Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
File size: 4,573 Bytes
0061e14 e00a798 b49d2fb e00a798 0061e14 e00a798 0061e14 e00a798 0061e14 e00a798 0061e14 e00a798 0061e14 b49d2fb e00a798 da4996e e00a798 0061e14 e00a798 0061e14 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 |
import json
import os
import pandas as pd
from datasets import load_dataset, get_dataset_config_names
from datasets.exceptions import DatasetNotFoundError
from tqdm.auto import tqdm
from src.display.formatting import has_no_nan_values, make_clickable_model
from src.display.utils import AutoEvalColumn, EvalQueueColumn
from src.envs import TOKEN
from src.leaderboard.read_evals import get_raw_eval_results
from src.logger import get_logger
logger = get_logger(__name__)
def get_leaderboard_df(results_dataset_name: str) -> pd.DataFrame:
"""Creates a dataframe from all the individual experiment results"""
try:
configs = get_dataset_config_names(results_dataset_name, token=TOKEN)
except (DatasetNotFoundError, FileNotFoundError):
# Return an empty DataFrame with expected columns
return pd.DataFrame(
columns=[
"System Name",
"System Type",
"Organization",
"Success Rate (%)",
"Problems Solved",
"Submitted On",
]
)
rows = []
for submission_id in tqdm(configs, total=len(configs), desc="Processing Submission Results"):
submission_ds = load_dataset(results_dataset_name, submission_id, split="train", token=TOKEN)
submission_df = pd.DataFrame(submission_ds)
if submission_df.empty or "did_pass" not in submission_df.columns or submission_df.did_pass.isna().any():
logger.warning(f"Skipping {submission_id} due to invalid did_pass values")
continue
success_rate = 100 * submission_df["did_pass"].mean()
num_solved = submission_df["did_pass"].sum()
first_row = submission_df.iloc[0]
rows.append(
{
"System Name": first_row["system_name"],
"System Type": first_row["system_type"],
"Organization": first_row["organization"],
"Success Rate (%)": success_rate,
"Problems Solved": num_solved,
"Submitted On": pd.to_datetime(first_row["submission_ts"]).strftime("%Y-%m-%d %H:%M"),
}
)
full_df = pd.DataFrame(rows)
# TODO: forbid multiple submissions under the same name?
# Keep only the latest entry per unique (System Name, System Type, Organization) triplet
final_df = (
full_df.sort_values("Submitted On", ascending=False)
.drop_duplicates(subset=["System Name", "System Type", "Organization"], keep="first")
.sort_values(by=[AutoEvalColumn.success_rate.name], ascending=False)
.reset_index(drop=True)
)
cols_to_round = ["Success Rate (%)"]
final_df[cols_to_round] = final_df[cols_to_round].round(decimals=2)
return final_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 os.path.isfile(e) and not e.startswith(".")]
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]
|