关于「似然比串联:比重、折射率与荧光的贝叶斯鉴别」,《翡翠鉴别与赌石赢家诅咒》给出的定量结论是什么?

结论:第三章把比重、折射率与荧光串成似然比:同一组读数在游客市场给出 0.8622 的翡翠后验,在信誉商家给出 0.9657。

数据口径:书中参数为示意;全部计算由书中程序生成;本书不是个人化的法律或理财建议,也不是任何一件石头的鉴定意见。 第六章把三项修正折成出价上限:挑窗、六人竞价、证书可靠时为窗口估值的 0.309,只做门店检验时为 0.274;在天真买家占多数的公盘里应当离场;凯利比例在示意优势下为 0.090。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.8622, 0.9657

sample A: 3.33/1.66/none
priorP(jade)=0.2000top=B-jade0.3000
+ SGP(jade)=0.5209top=jadeite0.5209
+ SG, RIP(jade)=0.5684top=jadeite0.5684
+ SG, RI, UVP(jade)=0.8622top=jadeite0.8622
sample B: 3.30/1.66/glow
priorP(jade)=0.2000top=B-jade0.3000
+ SGP(jade)=0.2756top=B-jade0.7222
+ SG, RIP(jade)=0.3155top=B-jade0.6845
+ SG, RI, UVP(jade)=0.0280top=B-jade0.9720
sample C: 2.96/1.61/none
priorP(jade)=0.2000top=B-jade0.3000
+ SGP(jade)=0.0000top=nephrite0.9496
+ SG, RIP(jade)=0.0000top=nephrite0.9960
+ SG, RI, UVP(jade)=0.0000top=nephrite0.9970
sample D: 2.66/1.54/glow
priorP(jade)=0.2000top=B-jade0.3000
+ SGP(jade)=0.0000top=quartzite0.9053
+ SG, RIP(jade)=0.0000top=quartzite0.9864
+ SG, RI, UVP(jade)=0.0000top=quartzite0.9891
sample A with a dealer prior (0.60 jadeite): P(jade)=0.9657
LR of 'no glow', jadeite vs B-jade: 4.75

核心公式

\[\frac{P(A\mid E)}{P(B\mid E)} = \frac{P(A)}{P(B)} \times \frac{P(E\mid A)}{P(E\mid B)}\]
查看完整验算程序
# Chapter 3. p31: Chaining likelihood ratios: SG, spot RI, long-wave UV.
# 类参数(示意,量级取宝石学通行值):比重同 p22;点测折射率:翡翠
# 1.66(Hughes 等 2000)、B 货 1.66、软玉 1.61、石英岩 1.54、蛇纹石
# 1.56、玻璃 1.55(宽);长波紫外荧光出现概率:翡翠 0.05、B 货 0.80、
# 软玉 0.05、染色石英岩 0.30、蛇纹石 0.20、玻璃 0.30(均为示意)。
# 读数误差:比重 0.0079(p21,10 g、0.01 g 天平),折射率 0.005。
# 先验:一个游客市场的假想构成(示意);另给一个信誉商家的先验作对照。
# 各检验在类内条件独立时,
# 后验 = 先验乘以各项似然再归一;表中给出四件样品的逐步后验。
# 第八章流程卡以翡翠后验 0.95 为门店放行阈值,0.05 为排除阈值。
import math
from statistics import NormalDist

CLS = ("jadeite", "B-jade", "nephrite", "quartzite", "serpentine",
       "glass")
SG = ((3.34, 0.03), (3.30, 0.04), (2.95, 0.03), (2.65, 0.02),
      (2.57, 0.06), (2.60, 0.30))
RI = ((1.66, 0.006), (1.66, 0.008), (1.61, 0.006), (1.54, 0.005),
      (1.56, 0.010), (1.55, 0.050))
UV = (0.05, 0.80, 0.05, 0.30, 0.20, 0.30)
PRIOR = (0.20, 0.30, 0.10, 0.20, 0.10, 0.10)
DEALER = (0.60, 0.20, 0.05, 0.05, 0.05, 0.05)
SM_SG, SM_RI = 0.0079, 0.005


def norm(ws):
    t = math.fsum(ws)
    return [w / t for w in ws]


def update(p, sg=None, ri=None, uv=None):
    out = []
    for i, w in enumerate(p):
        if sg is not None:
            w *= NormalDist(SG[i][0], math.hypot(SG[i][1], SM_SG)).pdf(sg)
        if ri is not None:
            w *= NormalDist(RI[i][0], math.hypot(RI[i][1], SM_RI)).pdf(ri)
        if uv is not None:
            w *= UV[i] if uv else 1 - UV[i]
        out.append(w)
    return norm(out)


def show(tag, p):
    k = max(range(len(p)), key=lambda i: p[i])
    print("  %-14s P(jade)=%.4f  top=%-10s %.4f"
          % (tag, p[0], CLS[k], p[k]))


def main():
    cases = (("A: 3.33/1.66/none", 3.33, 1.66, False),
             ("B: 3.30/1.66/glow", 3.30, 1.66, True),
             ("C: 2.96/1.61/none", 2.96, 1.61, False),
             ("D: 2.66/1.54/glow", 2.66, 1.54, True))
    for name, sg, ri, uv in cases:
        print("sample " + name)
        p = list(PRIOR)
        show("prior", p)
        p = update(p, sg=sg)
        show("+ SG", p)
        p = update(p, ri=ri)
        show("+ SG, RI", p)
        p = update(p, uv=uv)
        show("+ SG, RI, UV", p)
    p = update(list(DEALER), sg=3.33, ri=1.66, uv=False)
    print("sample A with a dealer prior (0.60 jadeite): P(jade)=%.4f"
          % p[0])
    lr = (1 - UV[0]) / (1 - UV[1])
    print("LR of 'no glow', jadeite vs B-jade: %.2f" % lr)


if __name__ == "__main__":
    main()

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《翡翠静水比重贝叶斯鉴别、赌石开窗信号博弈与原石拍卖赢家诅咒母典》第 3 章「似然比串联:比重、折射率与荧光的贝叶斯鉴别」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-206 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026206,
  author    = {Fan, Yafu},
  title     = {Hydrostatic Specific Gravity and Bayesian Jadeite Identification, Window-Cut Signalling Games and the Winner's Curse in Rough-Jade Tenders},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论