src/populate.py

import json
import os

import pandas as pd

from src.display.formatting import has_no_nan_values, make_clickable_model
from src.display.utils import AutoEvalColumn, EvalQueueColumn
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"""
    raw_data = get_raw_eval_results(results_path, requests_path)
    all_data_json = [v.to_dict() for v in raw_data]

    df = pd.DataFrame.from_records(all_data_json)
    
    if not df.empty:
        df = df.sort_values(by=[AutoEvalColumn.average_score.name], ascending=False)
        

        # filter out if any of the benchmarks have not been produced
        df = df[has_no_nan_values(df, benchmark_cols)]

        df.insert(0, "Rank", range(1, len(df) + 1))
  
        
        ##round any float column 
        for col in df.columns:
            if df[col].dtype == "float64":
                df[col] = df[col].round(2)

        df["Benchmark Score (0-10)"] = df["Benchmark Score (0-10)"].astype(str)
        print(df["Benchmark Score (0-10)"])

        print("###############\n\n\n\n\n\n###############")

        print(df)
        print(df.info())


        return df
    else:
        return pd.DataFrame(columns=cols)


def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
    """Creates the different dataframes for the evaluation queues requestes"""
    entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")]
    all_evals = []

    for entry in entries:

        if ".json" in entry:
            file_path = os.path.join(save_path, entry)
            with open(file_path) as fp:
                data = json.load(fp)

            data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
            data[EvalQueueColumn.revision.name] = data.get("revision", "main")

            all_evals.append(data)
        elif os.path.isdir(f"{save_path}/{entry}"):
            # this is a folder

            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if os.path.isfile(f"{save_path}/{entry}/{e}") ]#and not e.startswith(".")
            print(f"Sub entries: {sub_entries}")
            for sub_entry in sub_entries:
                file_path = os.path.join(save_path, entry, sub_entry)
                print(f"{file_path}")

                with open(file_path) as fp:
                    data = json.load(fp)

                data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
                data[EvalQueueColumn.revision.name] = data.get("revision", "main")
                all_evals.append(data)
    

    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]
    print(pending_list)
    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"]
    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]