Spaces:
Running
Running
| 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 | |
| from src.envs import RESULTS_REPO | |
| 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] | |
| # print(raw_data) | |
| df = pd.DataFrame.from_records(json.load(open(raw_data[0]))) | |
| print(list(df.columns)) | |
| df['95% CI'] = " " | |
| # df['model']="nothing" | |
| # df.columns = cols | |
| # df.iloc[0]= create dummy | |
| # print(dir(AutoEvalColumn)) | |
| df = df.sort_values(by=[AutoEvalColumn.task0.name], ascending=False) | |
| decimal = 1 | |
| for i,row in df.iterrows(): | |
| if 'lower' not in row: | |
| continue | |
| interval = '+'+str(round(row['upper'] - row['score'], decimal))+' / '+str(round(row['lower'] - row['score'], decimal)) | |
| df.at[i,'95% CI'] = interval | |
| df = df[cols].round(decimals=2) | |
| # 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(".") and not entry.endswith(".jsonl")] | |
| all_evals = [] | |
| for entry in entries: | |
| if ".json" in entry and 'toeval' not in entry: | |
| file_path = os.path.join(save_path, entry) | |
| with open(file_path) as fp: | |
| print(file_path) | |
| 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 and 'toeval' not in entry and 'results' not in entry: | |
| # this is a folder | |
| sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if not e.startswith(".")] | |
| for sub_entry in sub_entries: | |
| if 'toeval' in sub_entry: | |
| continue | |
| file_path = os.path.join(save_path, entry, sub_entry) | |
| with open(file_path) as fp: | |
| # print(file_path) | |
| 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] | |