关于「实证闭环:新品冷启动需求分布、最大熵需求重建器与预注册证伪方案」,《最大熵需求重建》给出的定量结论是什么?

结论:第八章给出新品冷启动闭环、蒸馏小工具「最大熵需求重建器」与预注册证伪方案,校准检验分辨超出比例 0.20 需 86 个新品。书中以标注框区分已发表的证据、可运行的 Python 计算(20 段程序)、原典与作者的假说。

数据口径:本书推翻这一常识:正态捷径私加了对称形状与标准差,示意参数下均值 80、下限 30 时它把 0.133 的概率放在不可能的区间,均值 40 时升到 0.412;均匀捷径隐含均值在区间中点,直接违背已知均值。 只承认上下限与均值时,杰恩斯 1957 年提出的最大熵原则给出截断指数分布,示意甜品店的第 90 百分位客单为 146.0 元。

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: 0.20, 86

A. items needed to test P90 calibration (null share 0.10)
true share 0.05239 items
true share 0.15316 items
true share 0.2086 items
true share 0.2541 items
B. items needed for a paired regret comparison
effect size 0.2199 items
effect size 0.390 items
effect size 0.534 items
effect size 0.815 items
C. rule of thumb from p51 and p52: below about 30 orders use
max-ent from mean and bounds; from about 50 orders the
sample's own percentile wins on these illustrative truths
查看完整验算程序
# Chapter 8. p82: how many items and orders a fair test needs.
# Model: two pre-registered tests, normal approximations, two-sided
# alpha 0.05 and power 0.80. A. Calibration: if the maximum-entropy
# P90 is honest, a share 0.10 of new items sell above it; how many
# items detect a true share of p1? B. Paired comparison of profit
# regret between max-ent and normal ordering across items, effect size
# d = mean difference / sd of differences. C. Orders needed before the
# sample's own P90 beats max-ent, read off p51 and p52 (illustrative).
import math
from statistics import NormalDist

Z = NormalDist()


def n_prop(p0, p1, alpha=0.05, power=0.80):
    za, zb = Z.inv_cdf(1 - alpha / 2), Z.inv_cdf(power)
    num = za * math.sqrt(p0 * (1 - p0)) + zb * math.sqrt(p1 * (1 - p1))
    return math.ceil((num / (p1 - p0)) ** 2)


def n_paired(d, alpha=0.05, power=0.80):
    za, zb = Z.inv_cdf(1 - alpha / 2), Z.inv_cdf(power)
    return math.ceil(((za + zb) / d) ** 2 + za * za / 2)


if __name__ == "__main__":
    print("A. items needed to test P90 calibration (null share 0.10)")
    for p1 in (0.05, 0.15, 0.20, 0.25):
        print("  true share {:.2f}: {:>5} items".format(p1, n_prop(0.10, p1)))
    print("B. items needed for a paired regret comparison")
    for d in (0.2, 0.3, 0.5, 0.8):
        print("  effect size {:.1f}: {:>5} items".format(d, n_paired(d)))
    print("C. rule of thumb from p51 and p52: below about 30 orders use")
    print("   max-ent from mean and bounds; from about 50 orders the")
    print("   sample's own percentile wins on these illustrative truths")

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《最大熵原理与缺数据需求重建、仅知均价与上下限时最不偏的客单分布推断母典》第 8 章「实证闭环:新品冷启动需求分布、最大熵需求重建器与预注册证伪方案」(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-301 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026301,
  author    = {Fan, Yafu},
  title     = {The Maximum-Entropy Principle and Demand Reconstruction from Missing Data: the Least-Biased Order-Value Distribution Given Only a Mean and Bounds},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论