关于「身高—围度两维主成分网格:一根轴的标签丢掉了什么」,《尺码体系与退货最小化》给出的定量结论是什么?

结论:第三章说明身高与胸围的第一主成分只承载 0.654 的方差,一维尺码用十六档也只覆盖 0.6022,两维网格用 11 档即覆盖 0.9。

数据口径:书中的品牌、门店、价格与身体尺寸参数均为示意;全部计算由书中程序生成;本书是非个人化经营建议,只涉及服装尺寸的测量,不作任何健康判断。 第二章把一次选码的误差拆成五块:身体尺码表的退货概率是 0.2170,细尺码表 0.1760,完美尺码表也只有 0.1636,而品牌残差法的闭式解是 0.0302,如实模拟留存单品的筛选后是 0.0499。

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: 0.6022, 0.9, 11

sizescover 1Dcover 2Dgain
30.45990.45990.0000
40.49240.54410.0517
60.54280.68350.1407
80.56840.78980.2214
100.58070.87280.2921
120.59040.91920.3288
160.60220.96370.3616
first 6 2D sizes: (170,89) (172,96) (163,96) (181,96) (174,103) (165,103)
sizes to cover 0.8: 1D none of 16, 2D 9
sizes to cover 0.9: 1D none of 16, 2D 11
查看完整验算程序
# Chapter 3. p32: How many sizes cover how much of the population?
# Bivariate normal height-chest population (means 172 and 96 cm, sd 6.5
# and 7.0, rho 0.3; illustrative), binned on a 1 cm lattice. A size at
# (h, c) fits a body when |dh| <= 4 and |dc| <= 3 cm. Sizes are chosen
# greedily (each new size covers the most still-uncovered mass), either
# (1D) only on the regression line of height on chest, as in a chest
# label with graded length, or (2D) anywhere: a nearest-neighbour grid.
import math

MH, MC, SH, SC, RHO = 172.0, 96.0, 6.5, 7.0, 0.3
TH, TC = 4, 3
HS = list(range(140, 205))
CS = list(range(66, 127))


def dens(h, c):
    x, y = (h - MH) / SH, (c - MC) / SC
    q = (x * x - 2 * RHO * x * y + y * y) / (1 - RHO * RHO)
    return math.exp(-q / 2) / (2 * math.pi * SH * SC
                               * math.sqrt(1 - RHO * RHO))


mass = {(h, c): dens(h, c) for h in HS for c in CS}
total = math.fsum(mass.values())
for key in mass:
    mass[key] /= total


def greedy(cands, k):
    covered, picks, out = set(), [], []
    for _ in range(k):
        best, gain = None, -1.0
        for (h0, c0) in cands:
            g = math.fsum(mass[(h, c)]
                          for h in range(h0 - TH, h0 + TH + 1)
                          for c in range(c0 - TC, c0 + TC + 1)
                          if (h, c) in mass and (h, c) not in covered)
            if g > gain + 1e-15:
                best, gain = (h0, c0), g
        picks.append(best)
        for h in range(best[0] - TH, best[0] + TH + 1):
            for c in range(best[1] - TC, best[1] + TC + 1):
                covered.add((h, c))
        out.append(math.fsum(mass[x] for x in covered if x in mass))
    return picks, out


line = [(int(round(MH + RHO * SH / SC * (c - MC))), c)
        for c in range(76, 117)]
grid = [(h, c) for h in range(150, 195) for c in range(76, 117)]
p1, cov1 = greedy(line, 16)
p2, cov2 = greedy(grid, 16)
print("sizes  cover 1D  cover 2D  gain")
for k in (3, 4, 6, 8, 10, 12, 16):
    print("%5d  %.4f    %.4f    %.4f"
          % (k, cov1[k - 1], cov2[k - 1], cov2[k - 1] - cov1[k - 1]))
print("first 6 2D sizes: " + " ".join(
    "(%d,%d)" % p for p in sorted(p2[:6], key=lambda t: t[1])))
for goal in (0.8, 0.9):
    need = [next((str(i + 1) for i, v in enumerate(c) if v >= goal),
                 "none of 16") for c in (cov1, cov2)]
    print("sizes to cover %.1f: 1D %s, 2D %s" % (goal, need[0], need[1]))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《人体测量主成分尺码体系、虚荣尺码漂移与跨境退货率最小化母典》第 3 章「身高—围度两维主成分网格:一根轴的标签丢掉了什么」(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-215 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026215,
  author    = {Fan, Yafu},
  title     = {Anthropometric Principal-Component Sizing, Vanity-Size Drift and Cross-Border Return-Rate Minimisation},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论