Spaces:
Running
Running
File size: 4,797 Bytes
84f40ff 6d97820 84f40ff 6d97820 84f40ff 6d97820 84f40ff 6d97820 84f40ff 6d97820 84f40ff 6d97820 59ceb6d 139368d 6d97820 84f40ff |
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 |
import json
import os
from ast import literal_eval
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)
#df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
#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 raw_data, df
'''
def get_leaderboard_df(EVAL_REQUESTS_PATH, tasks) -> pd.DataFrame:
model_result_filepaths = []
for root,_, files in os.walk(EVAL_REQUESTS_PATH):
if len(files) == 0 or any([not f.endswith(".json") for f in files]):
continue
for file in files:
model_result_filepaths.append(os.path.join(root, file))
model_res = []
for model in model_result_filepaths:
import json
with open(model) as f:
model_res.append(json.load(f))
for model in model_res:
model["test"] = literal_eval(model["test"])
model["valid"] = literal_eval(model["valid"])
model["params"] = int(model["params"])
model['submitted_time'] = model['submitted_time'].split('T')[0]
#model['paper_url'] = '[Link](' + model['paper_url'] + ')'
#model['github_url'] = '[Link](' + model['github_url'] + ')'
name2short_name = {task.value.benchmark: task.value.col_name for task in tasks}
for model in model_res:
model.update({name2short_name[i]: str(model['test'][i][0])[:4] + '±' + str(model['test'][i][1])[:4] if i in model['test'] else '-' for i in name2short_name})
columns_to_show = ['model', 'author', 'email', 'paper_url', 'github_url', 'submitted_time', 'params'] + list(name2short_name.values())
# Check if model_res is empty
if len(model_res) > 0:
df_res = pd.DataFrame([{col: model[col] for col in columns_to_show} for model in model_res])
else:
# Initialize an empty DataFrame with the desired columns
df_res = pd.DataFrame(columns=columns_to_show)
#df_res = pd.DataFrame([{col: model[col] for col in columns_to_show} for model in model_res])
print(df_res)
ranks = df_res[list(name2short_name.values())].rank()
df_res.rename(columns={'model': 'Model', 'author': 'Author', 'email': 'Email', 'paper_url': 'Paper URL', 'github_url': 'Github URL', 'submitted_time': 'Time', 'params': '# of Params'}, inplace=True)
df_res['Average Rank⬆️'] = ranks.mean(axis=1)
df_res.sort_values(by='Average Rank⬆️', ascending=True, inplace=True)
return df_res
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(".")]
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]
|