File size: 3,805 Bytes
be62d39 |
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 |
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.filter_models import filter_models
from src.leaderboard.read_evals import get_raw_eval_results, EvalResult
'''
This function, get_leaderboard_df, is designed to read and process evaluation results from a specified results path and requests path,
ultimately producing a leaderboard in the form of a pandas DataFrame. The process involves several steps, including filtering, sorting,
and cleaning the data based on specific criteria. Let's break down the function step by step:
'''
## TO-DO: if raw_data is [], return dummy df with correct columns so that the UI shows the right columns
def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> tuple[list[EvalResult], pd.DataFrame]:
print(f"results_path = {results_path}")
raw_data = get_raw_eval_results(results_path, requests_path)
all_data_json = [v.to_dict() for v in raw_data] # if v.is_complete()]
# all_data_json.append(baseline_row)
filter_models(all_data_json)
print(f"all_data_json = {all_data_json}")
df = pd.DataFrame.from_records(all_data_json)
task_attributes = []
# Iterate over all attributes of AutoEvalColumn class
for attr_name in dir(AutoEvalColumn):
# Retrieve the attribute object
attr = getattr(AutoEvalColumn, attr_name)
# Check if the attribute has 'is_task' attribute and it is True
if hasattr(attr, 'is_task') and getattr(attr, 'is_task'):
task_attributes.append(attr)
# Now task_attributes contains all attributes where is_task=True
# print(task_attributes)
task_col_names_all = [str(item.name) for item in task_attributes]
# import pdb; pdb.set_trace()
# Add empty columns with specified names
for col_name in task_col_names_all:
if col_name not in df.columns:
df[col_name] = None
return raw_data, df
def get_evaluation_queue_df(save_path: str, cols: list) -> tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame]:
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]
|