src/populate.py

import json
import os

import pandas as pd

from src.display.formatting import has_no_nan_values, make_clickable_model, model_hyperlink
from src.display.utils import AutoEvalColumn, EvalQueueColumn
from src.leaderboard.read_evals import get_raw_eval_results

def calc_average(row: pd.Series, benchmark_cols: list) -> float:
    """Calculates the average of the benchmark columns that exist in the row"""
    return row[[col for col in benchmark_cols if col in row]].mean()

def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols_paired: list, cols_paired: 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]
    all_data_json = []
    benchmark_cols = [col[0] for col in benchmark_cols_paired]
    with open('./master_table.json') as f:
        content = json.load(f)
        for key, val in content.items():
            val['eval_name'] = val['id']
            del val['id']
            
            if 'link' in val and val['link'].strip():
                val['Algorithm'] = model_hyperlink(val['link'], val['name'])
            else:
                val['Algorithm'] = val['name']
            
            del val['name']
            
            # fill in the missing benchmark columns as 0
            for display_name, benchmark in benchmark_cols_paired:
                if benchmark not in val:
                    val[display_name] = 0
                else:
                    val[display_name] = val[benchmark]
                    del val[benchmark]
            
            # change all the keys to the display names
            for display_name, col in cols_paired:
                if display_name in val:
                    pass
                elif col in val:
                    val[display_name] = val[col]
                    del val[col]
                else:
                    val[display_name] = None
            
            all_data_json.append(val)
    
    print(f'All data json: {all_data_json}')

    df = pd.DataFrame.from_records(all_data_json)
    df[AutoEvalColumn.average.name] = df.apply(lambda row: calc_average(row, benchmark_cols), axis=1)
    print(df, AutoEvalColumn.average.name)
    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
    df = df[cols].round(decimals=4)

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


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 ".md" not in entry:
            # this is a folder
            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if not e.startswith(".")]
            print(sub_entries)
            for sub_entry in sub_entries:
                file_path = os.path.join(save_path, entry, sub_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)

    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"]
    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]