qualify_tool.py 4.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081
  1. """Reproducible comparison with the supplied legacy parser and ingestion benchmark.
  2. Usage: .venv/bin/python scripts/qualify_tool.py --legacy ../reference_20261117/component_trade_promax --rows 100000
  3. All generated data is synthetic. Timings do not estimate production throughput.
  4. """
  5. import argparse
  6. import csv
  7. import importlib.util
  8. import json
  9. import sys
  10. import time
  11. from pathlib import Path
  12. from decimal import Decimal
  13. ROOT=Path(__file__).resolve().parents[1]
  14. sys.path.insert(0,str(ROOT/'tools-service'))
  15. from toolsvc.customs.engine.rules import Rules
  16. from toolsvc.customs.engine.storage import ingest, Store
  17. def main():
  18. parser=argparse.ArgumentParser()
  19. parser.add_argument('--legacy',type=Path,required=True)
  20. parser.add_argument('--rows',type=int,default=100000)
  21. parser.add_argument('--before',type=Path)
  22. args=parser.parse_args()
  23. out=ROOT/'.run/qualification'; out.mkdir(parents=True,exist_ok=True)
  24. folder=ROOT/'tools-service/toolsvc/customs/rules'
  25. mapping=folder/'数据映射规则.xlsx'; fields=folder/'进出口数据字段需求20260819.xlsx'
  26. sys.path.insert(0,str(args.legacy.resolve()))
  27. from trade0909.rules import Rules as LegacyRules
  28. legacy=LegacyRules(mapping, fields); current=Rules(mapping, fields)
  29. # Independent baseline: agent implementation before this change.
  30. original=args.before
  31. before=None
  32. if original and original.exists():
  33. spec=importlib.util.spec_from_file_location('before_rules',original)
  34. mod=importlib.util.module_from_spec(spec); spec.loader.exec_module(mod)
  35. before=mod.Rules(mapping, fields)
  36. texts=[
  37. '4|0|多层|陶瓷|MURATA|TEST-1|||',
  38. '品牌:MURATA;型号:TEST-1', '品牌名称:村田;型号规格:TEST-1',
  39. '村田(MURATA)|TEST-1', 'MURATA/TEST-1/计算机用',
  40. 'MURATA_TEST-1', '0 0 多层片式 瓷介 MURATA TEST-1',
  41. '品牌为MURATA,型号为TEST-1', '(品牌)MURATA(型号)TEST-1',
  42. '品牌|MURATA|型号|TEST-1', '品牌:MURTA;型号:TEST-1',
  43. '品牌:MURATA,SAMSUNG;型号:TEST-1',
  44. '品牌:ABACUS;型号:A5E30295495R2W', '品牌:无品牌;型号:无型号',
  45. '品牌:MURATA;型号:1/10W', '品牌:MURATA;用途:5G基站',
  46. ]
  47. records=[]
  48. keys=['brand','model','brand_country','c1','c2','c3','c4','app','mapped']
  49. for text in texts:
  50. old=legacy.parse('8532241000',text); new=current.parse('8532241000',text)
  51. prev=before.parse('8532241000',text) if before else None
  52. records.append(dict(raw=text,legacy={k:old[k] for k in keys},current={k:new[k] for k in keys},
  53. legacy_equal=all(old[k]==new[k] for k in keys),
  54. before_pair=[prev['brand'],prev['model']] if prev else None))
  55. source=out/'synthetic.csv'
  56. with source.open('w',encoding='utf-8-sig',newline='') as f:
  57. w=csv.writer(f);w.writerow(['I_E_FLAG','HS_CODE','YEAR','WEEK','MONORSUM','COUNT','G_MODEL'])
  58. for i in range(args.rows):
  59. w.writerow(['I','8532241000','2025',str(i%52+1),'0.10','1',texts[i%len(texts)]])
  60. bench=out/f'bench-{time.time_ns()}';bench.mkdir()
  61. start=time.perf_counter()
  62. db,stats=ingest(dict(input_csv=str(source),directions=['I'],start_year=2025,end_year=2026,cutoff_date='2026-12-31'),current,bench)
  63. elapsed=time.perf_counter()-start
  64. totals=Store(db).total({});db.close()
  65. assert stats['included']==args.rows
  66. assert totals['amount']==Decimal('0.1')*args.rows
  67. report=dict(comparison=records, equal_to_legacy=sum(r['legacy_equal'] for r in records),
  68. cases=len(records),before_recognized_pairs=sum(bool(r['before_pair'] and all(r['before_pair'])) for r in records),
  69. current_recognized_pairs=sum(bool(r['current']['brand'] and r['current']['model']) for r in records),
  70. synthetic_benchmark=dict(rows=args.rows,distinct_declarations=len(texts),seconds=round(elapsed,3),
  71. rows_per_second=round(args.rows/elapsed),amount=str(totals['amount']),included=stats['included'],
  72. scope='stream reading, parsing, audited SQLite ingestion, monthly aggregation and indexes; excludes full reports and uploads'))
  73. (out/'report.json').write_text(json.dumps(report,ensure_ascii=False,indent=2))
  74. print(json.dumps({k:v for k,v in report.items() if k!='comparison'},ensure_ascii=False,indent=2))
  75. assert report['equal_to_legacy']==len(records)
  76. if __name__=='__main__':main()