main")34 35        all_evals.append(data)36 37    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]38    running_list = [e for e in all_evals if e["status"] == "RUNNING"]39    finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"]40    df_pending = pd.DataFrame.from_records(pending_list, columns=cols)41    df_running = pd.DataFrame.from_records(running_list, columns=cols)42    df_finished = pd.DataFrame.from_records(finished_list, columns=cols)43    return df_finished[cols], df_running[cols], df_pending[cols]44