Recursalberg / scripts /gutenberg_process.py
KaraKaraWitch's picture
Upload scripts/gutenberg_process.py with huggingface_hub
232f4f2 verified
raw
history blame
6.96 kB
import concurrent.futures as fut
import pathlib
import ftfy
import orjson
import re
import psutil
from bs4 import BeautifulSoup, Comment
import markdownify
from text_toolkit import normalize_puncts, clean_newlines, merge_consecutive_lines
class md_nolinks(markdownify.MarkdownConverter):
def convert_a(self, _, text, __):
_, _, text = markdownify.chomp(text)
if not text:
return ""
return text
def convert_img(self, el, text, convert_as_inline):
return ""
md = md_nolinks()
def blackbox(soup: pathlib.Path, filename: pathlib.Path):
p = psutil.Process()
p.cpu_affinity([i for i in range(0, 64)])
path = pathlib.Path(soup)
data_bytes = path.read_bytes()
try:
utf8 = data_bytes.decode("utf-8")
utf8, explain = ftfy.fix_and_explain(utf8)
except UnicodeDecodeError:
# print("Attempt Black magic...")
magic_data, magic_explain = ftfy.guess_bytes(data_bytes)
# print(magic_explain)
utf8, explain = ftfy.fix_and_explain(magic_data)
# Skip
soup = BeautifulSoup(utf8, "lxml")
pre = soup.find("pre")
[i.decompose() for i in soup.find_all("span", attrs={"class": "pagenum"})]
[i.extract() for i in soup.findAll(text=lambda text: isinstance(text, Comment))]
composed_content = []
if pre:
# Gutenberg pre html block
txt: str = pre.get_text()
# rgx = re.compile(r"\*\*+START\*\*\*+([\S\s]+)\*END\*")
if "gutenberg e" in txt.lower():
if not pre.parent:
raise Exception
has_pre = False
for child in pre.next_siblings:
if isinstance(child, str):
composed_content.append(child.lstrip())
if child.name == "pre":
# print(child)
has_pre = True
break
if child.name == "table":
continue
if child.name is None:
continue
child_filt = child.get_text().lower()
if (
child.name in ["div", "p"]
and "transcriber" in child_filt
and "note" in child_filt
):
continue
if "pagenum" in child.get("class", ""):
continue
composed_content.append(child)
else:
# For HR rules.
look_ahead = 5
sibs = []
rule = None
for sibling in pre.next_siblings:
if look_ahead > 0:
if sibling.name == "hr":
rule = sibling
break
else:
sibs.append(sibling)
if not rule:
raise Exception
else:
pass
for t_block in rule.next_siblings:
if isinstance(t_block, str):
composed_content.append(t_block.lstrip(" "))
continue
if not t_block.get_text().strip():
continue
if t_block.name == "hr" and "full" in t_block.get("class", ""):
break
gt = t_block.get_text().lower()
if t_block.name == "pre" and "gutenberg" in gt:
break
elif t_block.name == "p" and "end of" in gt and "gutenberg" in gt:
break
else:
composed_content.append(t_block)
else:
print("no pre", filename)
# We have to iter the book page it like some sort of caveman...
body = soup.find("body")
if not body:
raise Exception("Invisible man?")
# ???
pre_ctx = []
ctx = []
post_ctx = []
mode = 0
for children in body:
if isinstance(children, str):
continue
if isinstance(children, type(None)):
continue
gt = children.get_text()
if children.name == "table":
children.decompose()
if (
(children.name == "div" or children.name == "p")
and "start of" in gt.lower()
and "gutenberg" in gt.lower()
):
mode = 0.5
pre_ctx.append(children)
if (
(children.name == "div" or children.name == "p")
and "end of" in gt.lower()
and "gutenberg" in gt.lower()
and mode == 1
):
mode = 2
post_ctx.append(children)
if mode == 0:
pre_ctx.append(children)
elif mode == 0.5:
mode = 1
pre_ctx.append(children)
elif mode == 1:
ctx.append(children)
elif mode == 2:
post_ctx.append(children)
z_ctx = []
for pp in ctx:
if pp.name is None:
continue
z_ctx.append(pp)
composed_content = z_ctx
pure = []
for content in composed_content:
if isinstance(content, str):
pure.append(content)
continue
if not content.name:
continue
if "toc" in content.get("class", "") and content.name in ["ul", "ol"]:
continue
pure.append(content)
pp = normalize_puncts(
"\n".join(
[
md.convert_soup(soup_content)
if not isinstance(soup_content, str)
else soup_content
for soup_content in pure
]
)
)
pp = clean_newlines(pp)
pp = merge_consecutive_lines(pp)
pp = clean_newlines(pp)
pathlib.Path(filename).write_text(pp, encoding="utf-8")
return pp
pathlib.Path("gutenberg_processed").mkdir(exist_ok=True)
if __name__ == "__main__":
with fut.ProcessPoolExecutor(max_workers=64) as pool:
futures = []
for gutenbook in orjson.loads(pathlib.Path("final.json").read_bytes()):
path, f_type = gutenbook
if f_type == "html":
# print("processing", path)
path = pathlib.Path(path)
list_html = list(path.glob("*.htm"))
if len(list_html) > 1:
print(list_html)
list_html = list_html[0]
futures.append(
pool.submit(
blackbox,
list_html,
str(
pathlib.Path("gutenberg_processed")
/ list_html.with_suffix(".txt").name
),
)
)
print("Waiting for the future to arrive...")
fut.wait(futures)