关于「保证的价钱:区间宽度、得分形状与正态成立时多付的那一截」,《共形预测与订单区间》给出的定量结论是什么?

结论:对数正态零点五一行,神谕宽度是 2.289,正态区间的中位宽度是 2.367,共形区间是 2.960。对数正态一的一行,神谕 6.958,正态中位 7.008,共形中位 10.301。最后一行把帕累托 2.1 的中位宽度除以神谕:正态 0.99,共形 1.49。

数据口径:书中参数为示意;全部计算由书中程序生成;本书属于非个人化经营建议。 本书作蓝图修正:正态区间的缺口是本书程序的模拟结果,集中在右尾与单侧上界;东方原典句出自《元史·郭守敬传》。

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: 0.99, 1.49, 10.301, 2.1, 2.289, 2.367

width in units of the shape's scale, n30
shape a oracle normal mean/med conformal mean/med
lognorm0.5 2.289 2.446 / 2.367 3.297 / 2.960
lognorm1.0 6.958 7.999 / 7.008 12.979 / 10.301
pareto3.0 2.411 2.935 / 2.433 4.873 / 3.602
pareto2.1 4.781 6.623 / 4.737 11.452 / 7.138
Pareto 2.1 median width over oracle: normal 0.99, conformal 1.49
查看完整验算程序
# Chapter 3. p31: what the guarantee costs in interval width.
# Model: n = 30 past values; normal-theory PI (t = 2.0452) versus
# split conformal (10 fit a median centre, 20 calibrate, k = 20).
# The yardstick is the oracle equal-tailed interval from the true
# quantiles at 0.025 and 0.975. Widths are averaged (mean and median)
# over 8,000 training sets. Shapes are illustrative.
import math
import random
import statistics

ND = statistics.NormalDist()


def make(kind, a):
    if kind == "lognorm":
        return (lambda r: math.exp(r.gauss(0.0, a)),
                lambda p: math.exp(a * ND.inv_cdf(p)))
    return (lambda r: (1.0 - r.random()) ** (-1.0 / a),
            lambda p: (1.0 - p) ** (-1.0 / a))


def widths(kind, a, reps=8000):
    rng = random.Random(20261113)
    draw, q = make(kind, a)
    wn, wc = [], []
    for _ in range(reps):
        xs = [draw(rng) for _ in range(30)]
        f = statistics.stdev(xs) * math.sqrt(1.0 + 1.0 / 30)
        wn.append(2 * 2.0452 * f)
        c = statistics.median(xs[:10])
        sc = sorted(abs(x - c) for x in xs[10:])
        wc.append(2 * sc[19])
    oracle = q(0.975) - q(0.025)
    return oracle, wn, wc


if __name__ == "__main__":
    print("width in units of the shape's scale, n = 30")
    print("shape     a    oracle  normal mean/med  conformal mean/med")
    for kind, a in (("lognorm", 0.5), ("lognorm", 1.0),
                    ("pareto", 3.0), ("pareto", 2.1)):
        o, wn, wc = widths(kind, a)
        print("{:<8s} {:3.1f}   {:6.3f}  {:6.3f} / {:6.3f}   "
              "{:6.3f} / {:6.3f}"
              .format(kind, a, o, statistics.fmean(wn),
                      statistics.median(wn), statistics.fmean(wc),
                      statistics.median(wc)))
    o, wn, wc = widths("pareto", 2.1)
    print("Pareto 2.1 median width over oracle: normal {:.2f}, "
          "conformal {:.2f}".format(statistics.median(wn) / o,
                                    statistics.median(wc) / o))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《共形预测与无分布区间、订单量与客单价有限样本覆盖保证母典》第 3 章「保证的价钱:区间宽度、得分形状与正态成立时多付的那一截」(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-306 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026306,
  author    = {Fan, Yafu},
  title     = {Conformal Prediction and Distribution-Free Intervals: Finite-Sample Coverage Guarantees for Order Counts and Order Values},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论