StackingExchange / scripts /stack_parser.py
KaraKaraWitch's picture
Upload scripts/stack_parser.py with huggingface_hub
d2566f6 verified
raw
history blame
16.8 kB
import multiprocessing
import random
import traceback
from lxml import etree
import pathlib
import orjson
import tqdm
import typer
import urllib.parse
from sqlitedict import SqliteDict
app = typer.Typer()
def fast_iter(context, func, *args, **kwargs):
"""
http://lxml.de/parsing.html#modifying-the-tree
Based on Liza Daly's fast_iter
http://www.ibm.com/developerworks/xml/library/x-hiperfparse/
See also http://effbot.org/zone/element-iterparse.htm
"""
for event, elem in context:
func(elem, *args, **kwargs)
# It's safe to call clear() here because no descendants will be
# accessed
elem.clear()
# Also eliminate now-empty references from the root node to elem
for ancestor in elem.xpath("ancestor-or-self::*"):
while ancestor.getprevious() is not None:
del ancestor.getparent()[0]
del context
@app.command()
def decadence(stack_folder: pathlib.Path, output_file: pathlib.Path):
post_history = stack_folder / "PostHistory.xml"
if not post_history.exists():
raise Exception()
# posts_hist = {}
print("Working on", stack_folder.name)
sql_dict_posts = SqliteDict(
f"tmp/{stack_folder.name}.hist.tmp.sqlite", flag="n", tablename="dict_hist"
)
try:
with tqdm.tqdm() as pbar:
ctx = etree.iterparse(post_history, tag="row")
def element_processor2(elem):
p_type = int(elem.get("PostHistoryTypeId"))
pid: str = elem.get("PostId")
if p_type in range(1, 4):
if pid in sql_dict_posts:
post_data = sql_dict_posts[pid]
else:
post_data = sql_dict_posts[pid] = {}
if p_type == 1:
post_data["title"] = urllib.parse.unquote(
elem.get("Text", "")
).replace("\r\n", "\n")
elif p_type == 2:
post_data["body"] = urllib.parse.unquote(
elem.get("Text", "")
).replace("\r\n", "\n")
elif p_type == 3:
post_data["tags"] = urllib.parse.unquote(
elem.get("Text", "")
).replace("\r\n", "\n")
# print(post_data)
sql_dict_posts[pid] = post_data
pbar.update(1)
elif p_type in range(4, 10):
if pid in sql_dict_posts:
post_data = sql_dict_posts[pid]
else:
# print(f"[W] {pid}, {stack_folder.name} appears to be a partial?")
post_data = sql_dict_posts[pid] = {}
post_data["partial"] = True
if p_type in [4, 7]:
post_data["title"] = urllib.parse.unquote(
elem.get("Text", "")
).replace("\r\n", "\n")
elif p_type in [5, 8]:
post_data["body"] = urllib.parse.unquote(
elem.get("Text", "")
).replace("\r\n", "\n")
elif p_type in [6, 9]:
post_data["tags"] = urllib.parse.unquote(
elem.get("Text", "")
).replace("\r\n", "\n")
sql_dict_posts[pid] = post_data
pbar.update(1)
elif p_type == 35:
if pid in sql_dict_posts:
post_data = sql_dict_posts[pid]
else:
print(
f"[W] {pid}, {stack_folder.name} requested to be redirected but redirect doesn't seem to exist?"
)
post_data = {}
if not post_data.get("partial", False):
print(f"[I] {pid}, {stack_folder.name} Not partial?")
return
print(post_data)
elif p_type == 10:
if pid not in sql_dict_posts:
return
post_data = sql_dict_posts[pid]
if not post_data:
raise Exception
post_data["closed"] = True
sql_dict_posts[pid] = post_data
pbar.update(1)
elif p_type == 11:
if pid not in sql_dict_posts:
return
post_data = sql_dict_posts[pid]
if not post_data:
raise Exception
post_data["closed"] = False
sql_dict_posts[pid] = post_data
pbar.update(1)
elif p_type == 12:
if pid not in sql_dict_posts:
return
post_data = sql_dict_posts[pid]
if not post_data:
raise Exception
post_data["deleted"] = True
sql_dict_posts[pid] = post_data
pbar.update(1)
elif p_type == 13:
if pid not in sql_dict_posts:
return
post_data = sql_dict_posts[pid]
if not post_data:
raise Exception
post_data["deleted"] = False
sql_dict_posts[pid] = post_data
pbar.update(1)
elif p_type == 14:
if pid not in sql_dict_posts:
return
post_data = sql_dict_posts[pid]
if not post_data:
raise Exception
post_data["locked"] = True
sql_dict_posts[pid] = post_data
pbar.update(1)
elif p_type == 15:
if pid not in sql_dict_posts:
return
post_data = sql_dict_posts[pid]
if not post_data:
raise Exception
post_data["locked"] = False
sql_dict_posts[pid] = post_data
pbar.update(1)
# print(etree.tostring(element))
if pbar.n % 5000 == 0 and pbar.n != 0:
# print(sql_dict_posts.conn.reqs.qsize())
sql_dict_posts.commit()
# elem.clear()
fast_iter(ctx, element_processor2)
except Exception as e:
print("[!] ERR: ", traceback.format_exception(e))
flagged = set()
for k, v in sql_dict_posts.items():
if not v.get("body").strip():
flagged.add(k)
if v.get("deleted", False):
flagged.add(k)
for fg in flagged:
del sql_dict_posts[fg]
sql_dict_posts.sync()
print("Reconstruct done!")
posts = stack_folder / "Posts.xml"
if not posts.exists():
raise Exception()
# reconstruct = {}
main_qn_posts = {}
try:
# Prepare all the Main Question posts
with tqdm.tqdm() as bar:
ctx2 = etree.iterparse(posts, tag="row")
def element_processor(element):
pid = element.get("Id")
p_type = int(element.get("PostTypeId"))
if p_type == 1:
if pid not in sql_dict_posts:
main_qn_posts[pid] = {"Stub":True}
print(
f"[!] Question: {int(element.get('Id'))} {stack_folder.name} does not exist but referanced?"
)
return
main_qn_posts[pid] = sql_dict_posts[pid]
main_qn_posts[pid]["Stub"] = False
main_qn_posts[pid]["Id"] = pid
main_qn_posts[pid]["CreationDate"] = element.get("CreationDate")
main_qn_posts[pid]["Score"] = int(element.get("Score"))
main_qn_posts[pid]["Accepted"] = (
int(element.get("AcceptedAnswerId"))
if element.get("AcceptedAnswerId")
else None
)
main_qn_posts[pid]["Counts"] = {
"Views": int(element.get("ViewCount"))
if element.get("ViewCount")
else 0,
"Answers": int(element.get("AnswerCount"))
if element.get("AnswerCount")
else 0,
"Comments": int(element.get("CommentCount"))
if element.get("CommentCount")
else 0,
}
bar.update(1)
fast_iter(ctx2, element_processor)
except Exception as e:
print("[!] ERR: ", traceback.format_exception(e))
print("2nd Pass Posts Done")
# Match all the Answer posts
try:
with tqdm.tqdm() as bar:
ctx3 = etree.iterparse(posts, tag="row")
def element_processor3(element):
pid = element.get("Id")
p_type = int(element.get("PostTypeId"))
if p_type == 2:
parent_id = element.get("ParentId")
if parent_id not in main_qn_posts:
print(
f"[!] Answer: {int(element.get('Id'))} {stack_folder.name} has no parent attached to it!"
)
return
answers = main_qn_posts[parent_id].setdefault("answers", [])
if pid in sql_dict_posts:
rec_answer = sql_dict_posts[pid]
else:
rec_answer = None
if rec_answer is None:
print(
f"[!] Answer: {int(element.get('Id'))} {stack_folder.name} does not have a PostHistory.xml Assoc. with it!"
)
main_qn_posts[parent_id]["answers"] = answers
return
rec_answer["Id"] = int(element.get("Id"))
rec_answer["Score"] = (
int(element.get("Score")) if element.get("Score") else 0
)
rec_answer["Counts"] = {
"Views": int(element.get("ViewCount"))
if element.get("ViewCount")
else 0,
"Comments": int(element.get("CommentCount"))
if element.get("CommentCount")
else 0,
}
answers.append(rec_answer)
main_qn_posts[parent_id]["answers"] = answers
fast_iter(ctx3, element_processor3)
except Exception as e:
print("[!] ERR: ", traceback.format_exception(e))
print("2nd Pass done!")
try:
with open(output_file, "wb") as f:
for k, v in main_qn_posts.items():
f.write(orjson.dumps(v) + b"\n")
sql_dict_posts.close()
pathlib.Path(f"tmp/{stack_folder.name}.hist.tmp.sqlite").unlink()
except Exception as e:
print("[!] ERR: ", traceback.format_exception(e))
@app.command()
def convo_processor(base_file: pathlib.Path, qa_file: pathlib.Path):
with open(qa_file, "wb") as fw:
with open(base_file, "rb") as f:
for line in f:
data = orjson.loads(line)
if not data.get("body") or not data.get("title"):
continue
if data["body"].lower().startswith(data["title"].lower()):
question = f'{data["title"].strip()}\n\n{data["body"][len(data["title"]) :].strip()}'
else:
question = f'{data["title"].strip()}\n\n{data["body"].strip()}'
if data.get("answers"):
for answer in data.get("answers"):
if "Score" not in answer:
print("no score?",data)
continue
if answer["Score"] >= 0:
zz = {
"conversation": [
{"sender": "user", "message": question},
{"sender": "assistant", "message": answer},
],
"meta": {
"q_score": data["Score"],
"a_score": answer["Score"],
"s_score": data["Score"] + answer["Score"]
},
}
fw.write(orjson.dumps(zz) + b"\n")
pass
@app.command()
def qa_processor(base_file: pathlib.Path, qa_file: pathlib.Path, formats: pathlib.Path):
question_formats = orjson.loads(formats.read_bytes())
with open(qa_file, "wb") as fw:
with open(base_file, "rb") as f:
for line in f:
data = orjson.loads(line)
if not data.get("body") or not data.get("title"):
continue
if data["body"].lower().startswith(data["title"].lower()):
question = f'{data["title"].strip()}\n\n{data["body"][len(data["title"]) :].strip()}'
else:
question = f'{data["title"].strip()}\n\n{data["body"].strip()}'
if data.get("answers"):
for answer in data.get("answers"):
if answer["Score"] >= 0:
fmt = random.choice(question_formats)
fmt = fmt.format(question=question, answer=answer["body"])
zz = {
"text": fmt,
"meta": {
"q_score": data["Score"],
"a_score": answer["Score"],
},
}
fw.write(orjson.dumps(zz) + b"\n")
pass
def err_handler(e):
traceback.print_exception(e)
@app.command()
def convo_stackstack(folder: pathlib.Path):
with multiprocessing.Pool(processes=64) as pool:
fn = []
for item in folder.iterdir():
if item.is_file() and item.suffix.endswith("jsonl"):
# print(item)
fn.append(
pool.apply_async(
convo_processor,
(
item,
pathlib.Path("convo")
/ pathlib.Path(f"{item.stem}_convo.jsonl")
),
error_callback=err_handler)
)
for f in fn:
f.wait()
pool.close()
pool.join
@app.command()
def qa_stackstack(folder: pathlib.Path):
with multiprocessing.Pool(processes=64) as pool:
fn = []
for item in folder.iterdir():
if item.is_file() and item.suffix.endswith("jsonl"):
# print(item)
fn.append(
pool.apply_async(
qa_processor,
(
item,
pathlib.Path("qa")
/ pathlib.Path(f"{item.stem}_qa.jsonl"),
pathlib.Path("staccato_format.json")
),
error_callback=err_handler)
)
for f in fn:
f.wait()
pool.close()
pool.join()
@app.command()
def stackstack(folder: pathlib.Path):
with multiprocessing.Pool(processes=64) as pool:
fn = []
for item in folder.iterdir():
if item.is_dir() and "meta." not in item.name:
# print(item)
fn.append(
pool.apply_async(
decadence,
(
item,
pathlib.Path("parsed")
/ pathlib.Path(f"{item.name}_raw.jsonl"),
),
)
)
for f in fn:
f.wait()
pool.close()
pool.join()
app()