DatasetCardForm / formatting /reformat_json.py
Sebastian Gehrmann
.
75a8362
import argparse
import json
import pathlib
import os
parser = argparse.ArgumentParser(
description="Format the output of the data card tool as .md for the hub."
)
parser.add_argument("--input_path", "-i", type=pathlib.Path, required=False)
parser.add_argument("--output_path", "-o", type=pathlib.Path, required=False)
args = parser.parse_args()
def read_json_file(json_path: pathlib.Path):
"""Load a json file and return it as object."""
with open(json_path, "r") as f:
data = json.load(f)
return data
def save_file(json_path: pathlib.Path, json_obj: str):
"""Takes a string and saves it as .md file."""
with open(json_path, "w") as f:
f.write(json.dumps(json_obj, indent=2))
def construct_json(dataset_name: str, data_card_data: dict, text_by_key: dict):
"""Constructs the json file
This function iterates through text_by_key and extracts all answers from
the data_card_data object. It uses the levels of hierarchy as indicator for
the heading indentation and does not change the order in which anything
appears.
Args:
data_card_data: Output from the data card tool
text_by_key: configuration defined in key_to_question.json
Returns:
data_card_md_string: json content
"""
try:
website_link = data_card_data["overview"]["where"]["website"]
except KeyError:
website_link = ""
try:
paper_link = data_card_data["overview"]["where"]["paper-url"]
except KeyError:
paper_link = ""
try:
authors = data_card_data["overview"]["credit"]["creators"]
except KeyError:
authors = ""
try:
summary = data_card_data["overview"]["what"]["dataset"]
except KeyError:
summary = "Placeholder"
# Add summary blurb with loading script and link to GEM loader
summary +=f"\n\nYou can load the dataset via:\n```\nimport datasets\ndata = datasets.load_dataset('GEM/{dataset_name}')\n```\nThe data loader can be found [here](https://huggingface.co/datasets/GEM/{dataset_name})."
new_json = {
"name": dataset_name,
"summary": summary,
"sections": [
],
}
if website_link:
new_json["website"] = website_link
if paper_link:
new_json["paper"] = paper_link
if authors:
new_json["authors"] = authors
total_questions = 0
total_words = 0
for main_key, main_content in text_by_key.items():
l2_data = {
"title": main_content["section-title"],
"level": 2,
"subsections": []
}
if main_key not in data_card_data:
continue
for second_key, second_content in main_content.items():
if second_key == "section-title":
continue
# Skip summary data since it is already in the header.
if main_key == "overview" and second_key == "what":
continue
l3_data = {
"title": second_content["section-title"],
"level": 3,
"fields": []
}
for final_key, final_content in second_content.items():
if final_key == "section-title":
continue
try:
total_questions += 1
answer = data_card_data[main_key][second_key].get(final_key, "N/A")
except:
# print(main_key, second_key, final_key)
# print("==="*50)
# print(data_card_data)
continue
# Skip empty answers.
if isinstance(answer, str):
if answer.lower() == "n/a":
continue
if not answer:
continue
if isinstance(answer, list):
answer = ", ".join([f"`{a}`" for a in answer])
json_answer = {
"title": final_content["title"],
"level": 4,
"content": answer,
"flags": final_content["flags"],
"info": final_content["info"],
"scope": final_content["scope"],
}
total_words += len(answer.split())
l3_data["fields"].append(json_answer)
l2_data["subsections"].append(l3_data)
new_json["sections"].append(l2_data)
print(f"Total questions {total_questions}")
print(f"total words: {total_words}")
return new_json, total_words
if __name__ == "__main__":
text_by_key = read_json_file(
os.path.join(os.path.dirname(__file__), "key_to_question.json")
)
total_words_across_everything = 0
for dataset in os.listdir("../../../GEMv2"):
data_card_path = f"../../../GEMv2/{dataset}/{dataset}.json"
if os.path.exists(data_card_path):
print(f"Now processing {dataset}.")
new_path = f"datacards/{dataset}.json"
data_card_data = read_json_file(data_card_path)
data_card_json, total_cur_words = construct_json(dataset, data_card_data, text_by_key)
total_words_across_everything += total_cur_words
save_file(new_path, data_card_json)
else:
print(f"{dataset} has no data card!")
print(total_words_across_everything)
# data_card_json = construct_json(data_card_data, text_by_key)
# save_file(args.output_path, data_card_json)