关于「平均买家的盲区:拉丁超立方合成客群与组合覆盖率」,《红蓝军合成客群沙盘》给出的定量结论是什么?
结论:程序 p11 显示,平均买家在十二种示意破绽上一无所获,24 人的合成客群平均找到约 7 种。程序 p12 显示,拉丁超立方在一维上做到全覆盖,在两维上小胜,在三维上与简单随机打平。程序 p13 说明它真正的长处在于估计平滑的平均量,方差比可达 336.86,而对角落破绽毫无帮助。那么,要系统地覆盖两维组合,最少需要多少位买家?合成出来的买家,又会不会比真实买家更整齐,把尾部本身磨平?
数据口径:本书检验相反的立项命题:只用平均买家会漏掉尾部破绽,拉丁超立方合成客群在同样测试量下覆盖更多人格组合(McKay、Beckman 与 Conover 1979 年刊于《技术计量学》),多轮砍价有 Rubinstein 1982 年给出的贴现均衡。 书中参数为示意;全部计算由书中程序生成,不调用任何模型接口;本书是非个人化经营建议。
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: 24, 336.86
| output method mean sd | |
| A SRS | 0.47669 0.02459 |
| A LHS | 0.47648 0.00134 |
| B SRS | 0.00771 0.01765 |
| B LHS | 0.00808 0.01774 |
| variance ratio SRS/LHS, smooth output | 336.86 |
| variance ratio SRS/LHS, corner output | 0.99 |
| crowds that never meet the corner: SRS 0.829, LHS 0.818 | |
| exact SRS miss chance (1 - 0.008)^24 | 0.825 |
查看完整验算程序
# Chapter 1. p13: where LHS pays off and where it does not.
# McKay, Beckman and Conover (1979, Technometrics) proved that LHS
# lowers the variance of the sample mean when the output is monotone
# in each input. Three buyer traits on [0, 1] (budget room, patience,
# trust; illustrative). Output A: a smooth, monotone chance of buying,
# logistic in the traits. Output B: a corner leak that fires only when
# all three traits exceed 0.8 (true rate 0.008). Crowd of 24, 6000
# seeded replications; variance ratio SRS / LHS above 1 favours LHS.
import math
import random
import statistics
N, REPS, K = 24, 6000, 3
def conv(x):
z = -1.6 + 1.2 * x[0] + 0.8 * x[1] + 1.0 * x[2]
return 1 / (1 + math.exp(-z))
def corner(x):
return 1.0 if min(x) > 0.8 else 0.0
def srs(rng):
return [[rng.random() for _ in range(K)] for _ in range(N)]
def lhs(rng):
cols = []
for _ in range(K):
perm = list(range(N))
rng.shuffle(perm)
cols.append([(perm[i] + rng.random()) / N for i in range(N)])
return [[c[i] for c in cols] for i in range(N)]
rng = random.Random(20261113)
est = {(f, m): [] for f in ("A", "B") for m in ("SRS", "LHS")}
miss = {"SRS": 0, "LHS": 0}
for _ in range(REPS):
for m, gen in (("SRS", srs), ("LHS", lhs)):
crowd = gen(rng)
est[("A", m)].append(math.fsum(map(conv, crowd)) / N)
b = math.fsum(map(corner, crowd)) / N
est[("B", m)].append(b)
miss[m] += b == 0
print("output method mean sd")
for f in ("A", "B"):
for m in ("SRS", "LHS"):
v = est[(f, m)]
print("%-6s %-6s %.5f %.5f"
% (f, m, statistics.fmean(v), statistics.pstdev(v)))
for f, label in (("A", "smooth"), ("B", "corner")):
r = (statistics.pvariance(est[(f, "SRS")])
/ statistics.pvariance(est[(f, "LHS")]))
print("variance ratio SRS/LHS, %s output: %.2f" % (label, r))
print("crowds that never meet the corner: SRS %.3f, LHS %.3f"
% (miss["SRS"] / REPS, miss["LHS"] / REPS))
print("exact SRS miss chance (1 - 0.008)^24: %.3f" % (0.992 ** 24))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《红蓝军合成客群对抗沙盘、参数化挑剔买家多轮砍价博弈与漏单破绽预检母典》第 1 章「平均买家的盲区:拉丁超立方合成客群与组合覆盖率」(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-259 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026259,
author = {Fan, Yafu},
title = {Red-Blue Synthetic-Customer Sandboxes: Multi-Round Bargaining against Parameterised Demanding Buyers and Pre-Launch Leak Detection},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}