关于「期望信息增益:探针价格该设在学得最多的地方」,《先手探针与隐藏事实》给出的定量结论是什么?

结论:第三章用林德利信息比较探针价格:信息最多的价格是 45,收入最高的价格是 35。

数据口径:书中参数为示意,全部计算由书中程序生成,数据全部是合成数据。 本书是非个人化经营建议,不构成法律、税务或投资意见,不向任何人承诺收入。

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: 35, 45

prior entropy 3.462 bits, prior mean theta 46.02
priceEIG n=40 (bits)revenue per visitor
150.80114.54
201.12818.78
251.46822.22
301.78324.47
352.03525.26
402.19624.55
452.25322.57
502.20719.68
552.07216.34
601.86812.91
651.6199.68
701.3506.85
751.0804.50
800.8262.72
850.6001.49
best probe price 45 with 2.253 bits
n =10: EIG at 45 = 1.419 bits
n =40: EIG at 45 = 2.253 bits
n = 160: EIG at 45 = 3.024 bits
查看完整验算程序
# Chapter 3. p31: expected information gain of a probe price (Lindley).
# Illustrative. The unknown is the median willingness to pay theta of a
# segment, on a grid from 20 to 80 units in steps of 5, with a prior near
# 40 (log-normal shape, spread 0.35). A visitor shown price p buys
# with chance 1 / (1 + exp((p - theta) / 6)). A probe shows one real,
# purchasable price to n visitors and counts buyers k ~ Binomial(n, .).
# EIG(p) = mutual information between theta and k, in bits.
import math

THETAS = list(range(20, 81, 5))
S = 6.0


def prior():
    w = [math.exp(-(math.log(t) - math.log(40.0)) ** 2 / (2 * 0.35 ** 2))
         for t in THETAS]
    tot = math.fsum(w)
    return [x / tot for x in w]


def p_buy(p, t):
    return 1.0 / (1.0 + math.exp((p - t) / S))


def eig(pri, p, n):
    like = []
    for t in THETAS:
        q = p_buy(p, t)
        like.append([math.comb(n, k) * q ** k * (1 - q) ** (n - k)
                     for k in range(n + 1)])
    marg = [math.fsum(pri[i] * like[i][k] for i in range(len(THETAS)))
            for k in range(n + 1)]
    terms = []
    for i in range(len(THETAS)):
        for k in range(n + 1):
            if like[i][k] > 0 and marg[k] > 0:
                terms.append(pri[i] * like[i][k]
                             * math.log2(like[i][k] / marg[k]))
    return math.fsum(terms)


def entropy(pr):
    return -math.fsum(x * math.log2(x) for x in pr if x > 0)


if __name__ == "__main__":
    pri = prior()
    mean = math.fsum(t * w for t, w in zip(THETAS, pri))
    print("prior entropy {:.3f} bits, prior mean theta {:.2f}".format(
        entropy(pri), mean))
    print(" price  EIG n=40 (bits)  revenue per visitor")
    best = None
    for p in range(15, 86, 5):
        g = eig(pri, p, 40)
        rev = p * math.fsum(w * p_buy(p, t) for t, w in zip(THETAS, pri))
        print("  {:>3}       {:.3f}             {:>6.2f}".format(p, g, rev))
        if best is None or g > best[1]:
            best = (p, g)
    print("best probe price {} with {:.3f} bits".format(*best))
    for n in (10, 40, 160):
        print("  n = {:>3}: EIG at {} = {:.3f} bits".format(
            n, best[0], eig(pri, best[0], n)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《高能动性主动推理、意念强度先手扰动与商业隐藏事实勘探母典》第 3 章「期望信息增益:探针价格该设在学得最多的地方」(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-395 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026395,
  author    = {Fan, Yafu},
  title     = {High-Agency Active Inference, Intentional Perturbation Probes and Hidden Market-Truth Exploration},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论