关于「四种合成规则:加性、乘性、短板与软最小值的拟合比较」,《三秒薄切片信任》给出的定量结论是什么?
结论:第四章用幂平均把加性、乘性与短板放在一条刻度上,补短板相对补强项的收益在几何平均下是 2.8 倍,在 p 为 -4 时是 304.8 倍。
数据口径:书中参数为示意;全部计算由书中程序生成;本书只写如何让真实、可兑现的信号在前三秒清晰一致,不写任何伪造资质或假权威的手法,不宣称任何气味的健康功效,是非个人化经营建议。 第一章区分信度与效度,示意参数下三秒判断只达到效度上限的 0.612,与长期评价的一致率为 0.579。
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: 2.8, 304.8
| A. p balanced lopsided gap | |
| +1 | 0.700 0.700 0.000 |
| geo | 0.700 0.624 0.076 |
| -1 | 0.700 0.540 0.160 |
| -2 | 0.700 0.470 0.230 |
| -4 | 0.700 0.392 0.308 |
| -8 | 0.700 0.344 0.356 |
| min | 0.700 0.300 0.400 |
| B. lopsided (0.9, 0.9, 0.3): gain from +0.1 on one channel | |
| p raise V raise A raise O(weak) O/V ratio | |
| +1 | 0.0333 0.0333 0.0333 1.0 |
| geo | 0.0223 0.0223 0.0628 2.8 |
| -1 | 0.0110 0.0110 0.0953 8.6 |
| -2 | 0.0041 0.0041 0.1166 28.3 |
| -4 | 0.0004 0.0004 0.1242 304.8 |
| -8 | 0.0000 0.0000 0.1146 >999 |
| min | 0.0000 0.0000 0.1000 >999 |
查看完整验算程序
# Chapter 4. p42: the power mean M_p as one dial from sum to barrel.
# M_p(x) = (mean of x_i^p)^(1/p); p = 1 is the plain mean, p -> 0 the
# geometric mean (the cube root of the product), p -> -inf the minimum.
# Two profiles with the same plain mean 0.7: balanced (0.7, 0.7, 0.7) and
# lopsided (0.9, 0.9, 0.3). Part B asks which channel to raise by 0.1 in the
# lopsided profile. Scores are illustrative.
import math
BAL = (0.7, 0.7, 0.7)
LOP = (0.9, 0.9, 0.3)
PS = (1.0, 0.0, -1.0, -2.0, -4.0, -8.0, None)
def m_p(x, p):
if p is None:
return min(x)
if p == 0.0:
return math.exp(math.fsum(math.log(v) for v in x) / len(x))
return (math.fsum(v ** p for v in x) / len(x)) ** (1.0 / p)
def label(p):
if p is None:
return "min"
return "geo" if p == 0.0 else "{:+.0f}".format(p)
if __name__ == "__main__":
print("A. p balanced lopsided gap")
for p in PS:
b, l = m_p(BAL, p), m_p(LOP, p)
print(" {:>4} {:.3f} {:.3f} {:.3f}".format(
label(p), b, l, abs(b - l)))
print("B. lopsided (0.9, 0.9, 0.3): gain from +0.1 on one channel")
print(" p raise V raise A raise O(weak) O/V ratio")
for p in PS:
base = m_p(LOP, p)
g = []
for i in range(3):
x = list(LOP)
x[i] = min(1.0, x[i] + 0.1)
g.append(m_p(x, p) - base)
if g[0] < g[2] / 999.0:
ratio = " >999"
else:
ratio = "{:6.1f}".format(g[2] / g[0])
print(" {:>4} {:.4f} {:.4f} {:.4f} {}".format(
label(p), g[0], g[1], g[2], ratio))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《三秒薄切片信任捕获:视听嗅三通道同频一致性与瞬间权威建立母典》第 4 章「四种合成规则:加性、乘性、短板与软最小值的拟合比较」(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-419 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026419,
author = {Fan, Yafu},
title = {Three-Second Thin-Slicing Trust: Visual-Auditory-Olfactory Congruence and Instant Authority Signalling},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}