关于「相关证据的贝叶斯合并:对数几率降权」,《十二体系的同源证据降权》给出的定量结论是什么?

结论:第四章推导相关证据的贝叶斯合并:相关系数 0.5 时,十二路证据的对数似然比从 1.5000 降为 0.2308;共享潜在因子时,朴素后验升到 0.9796,精确后验停在 0.6000。

数据口径:本书只分析符号体系的结构、信息量与概率,不做任何个人占断,不宣称任何体系有预测效力;书中参数为示意,全部计算由书中程序生成。 第六章用数据处理不等式说明,示意模型中十二套体系关于结果的信息恰为出生日本身的 0.5769 毫比特,一点不多;事后对上几乎必然,每句陈述只有 0.05 的机会对上时,十二套体系至少一句对上的概率是 0.9539。

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: 0.2308, 0.5, 1.5000

12 systems, each at its typical H1 value, mu0.5
rho naive LLR exact LLR closed form P(H1) naive exact
0.01.5000 1.5000 1.5000 0.818 0.818
0.21.5000 0.4688 0.4688 0.818 0.615
0.51.5000 0.2308 0.2308 0.818 0.557
0.81.5000 0.1531 0.1531 0.818 0.538
prior 0.1, rho 0.5: naive 0.332, exact 0.123
weight per system at rho 0.51 / (1 + 11 * 0.5) = 0.1538
mixed case rho 0.5: six at +mu, six at -mu, naive -1.500, exact -0.231
查看完整验算程序
# Chapter 4. p41: log-odds of m correlated Gaussian signals, naive vs exact.
# Each system emits z_i; under H1 the mean is mu, under H0 it is 0; the
# covariance is (1 - rho) I + rho J under both. The exact log likelihood
# ratio is mu * 1'S^-1 z - mu^2 * 1'S^-1 1 / 2; with equal correlation
# 1'S^-1 = 1' / (1 + (m - 1) rho), so naive summing overstates the log-odds
# by the design effect. The closed form is checked by Gaussian elimination.
# mu = 0.5 and the priors are illustrative.
import math


def solve(a, b):
    n = len(b)
    m = [row[:] + [b[i]] for i, row in enumerate(a)]
    for c in range(n):
        piv = max(range(c, n), key=lambda r: abs(m[r][c]))
        m[c], m[piv] = m[piv], m[c]
        for r in range(n):
            if r != c:
                f = m[r][c] / m[c][c]
                m[r] = [x - f * y for x, y in zip(m[r], m[c])]
    return [m[i][n] / m[i][i] for i in range(n)]


def llr(z, mu, rho, exact=True):
    m = len(z)
    if not exact:
        return math.fsum(mu * x - mu * mu / 2 for x in z)
    s = [[1.0 if i == j else rho for j in range(m)] for i in range(m)]
    w = solve(s, [1.0] * m)                       # S^-1 1
    return (mu * math.fsum(a * b for a, b in zip(w, z))
            - mu * mu * math.fsum(w) / 2)


def post(prior, l):
    o = prior / (1 - prior) * math.exp(l)
    return o / (1 + o)


mu, m = 0.5, 12
z = [mu] * m                     # 十二个体系都给出 H1 下的典型值
print("12 systems, each at its typical H1 value, mu = %.1f" % mu)
print(" rho  naive LLR  exact LLR  closed form  P(H1) naive  exact")
for rho in (0.0, 0.2, 0.5, 0.8):
    ln, le = llr(z, mu, rho, False), llr(z, mu, rho)
    cf = ln / (1 + (m - 1) * rho)
    print("%4.1f %10.4f %10.4f %12.4f %12.3f %6.3f"
          % (rho, ln, le, cf, post(0.5, ln), post(0.5, le)))
print("prior 0.1, rho 0.5: naive %.3f, exact %.3f"
      % (post(0.1, llr(z, mu, 0.5, False)), post(0.1, llr(z, mu, 0.5))))
print("weight per system at rho 0.5: 1 / (1 + 11 * 0.5) = %.4f"
      % (1 / 6.5))
print("mixed case rho 0.5: six at +mu, six at -mu, naive %.3f, exact %.3f"
      % (llr([mu] * 6 + [-mu] * 6, mu, 0.5, False),
         llr([mu] * 6 + [-mu] * 6, mu, 0.5)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《十二命理体系独立证据降权、同源生辰伪共振剥离与跨东西方符号系统正交合参潜变量母典》第 4 章「相关证据的贝叶斯合并:对数几率降权」(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-283 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026283,
  author    = {Fan, Yafu},
  title     = {Down-Weighting Non-Independent Evidence across Twelve Symbolic Systems: Spurious Resonance from a Shared Birth Input and Orthogonal Latent-Variable Synthesis},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论