Datasets:
File size: 4,054 Bytes
7a088d4 | 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 | """Audit retained Kanripo unknown-era texts without changing training inputs."""
import argparse
from collections import Counter
import csv
import json
from pathlib import Path
import re
import tarfile
from sentencepiece_unigram import sha, write_json, now
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--input',type=Path,default=Path('unified/pretrain-v1/build-002/pretrain.jsonl'))
parser.add_argument('--output',type=Path,required=True)
args = parser.parse_args()
args.output.mkdir(parents=True,exist_ok=False)
with open('kanripo/metadata/manifest.csv',encoding='utf-8-sig') as f:
catalog = {r['id']:r for r in csv.DictReader(f)}
sections = {}
with tarfile.open('kanripo/metadata/catalog.tar.gz') as archive:
for member in archive.getmembers():
if '/KR/' in member.name and member.isfile():
text = archive.extractfile(member).read().decode()
headings = re.findall(r'^\*\* (.+)$',text,re.M)
sections[member.name.split('/',1)[1]] = headings[0] if headings else ''
patterns = {
'post1949_year_mention':re.compile(r'(?<!\d)(1949|19[5-9]\d|20[0-2]\d)\s*年'),
'editorial_heading':re.compile(r'〔題解〕|整理者|校錄說明|整理說明'),
}
works, counts, prefixes = {}, Counter(), Counter()
total_files = 0
confirmed = None
with args.input.open() as f, (args.output/'documents.jsonl').open('w') as out:
for line_number,line in enumerate(f,1):
r = json.loads(line)
if r['source']!='kanripo' or r['date_evidence'].get('dynasty')!='時代不明':
continue
total_files += 1
rid = r['source_document_id'].split('/')[-1]
c = catalog[rid]
works[rid] = dict(id=rid,label=c['label'],catalog_file=c['catalog_file'],
section=sections.get(c['catalog_file']),reason=c['classification_source'],
dynasty_raw=c['dynasty_raw'])
prefixes[rid[:4]] += 1
hits = {}
for name,pattern in patterns.items():
matches = list(pattern.finditer(r['text']))
if matches:
hits[name] = [dict(match=m[0],offset=m.start(),
excerpt=r['text'][max(0,m.start()-100):m.end()+140]) for m in matches[:8]]
counts[name+'_files'] += 1
if rid=='KR6v0002':
m = re.search(r'1995年3月8日',r['text'])
if m:
confirmed = dict(id=r['id'],line=line_number,text_sha256=r['text_sha256'],
excerpt=r['text'][max(0,m.start()-50):m.end()+180])
row = dict(id=r['id'],work_id=rid,input_line=line_number,text_sha256=r['text_sha256'],
catalog_section=works[rid]['section'],hits=hits)
out.write(json.dumps(row,ensure_ascii=False)+'\n')
with (args.output/'works.jsonl').open('w') as f:
for row in sorted(works.values(),key=lambda r:r['id']):
f.write(json.dumps(row,ensure_ascii=False)+'\n')
qing_works = [r for r in works.values() if r['catalog_file']=='KR/KR5i.txt']
report = dict(complete=True,finished=now(),input_sha256=sha(args.input),script_sha256=sha(__file__),
files=total_files,works=len(works),files_by_prefix=dict(prefixes),
works_by_prefix=dict(Counter(r['id'][:4] for r in works.values())),
candidate_counts=dict(counts),qing_section_work_count=len(qing_works),
confirmed_modern_editorial_example=confirmed,
limits='Year/heading matches are review candidates, not composition-date certification. Modern metadata DATE lines are not scanned; only retained training text is searched.')
write_json(args.output/'report.json',report)
print(json.dumps(report,ensure_ascii=False,indent=2),flush=True)
if __name__=='__main__':
main()
|