关于「原典考辨:轮扁斫轮与亚里士多德的公道」,《条件参数化规则蒸馏与在线校准》给出的定量结论是什么?

结论:第七章考辨《庄子》轮扁斫轮、《孟子》徒法不能以自行与亚里士多德的公道论,说不出的残差份额在示意模型中为 0.291。

数据口径:书中参数为示意,全部计算由书中程序生成。 本书是非个人化经营分析,不构成投资、法律或医疗意见,不对应任何真实客户或成交。

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: 0.291

total variance of the right speed: 0.86
factors written downlossshare of loss left
none (one speed)0.861.000
hardness0.500.581
hardness+moisture0.340.395
hardness+moisture+grain0.250.291
tacit share that no written rule removes: 0.291
apprentice learning the full written rule from n wheels:
n =5: extra loss 1.2447n x extra 6.22
n =10: extra loss 0.3421n x extra 3.42
n =40: extra loss 0.0601n x extra 2.41
n = 160: extra loss 0.0144n x extra 2.31
查看完整验算程序
# Chapter 7. p71: Wheelwright Bian as a variance budget (illustrative).
# The right chisel speed v* for a wheel = hardness + moisture + grain +
# a tacit part felt in the hand. Components are independent with
# variances 0.36, 0.16, 0.09 and 0.25 (arbitrary units). Loss of a
# wheel = (v - v*)^2. A written rule can condition on the recorded
# factors only; the tacit part stays as residual loss however good the
# rule is. An apprentice who estimates the written rule from n wheels
# (one slope per factor and an intercept, by least squares) pays an
# extra loss on top; Monte Carlo shows it shrinks roughly as 1 / n.
import math
import random

VAR = (("hardness", 0.36), ("moisture", 0.16), ("grain", 0.09))
TACIT = 0.25


def main():
    total = math.fsum([v for _, v in VAR] + [TACIT])
    print("total variance of the right speed: {:.2f}".format(total))
    print(" factors written down          loss   share of loss left")
    for k in range(len(VAR) + 1):
        left = math.fsum([v for _, v in VAR[k:]] + [TACIT])
        names = "+".join(n for n, _ in VAR[:k]) or "none (one speed)"
        print(" {:<28} {:.2f}     {:.3f}".format(names, left, left / total))
    print("tacit share that no written rule removes: {:.3f}".format(
        TACIT / total))
    print("apprentice learning the full written rule from n wheels:")
    rng = random.Random(7101)
    for n in (5, 10, 40, 160):
        extra = []
        for _ in range(2000):
            xs = [[rng.gauss(0, math.sqrt(v)) for _, v in VAR]
                  for _ in range(n)]
            ys = [math.fsum(x) + rng.gauss(0, math.sqrt(TACIT))
                  for x in xs]
            # independent factors: per-factor slope, centred
            mx = [math.fsum(x[j] for x in xs) / n for j in range(3)]
            my = math.fsum(ys) / n
            coef = []
            for j in range(3):
                sxy = math.fsum((x[j] - mx[j]) * (y - my)
                                for x, y in zip(xs, ys))
                sxx = math.fsum((x[j] - mx[j]) ** 2 for x in xs)
                coef.append(sxy / sxx)
            err = [(c - 1.0) ** 2 * v for c, (_, v) in zip(coef, VAR)]
            icpt = my - math.fsum(c * m for c, m in zip(coef, mx))
            extra.append(math.fsum(err) + icpt ** 2)
        mean = math.fsum(extra) / 2000
        print("  n = {:>3}: extra loss {:.4f}   n x extra {:.2f}".format(
            n, mean, n * mean))


if __name__ == "__main__":
    main()

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《高阶知识蒸馏进化、条件参数化编译与实盘转化反馈校准母典》第 7 章「原典考辨:轮扁斫轮与亚里士多德的公道」(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-377 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026377,
  author    = {Fan, Yafu},
  title     = {Next-Generation Knowledge Distillation, Context-Parametric Compilation and Live-Conversion Feedback Calibration},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论