sefaria / scrape_script.py
tomersagi's picture
added cleaning processes: html markup removal, skip non en/he languages
0f77462
import os
import sys
import pandas as pd
import pyarrow as pa
import pyarrow.parquet as pq
import json
import re
from langdetect import detect
def traverse_directory(root_path, callback):
for dirpath, _, filenames in os.walk(root_path):
for filename in filenames:
file_path = os.path.join(dirpath, filename)
callback(file_path)
def process_file(file_path):
if not file_path.endswith(".txt"):
return
with open(file_path, "r", encoding="utf-8") as file:
content = file.read()
dirname = os.path.dirname(file_path)
dir_name = os.path.basename(dirname)
top_level_directory = os.path.relpath(dirname, root_directory).split(os.sep)[0]
if dir_name.lower() == "english":
append_to_parquet(content, file_path, "en", top_level_directory)
elif dir_name.lower() == "hebrew":
append_to_parquet(content, file_path, "he", top_level_directory)
def append_to_parquet(content, file_path, lang, top_level_directory, data_dir = "data"):
if not os.path.exists(data_dir):
os.makedirs(data_dir)
if lang == "en":
parquet_file = os.path.join(data_dir, f"train_{top_level_directory}_english.parquet")
elif lang == "he":
parquet_file = os.path.join(data_dir, f"train_{top_level_directory}_hebrew.parquet")
else:
return
# Check if file_path ends with the pattern [xx].txt
file_pattern = re.search(r'\[[a-zA-Z]{2}\]\.txt$', file_path)
if file_pattern:
print(f"Warning: File '{file_path}' was skipped due to the detected [xy] pattern.")
return
# Check if the content is in English when lang is "en"
if lang == "en":
sample_text = content[:500] if len(content) > 500 else content
detected_lang = detect(sample_text)
if detected_lang != 'en' and detected_lang != 'id':
print(f"Warning: Non-English content detected in file '{file_path}'. Detected language: {detected_lang}")
return
# Apply cleaning rules
content = re.sub(r'<(?:span|b|big|small|strong|br|sup[^>]*)[^>]*>|</(?:span|b|big|small|strong|sup)>', '', content) # Remove HTML tags
content = re.sub(r'https?://\S+', '', content) # Remove HTML links
# Remove Hebrew cantillation marks
if lang == "he":
content = re.sub(r'[\u0591-\u05AF]', '', content)
# Remove chapter markers
chapter_markers = ['Chapter', 'Halakhah','Siman','Mitzvah']
for marker in chapter_markers:
content = re.sub(rf'^{marker} +\d+$', '', content, flags=re.MULTILINE)
metadata = {"file": file_path}
meta_json = json.dumps(metadata)
data = pd.DataFrame({"meta": [meta_json], "text": [content]})
table = pa.Table.from_pandas(data)
if not os.path.exists(parquet_file) or os.path.getsize(parquet_file) == 0:
with pq.ParquetWriter(parquet_file, table.schema, compression="snappy") as writer:
writer.write_table(table)
else:
pf = pq.ParquetFile(parquet_file)
old_table = pf.read()
combined_table = pa.concat_tables([old_table, table])
with pq.ParquetWriter(parquet_file, combined_table.schema, compression="snappy") as writer:
writer.write_table(combined_table)
print(f"Successfully saved: {file_path}")
if __name__ == "__main__":
if len(sys.argv) != 2:
print("Usage: python script.py <root_directory_path>")
sys.exit(1)
root_directory = sys.argv[1]
traverse_directory(root_directory, process_file)