关于「单因子潜变量与正交合参」,《十二体系的同源证据降权》给出的定量结论是什么?

结论:第五章发现单因子模型不成立,转而按共享周期把十二套体系分成 5 块做正交合参。

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

single-factor loadings (informational correlation scale)
  TROP 0.08 SIDR 0.08 MONB 0.08 ZODI 0.03 NSTA 0.03 NAYI 0.02
  DAYP 0.93 MANS 1.07 WEEK 0.37 TZOL 0.90 LIFE 0.15 HOUR 0.00
variance of the twelve carried by the factor: 0.251
largest eigenvalues of r: 3.195 2.957 2.147 1.328
eigenvalues above 1: 4 -> one factor is not enough
loadings above 1 (Heywood cases): MANS 1.07
查看完整验算程序
# Chapter 5. p51: one latent factor fitted to the twelve systems.
# Pairwise mutual information I (bits) as in p12 is turned into Linfoot's
# informational correlation r = sqrt(1 - 2^(-2 I)), the correlation a pair
# of Gaussian variables would need to share I bits. A single factor is fit
# by iterated principal axes (communalities start at the largest |r| in a
# row); the eigenvalues of r show how many factors the structure needs.
import datetime
import math
from collections import Counter

D0, D1 = datetime.date(1950, 1, 1), datetime.date(2009, 12, 31)
JIAZI = datetime.date(1949, 10, 1).toordinal()  # 甲子日(据通行历表)
GMT, JD = 584283, 1721425       # 卓尔金历 GMT 对应常数;序数日转儒略日
CUM = [0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334]
# 回归黄道太阳星座起始日(白羊为 0,固定日期近似)
TROP = [(3, 21), (4, 20), (5, 21), (6, 21), (7, 23), (8, 23),
        (9, 23), (10, 23), (11, 22), (12, 22), (1, 20), (2, 19)]
# 十二节近似日期,寅月(支序 2)起,子为 0
JIE = [(2, 4), (3, 6), (4, 5), (5, 6), (6, 6), (7, 7), (8, 8), (9, 8),
       (10, 8), (11, 7), (12, 7), (1, 6)]
# 六十甲子纳音五行,按每两年一组:M 金 F 火 W 木 E 土 A 水
NAYIN = "MFWEMFAEMWAEFWAMFWEMFAEMWAEFWA"
NAMES = ["TROP", "SIDR", "MONB", "ZODI", "NSTAR", "NAYIN",
         "DAYP", "MANS", "WEEK", "TZOL", "LIFE", "HOUR"]


def doy(m, d):
    return CUM[m - 1] + min(d, 28 if m == 2 else 31) - 1


def cyc(x, table, first):
    best, lab = -1, None
    for k, (m, d) in enumerate(table):
        s = doy(m, d)
        if best < s <= x:
            best, lab = s, (first + k) % 12
    if lab is None:
        s = max(doy(m, d) for m, d in table)
        lab = (first + [doy(m, d) for m, d in table].index(s)) % 12
    return lab


def digits(n):
    return [int(c) for c in str(n)]


def life(d):
    s = int(math.fsum(digits(d.year) + digits(d.month) + digits(d.day)))
    while s > 9 and s not in (11, 22, 33):
        s = int(math.fsum(digits(s)))
    return s


def labels(d):
    x = doy(d.month, d.day)
    n = d.toordinal()
    sid = d - datetime.timedelta(days=24)       # 恒星黄道:岁差约 24 天
    y = d.year if x >= doy(2, 4) else d.year - 1   # 以 2 月 4 日近似立春
    return (cyc(x, TROP, 0), cyc(doy(sid.month, sid.day), TROP, 0),
            cyc(x, JIE, 2), (y - 4) % 12, (10 - y % 9) % 9 + 1,
            NAYIN[((y - 4) % 60) // 2], (n - JIAZI) % 60, (n + 3) % 28,
            d.weekday(), (n + JD - GMT + 19) % 20, life(d))


def days():
    d = D0
    while d <= D1:
        yield d
        d += datetime.timedelta(days=1)  # step


def entropy(counter):
    tot = math.fsum(counter.values())
    return -math.fsum(c / tot * math.log2(c / tot)
                      for c in counter.values())


def mi_matrix(rows):
    cols = [None if NAMES[j] == "HOUR" else [r[j] for r in rows]
            for j in range(12)]
    h = [math.log2(12) if c is None else entropy(Counter(c)) for c in cols]
    mi = [[0.0] * 12 for _ in range(12)]
    for i in range(12):
        for j in range(i + 1, 12):
            if cols[i] is not None and cols[j] is not None:
                hij = entropy(Counter(zip(cols[i], cols[j])))
                mi[i][j] = mi[j][i] = max(0.0, h[i] + h[j] - hij)
    return mi


def linfoot(mi):
    return [[1.0 if i == j else math.sqrt(1 - 2 ** (-2 * mi[i][j]))
             for j in range(12)] for i in range(12)]


def top_eig(a, iters=500):
    v = [1.0] * len(a)
    lam = 0.0
    for _ in range(iters):
        w = [math.fsum(x * y for x, y in zip(row, v)) for row in a]
        lam = math.sqrt(math.fsum(x * x for x in w))
        v = [x / lam for x in w]
    return lam, v


rows = [labels(d) for d in days()]
R = linfoot(mi_matrix(rows))
h2 = [max(abs(R[i][j]) for j in range(12) if j != i) for i in range(12)]
for _ in range(50):                      # 迭代主轴法
    red = [[h2[i] if i == j else R[i][j] for j in range(12)]
           for i in range(12)]
    lam, v = top_eig(red)
    load = [x * math.sqrt(lam) for x in v]
    h2 = [min(0.995, x * x) for x in load]
print("single-factor loadings (informational correlation scale)")
print("  " + " ".join("%s %.2f" % (NAMES[j][:4], load[j])
                      for j in range(6)))
print("  " + " ".join("%s %.2f" % (NAMES[j][:4], load[j])
                      for j in range(6, 12)))
print("variance of the twelve carried by the factor: %.3f"
      % (math.fsum(x * x for x in load) / 12))
lams, a = [], [row[:] for row in R]
for _ in range(4):                       # 逐个取特征值并收缩
    lam, v = top_eig(a)
    lams.append(lam)
    a = [[a[i][j] - lam * v[i] * v[j] for j in range(12)]
         for i in range(12)]
print("largest eigenvalues of r: " + " ".join("%.3f" % x for x in lams))
print("eigenvalues above 1: %d -> one factor is not enough"
      % len([x for x in lams if x > 1]))
print("loadings above 1 (Heywood cases): " + " ".join(
    "%s %.2f" % (NAMES[j], load[j]) for j in range(12) if load[j] > 1))

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 5 章「单因子潜变量与正交合参」(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