关于「评分的两部分:语义跨度与预测力增益」,《结构洞、函子迁移与灵感共振》给出的定量结论是什么?

结论:第四章说明詹森-香农跨度以 1 比特为界、会饱和,从 200 个类比中按样本内拟合挑出的胜者,决定系数从 0.283 缩到新数据上的 0.046。

数据口径:书中的网络、领域、评分与势阱均为示意模型;全部计算由书中程序生成;本书不是投资、医疗或心理建议。 第六章用模拟检验克莱默斯公式(相差不超过 1.32 倍),并发现噪声太少是主要风险:低噪声下只有 0.017 的运行找到更好的状态,而先孵化后专注的节奏(0.730)并不比稳定的中等噪声(0.750)更好。

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: 0.046, 0.283

mean in-sample R^2 of the winner0.283
mean in-sample R^2 of the real one 0.128
share of runs the real one wins0.110
winner's R^2 on fresh data0.046
查看完整验算程序
# Chapter 4. p08: "Predictive gain" after searching many analogies. A
# target field has n = 30 observations. 200 candidate analogies each
# supply a predictor; 199 are unrelated, one is real with a true R^2 of
# 0.10. We pick the candidate with the largest in-sample R^2 and test it
# on 30 fresh observations. 1,000 repetitions, fixed seed 102.
import math
import random

rng = random.Random(102)
N, CANDS, REPS = 30, 200, 1000
B = (0.10 / 0.90) ** 0.5          # slope giving true R^2 = 0.10

def r2(x, y):
    mx, my = math.fsum(x) / len(x), math.fsum(y) / len(y)
    sxy = math.fsum((a - mx) * (b - my) for a, b in zip(x, y))
    sxx = math.fsum((a - mx) ** 2 for a in x)
    syy = math.fsum((b - my) ** 2 for b in y)
    return sxy * sxy / (sxx * syy)

def sample(real):
    x = [rng.gauss(0, 1) for _ in range(N)]
    y = [(B * a if real else 0.0) + rng.gauss(0, 1) for a in x]
    return x, y

if __name__ == "__main__":
    best_in = best_out = real_in = 0.0
    picked_real = 0
    for _ in range(REPS):
        xs = [rng.gauss(0, 1) for _ in range(N)]
        z = [rng.gauss(0, 1) for _ in range(N)]
        y = [B * a + e for a, e in zip(xs, z)]
        scores = [(r2(xs, y), 0)]
        for k in range(1, CANDS):
            scores.append((r2([rng.gauss(0, 1) for _ in range(N)], y), k))
        s, k = max(scores)
        best_in += s / REPS
        real_in += scores[0][0] / REPS
        picked_real += (k == 0)
        xo, yo = sample(k == 0)
        best_out += r2(xo, yo) / REPS
    print("   mean in-sample R^2 of the winner   {:.3f}".format(best_in))
    print("   mean in-sample R^2 of the real one {:.3f}".format(real_in))
    print("   share of runs the real one wins    {:.3f}".format(
        picked_real / REPS))
    print("   winner's R^2 on fresh data         {:.3f}".format(best_out))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《跨学科结构洞桥接、函子知识迁移与灵感随机共振》第 4 章「评分的两部分:语义跨度与预测力增益」(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-102 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026102,
  author    = {Fan, Yafu},
  title     = {Cross-Disciplinary Structural Holes, Functorial Knowledge Transfer and Inspiration Stochastic Resonance},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论