关于「加测与停测:便宜的证据在边界上最值钱」,《奢品多特征似然比初筛》给出的定量结论是什么?

结论:第五章说明便宜的加测在边界上最值钱(一项独立加测在秒收边界上值 444.4 元,在中间带正中只值 18.2 元),并说明序贯测量应在更宽的范围外才提前停下,净收益最高为 2519.4 元。

数据口径:书中全部概率只是外观概率参考,不是官方鉴定证书,不构成真伪结论,涉及争议或高价值货品须送品牌或专业鉴定机构;品牌一律以通称指代;书中参数为示意,全部计算由书中程序生成;本书是非个人化经营建议。 第二章是全书核心:示意参数下,朴素后验判到 0.99 以上、平均 0.996 的货,实际不符只有 0.756;放大倍数随类内相关升高,在相关系数 0.6 时达到 2.28;但排序并不吃亏,伤害集中在远离先验的边界上。

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: 18.2, 444.4

current q:0.02000.05390.10000.30000.60000.83620.9500
r=0.0 D2=1.00052.6444.4217.018.225.6148.89.3
r=0.1 D2=0.60223.0350.5123.42.69.3117.43.1
r=0.2 D2=0.3135.1255.747.80.01.485.60.4
r=0.3 D2=0.1130.1155.24.60.00.051.90.0
r=0.4 D2=0.0080.041.10.00.00.013.70.0
cost of E = 20; take E only where its value exceeds the cost
查看完整验算程序
# Chapter 5. p51: is one more cheap measurement worth taking before the
# referral decision? A fifth feature E (illustrative separation 1.0, cost
# 20 yuan of staff time) correlates r with each of the four features
# already scored (matrix R of p31). Given those four, E adds only its
# residual: delta = d_E - r' R^-1 d, residual variance v = 1 - r' R^-1 r,
# incremental separation delta^2 / v. Pre-posterior value at the current
# q: expected best profit after seeing E minus best profit now, with the
# three actions and loss matrix of p41 (numerical integration, fsum).
import math

D = [1.2, 1.0, 1.5, 0.9]
R = [[1.0, 0.3, 0.2, 0.5],
     [0.3, 1.0, 0.6, 0.2],
     [0.2, 0.6, 1.0, 0.2],
     [0.5, 0.2, 0.2, 1.0]]
DE, COST = 1.0, 20.0
M, BIG, FEE, WALK, A, B = 3600.0, 22000.0, 300.0, 0.25, 0.99, 0.01


def solve(a, b):
    n = len(b)
    m = [row[:] + [b[i]] for i, row in enumerate(a)]
    for c in range(n):
        p = max(range(c, n), key=lambda r: abs(m[r][c]))
        m[c], m[p] = m[p], m[c]
        for r in range(c + 1, n):
            f = m[r][c] / m[c][c]
            m[r] = [x - f * y for x, y in zip(m[r], m[c])]
    x = [0.0] * n
    for r in range(n - 1, -1, -1):
        x[r] = (m[r][n] - math.fsum(m[r][k] * x[k]
                                    for k in range(r + 1, n))) / m[r][r]
    return x


def dot(x, y):
    return math.fsum(a * b for a, b in zip(x, y))


def best(q):
    buy = M - q * (M + BIG)
    ref = -FEE + (1.0 - WALK) * (A * (1.0 - q) * M - B * q * BIG)
    return max(buy, ref, 0.0)


def voi(q, d2):
    # 增量对数似然比 l ~ N(+d2/2, d2) 于 N,N(-d2/2, d2) 于 G
    sd = math.sqrt(d2)
    step = 0.01
    terms = []
    for k in range(-1200, 1201):
        u = k * step
        phi = math.exp(-u * u / 2.0) / math.sqrt(2.0 * math.pi) * step
        for cls, pc in ((1, q), (0, 1.0 - q)):
            ell = (d2 / 2.0 if cls else -d2 / 2.0) + sd * u
            lo = math.log(q / (1.0 - q)) + ell
            qq = 1.0 / (1.0 + math.exp(-lo)) if lo > -700 else 0.0
            terms.append(pc * phi * best(qq))
    return math.fsum(terms) - best(q)


QS = [0.02, 0.0539, 0.10, 0.30, 0.60, 0.8362, 0.95]
print("current q:     " + " ".join("%7.4f" % q for q in QS))
for r in [0.0, 0.1, 0.2, 0.3, 0.4]:
    rv = [r] * 4
    delta = DE - dot(rv, solve(R, D))
    v = 1.0 - dot(rv, solve(R, rv))
    d2 = delta * delta / v
    vals = [voi(q, d2) for q in QS]
    print("r=%.1f D2=%.3f " % (r, d2) + " ".join("%7.1f" % x for x in vals))
print("cost of E = %.0f; take E only where its value exceeds the cost"
      % COST)

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《奢品真伪多特征似然比鉴定、走线节距五金光谱皮质纹理融合与送检秒收贝叶斯边界母典》第 5 章「加测与停测:便宜的证据在边界上最值钱」(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-281 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026281,
  author    = {Fan, Yafu},
  title     = {Multi-Feature Likelihood-Ratio Screening of Pre-Owned Luxury Goods: Fusing Stitch Pitch, Hardware Spectra and Leather Grain, and the Bayesian Boundary between Lab Referral and Instant Purchase},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论