Spaces:
Sleeping
Sleeping
| 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.about import Tasks, N_Tasks, Detail_Tasks | |
| def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list, version="1_correct") -> pd.DataFrame: | |
| """Creates a dataframe from all the individual experiment results""" | |
| cols = cols.copy() | |
| raw_data = get_raw_eval_results(results_path+"/"+version, requests_path) | |
| print(raw_data) | |
| tasks = list(N_Tasks) + list(Detail_Tasks) if "n_" in version else list(Tasks) | |
| if version == "1_correct_var": | |
| tasks = [t for t in tasks if t.value.col_name != "VCR"] | |
| all_data_json = [v.to_dict(tasks) for v in raw_data] | |
| print(all_data_json) | |
| df = pd.DataFrame.from_records(all_data_json) | |
| df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False) | |
| print(cols) | |
| if version != "1_correct": | |
| cols.remove("VCR") | |
| if version != "1_correct_var": | |
| benchmark_cols.remove("VCR") | |
| if version != "n_correct": | |
| for task in Detail_Tasks: | |
| cols.remove(task.value.col_name) | |
| print(df) | |
| print(cols) | |
| 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""" | |
| version = ["1_correct",] # "1_correct_var", "n_correct"] | |
| entries = [os.path.join(v, entry) for v in version for entry in os.listdir(os.path.join(save_path, v)) if not entry.startswith(".")] | |
| print(entries) | |
| all_evals = [] | |
| for entry in entries: | |
| if not ".txt" in entry: | |
| file_path = os.path.join(save_path, entry, "eval_request.json") | |
| 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") | |
| data[EvalQueueColumn.output_format.name] = data.get("output_format") | |
| data[EvalQueueColumn.dataset_version.name] = file_path.split("/")[-3] | |
| 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] | |