# -*- coding: utf-8 -*-
"""
用【同缸径的真实数据值域】作为官方值域的代理, 评估生成候选的保守程度。

为什么用代理: 官方图册是双栏排版, pdftotext 后左右栏交错, 正则解析多次尝试均不可靠
(把真实数据都判成越界 = 解析错误)。而【真实型号】是最可靠的一手证据:
某缸径的真实型号里从未出现过的取值, 生成它就是推测。

判定口径 (保守):
  某候选的每个参数值, 是否都在【同缸径的真实型号】中出现过?
  - 全部出现过  → 属"组合缺失", 补全风险低
  - 有值从未出现 → 属"引入新值", 补全风险高 (需人工/AI 核对官方图册)
"""
import csv, io, os, collections

CAT = r'D:/开发/OnebotCatalog/onebot-data/catalog'
DIMS = ['stroke', 'rod', 'orient', 'sensor', 'mount', 'style', 'pos', 'type']


def main():
    rows = list(csv.DictReader(io.open(os.path.join(CAT, 'csv', 'ULP.csv'), encoding='utf-8-sig')))
    real = [r for r in rows if r['step_file'] != 'PLACEHOLDER.step']
    gen = [r for r in rows if r['step_file'] == 'PLACEHOLDER.step']
    dims = [d for d in DIMS if d in rows[0]]

    # 每个缸径的真实取值集合
    domain = collections.defaultdict(lambda: collections.defaultdict(set))
    for r in real:
        for d in dims:
            if r[d] != '':
                domain[r['size']][d].add(r[d])

    print('=' * 74)
    print('真实数据里各缸径的取值域 (作为官方值域的代理)')
    print('=' * 74)
    for b in sorted(domain, key=int):
        parts = [f'{d}={sorted(domain[b][d])}' for d in dims if domain[b][d]]
        n = sum(1 for r in real if r['size'] == b)
        print(f'  ULP{b:>3s} ({n:3d} 个真实型号)')
        for p in parts:
            print(f'        {p}')

    # 评估生成候选
    safe, risky = 0, 0
    risky_detail = collections.Counter()
    risky_models = []
    for r in gen:
        b = r['size']
        novel = [f'{d}={r[d]}' for d in dims
                 if r[d] != '' and r[d] not in domain.get(b, {}).get(d, set())]
        if novel:
            risky += 1
            for x in novel:
                risky_detail[f'ULP{b} {x}'] += 1
            if len(risky_models) < 5:
                risky_models.append((r['model_code'], novel))
        else:
            safe += 1

    print()
    print('=' * 74)
    print(f'生成候选 {len(gen)} 个的保守度评估')
    print('=' * 74)
    print(f'  低风险 (全部取值在同缸径真实数据中出现过): {safe}  ({safe*100//len(gen)}%)')
    print(f'  高风险 (含该缸径从未出现的取值)          : {risky}  ({risky*100//len(gen)}%)')
    if risky_detail:
        print('\n  高风险取值 TOP15 (该缸径真实数据里从未出现):')
        for k, v in risky_detail.most_common(15):
            print(f'    {k:28s} {v:5d} 个候选')
        print('\n  高风险样本:')
        for mc, novel in risky_models:
            print(f'    {mc:18s} 新值 {novel}')


if __name__ == '__main__':
    main()