Datasets:
File size: 10,660 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 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 | #!/usr/bin/env python3
"""Clean an inventoried Kanripo snapshot without modifying raw files."""
import argparse
from collections import Counter, defaultdict
from contextlib import ExitStack
import csv
from datetime import datetime, timezone
import hashlib
import json
from pathlib import Path
import re
from kanripo_profile import dump
RULES = {
'dynasty': 'Exclude catalog dynasty 清/淸; unknown remains explicitly unknown',
'selection': 'Completed non-Qing catalog entries, inventoried KR_ID_digits[letter].txt only',
'management': 'Remove metadata/comment lines, pb/md/PB, pilcrow, fold/image tags, Org delimiters; keep link labels and heading text',
'body': 'Keep historical annotations, parentheses, slashes, missing characters, original glyphs and whitespace',
'unknown': 'Hold files with unresolved management syntax; never guess deletion ranges',
'unit': 'One source text file per record; no concatenation, deduplication, split or tokenize',
}
FLAGS = re.MULTILINE
LINE = r'[^\r\n]*(?:\r?\n|$)'
PATTERNS = [
('metadata_lines', re.compile(r'^[ \t]*#+\+[A-Za-z][A-Za-z0-9_-]*' + LINE, FLAGS), ''),
('mode_lines', re.compile(r'^[ \t]*#+[ \t]*-\*-'+ LINE, FLAGS), ''),
# A section title can be prefixed by an editorial Roman-numeral marker.
('roman_heading_prefix', re.compile(r'^[ \t]*#+[ \t]*[ivxlcdm]+#(?=<)', FLAGS | re.I), ''),
('comment_lines', re.compile(r'^[ \t]*#+[ \t]*[A-Za-z]' + LINE, FLAGS), ''),
('legacy_page_comments', re.compile(r'^[ \t]*%[ \t]*BR\d+(?:\.\d+)*[a-z]?[^\r\n]*(?:\r?\n|$)', FLAGS), ''),
('page_markers', re.compile(r'<(?:pb|md)(?::|[ \t]+)[^<>\r\n]*>', re.I), ''),
('fold_markers', re.compile(r'<折[ \t]+t="\d+"[ \t]*/>'), ''),
('image_markers', re.compile(r'<img:[^<>\r\n]*>', re.I), ''),
('org_heading_prefix', re.compile(r'^\*+[ \t]+', FLAGS), ''),
# Require a link type: [[...]] can also be nested character descriptions.
('org_labelled_links', re.compile(r'\[\[[A-Za-z][A-Za-z0-9_-]*:[^\[\]]{0,2000}\]\[([^\[\]]{0,2000})\]\]'), r'\1'),
('non_ascii_directive_prefix', re.compile(r'^[ \t]*#+\+(?=[\u3400-\u9fff])', FLAGS), ''),
('hash_heading_marks', re.compile(r'^[ \t]*#+|#+[ \t]*(?=\r?$)', FLAGS), ''),
('percent_page_prefix', re.compile(r'^[ \t]*%(?=¶?[ \t]*\r?$)', FLAGS), ''),
('pilcrow', re.compile('¶'), ''),
]
RESIDUAL = re.compile(r'<(?:pb|md|img)(?=[:\s>])|<折\s|\[\[[A-Za-z][A-Za-z0-9_-]*:|^\s*#\+|^%[ \t]*BR\d', re.M | re.I)
def sha(data):
return hashlib.sha256(data).hexdigest()
def clean(original):
counts = Counter()
text = original
if text.startswith('\ufeff'):
text = text[1:]
counts['initial_bom'] = 1
for name, pattern, replacement in PATTERNS:
text, n = pattern.subn(replacement, text)
if n:
counts[name] += n
issues = []
for match in RESIDUAL.finditer(text):
if len(issues) < 8:
issues.append(text[max(0, match.start()-35):match.end()+100])
return text, counts, issues
def load_inputs(root, audit):
audit_manifest = json.loads((audit / 'manifest.json').read_text())
if not audit_manifest['complete']:
raise ValueError('incomplete inventory')
data = (root / 'metadata/manifest.csv').read_bytes()
if sha(data) != audit_manifest['source_manifest_sha256']:
raise ValueError('source manifest changed since inventory')
rows = list(csv.DictReader(data.decode('utf-8-sig').splitlines()))
catalog = {r['id']: r for r in rows}
if len(catalog) != len(rows):
raise ValueError('duplicate catalog IDs')
inventory_data = (audit / 'files.jsonl').read_bytes()
files = [json.loads(line) for line in inventory_data.decode('utf-8').splitlines()]
if len(files) != len({r['raw_path'] for r in files}):
raise ValueError('duplicate inventory paths')
expected = {r['id'] for r in rows if r['status'] == 'complete' and r['dynasty'] not in {'清', '淸'}}
report = json.loads((audit / 'report.json').read_text())
if not {r['id'] for r in files} <= expected or len(files) != report['totals']['body_files']:
raise ValueError('inventory work coverage mismatch')
return catalog, files, data, inventory_data
def source_path(root, row):
path = root / row['raw_path']
rid = row['id']
if (not re.fullmatch(r'KR[1-6][a-z]\d{4}', rid)
or not re.fullmatch(re.escape(rid) + r'_\d+[a-z]?\.txt', path.name)
or path.parent != root / 'raw' / rid
or path.is_symlink() or path.parent.is_symlink()
or not path.resolve().is_relative_to(root / 'raw')):
raise ValueError(f'unsafe source path: {path}')
return path
def build(root, audit, output):
root, audit, output = root.resolve(), audit.resolve(), output.resolve()
catalog, files, metadata, inventory = load_inputs(root, audit)
if root.is_relative_to(output) or output.is_relative_to(root / 'raw') or output == audit:
raise ValueError('unsafe output directory')
output.mkdir(parents=True, exist_ok=False)
manifest = {'complete': False, 'started': datetime.now(timezone.utc).isoformat(),
'root': str(root), 'audit': str(audit), 'rules': RULES,
'source_manifest_sha256': sha(metadata), 'inventory_sha256': sha(inventory),
'script_sha256': sha(Path(__file__).read_bytes())}
dump(output / 'manifest.json', manifest)
totals, removed, dynasties = Counter(), Counter(), Counter()
exact = defaultdict(list)
selected_works, review_works = set(), set()
sample_counts = Counter()
with ExitStack() as stack:
out = {name: stack.enter_context((output / (name + '.jsonl')).open('w', encoding='utf-8'))
for name in ['documents', 'excluded', 'edits', 'review', 'preview', 'files']}
def emit(name, obj):
out[name].write(json.dumps(obj, ensure_ascii=False) + '\n')
body_works = {entry['id'] for entry in files}
for row in catalog.values():
reasons = []
if row['dynasty'] in {'清', '淸'}:
reasons.append('qing_user_excluded')
totals['qing_catalog_works'] += 1
if row['status'] == 'complete':
totals['qing_complete_works'] += 1
if row['status'] != 'complete':
reasons.append('download_not_complete')
totals['incomplete_catalog_works'] += 1
elif row['dynasty'] not in {'清', '淸'} and row['id'] not in body_works:
reasons.append('no_body_files')
totals['no_body_works'] += 1
if reasons:
emit('excluded', dict(id=row['id'], dynasty=row['dynasty'], label=row['label'], reasons=reasons))
totals['excluded_catalog_works'] += 1
for i, entry in enumerate(files):
rid = entry['id']
row = catalog[rid]
if row['status'] != 'complete' or row['dynasty'] in {'清', '淸'}:
raise ValueError(f'ineligible inventory entry: {rid}')
data = source_path(root, entry).read_bytes()
if sha(data) != entry['raw_sha256']:
raise ValueError(f'raw input changed: {entry["raw_path"]}')
original = data.decode('utf-8')
text, counts, issues = clean(original)
totals['input_files'] += 1
totals['original_chars'] += len(original)
removed.update(counts)
file_id = 'kanripo/' + rid + '/' + Path(entry['raw_path']).name
base = dict(id=file_id, source='kanripo', work_id=rid, label=row['label'],
dynasty=row['dynasty'], dynasty_raw=row['dynasty_raw'],
classification_source=row['classification_source'],
raw_path=entry['raw_path'], raw_sha256=entry['raw_sha256'])
audit_row = dict(base, original_chars=len(original), clean_chars=len(text), removed=counts,
clean_sha256=sha(text.encode('utf-8')))
if issues:
status = 'review'
emit('review', dict(audit_row, issues=issues))
review_works.add(rid)
elif not text.strip():
status = 'empty'
else:
status = 'included'
header = re.findall(r'^#+\+([^\r\n]+)', original, re.M)
emit('documents', dict(base, text=text, clean_sha256=audit_row['clean_sha256'],
chars=len(text), original_header=header))
exact[audit_row['clean_sha256']].append(file_id)
selected_works.add(rid)
totals['clean_chars'] += len(text)
dynasties[row['dynasty']] += len(text)
audit_row['status'] = status
emit('files', audit_row)
totals[status + '_files'] += 1
if counts:
emit('edits', audit_row)
for feature in counts:
if sample_counts[feature] < 2:
emit('preview', dict(id=file_id, feature=feature, before=original[:2500], after=text[:2500]))
sample_counts[feature] += 1
if (i+1) % 10000 == 0:
print(f'processed {i+1}/{len(files)}', flush=True)
duplicates = [dict(sha256=key, ids=ids) for key, ids in exact.items() if len(ids) > 1]
with (output / 'duplicates.jsonl').open('w', encoding='utf-8') as f:
for group in duplicates:
f.write(json.dumps(group, ensure_ascii=False) + '\n')
totals['included_works'] = len(selected_works)
totals['review_works'] = len(review_works)
totals['duplicate_groups'] = len(duplicates)
if sha((root / 'metadata/manifest.csv').read_bytes()) != sha(metadata):
raise ValueError('source manifest changed during build')
dump(output / 'report.json', dict(totals=totals, removed=removed, chars_by_dynasty=dynasties))
manifest.update(complete=True, finished=datetime.now(timezone.utc).isoformat(),
output_hashes={p.name: sha(p.read_bytes()) for p in output.glob('*.jsonl') if p.name != 'documents.jsonl'})
dump(output / 'manifest.json', manifest)
print(json.dumps(totals, ensure_ascii=False), flush=True)
if __name__ == '__main__':
p = argparse.ArgumentParser(description=__doc__)
p.add_argument('--root', type=Path, default=Path('kanripo'))
p.add_argument('--audit', type=Path, default=Path('kanripo/preprocessed-v1/inspection-002'))
p.add_argument('--output', type=Path, required=True)
a = p.parse_args()
build(a.root, a.audit, a.output)
|