关于「渠道先验、抽检与校准:谁被送检,决定店里能学到什么」,《奢品多特征似然比初筛》给出的定量结论是什么?
结论:先验决定同一份证据落在哪一区。示意参数下,往 N 方向 4.0 分贝班的证据,在回头寄卖客渠道秒收,在个人卖家与整批货渠道送检;整批货只有 0.400 的一致货能秒收,秒收区里的不符比例却最高。先验只能按渠道等经营事实来分,不能按卖家的个人特征。店里只知道送检货的结果,这会让两端的失准看不见;送检区本身已足以暴露朴素后验的过度自信,秒收区与婉拒区则需要随机抽检,而二十分之一的比例远远不够。墨菲分解显示,朴素后验的失分几乎全在可靠度上。下一章暂离收货台,到两部原典里看看,古人怎样谈论被重复计数的证据。
数据口径:书中全部概率只是外观概率参考,不是官方鉴定证书,不构成真伪结论,涉及争议或高价值货品须送品牌或专业鉴定机构;品牌一律以通称指代;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建议。 第二章是全书核心:示意参数下,朴素后验判到 0.99 以上、平均 0.996 的货,实际不符只有 0.756;放大倍数随类内相关升高,在相关系数 0.6 时达到 2.28;但排序并不吃亏,伤害集中在远离先验的边界上。
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: 0.400, 4.0
| channel prior buy needs decline needs | buy share N share | |
| (deciban) (deciban) | of G in buy | |
| repeat | 0.02 4.46 23.98 | 0.929 0.0083 |
| first-time | 0.10 -2.90 16.62 | 0.695 0.0163 |
| bulk lot | 0.30 -8.76 10.76 | 0.400 0.0224 |
| one item with weight 4.0 deciban toward N: | |
| repeat P(N) | 0.0488 zone buy |
| first-time P(N) | 0.2182 zone refer |
| bulk lot P(N) | 0.5184 zone refer |
查看完整验算程序
# Chapter 6. p61: the same evidence, three channels. Base rates of N by
# intake channel are illustrative: repeat consignor 0.02, first-time
# private seller 0.10, bulk lot from an unknown supplier 0.30. With the
# zones of p41 (buy below 0.0539, decline above 0.8362) the weight of
# evidence an item needs depends on the channel. The exact log ratio is
# normal with mean -/+ T2 / 2 and variance T2 under G / N, where T2 =
# d' R^-1 d = 3.140 for matrix R of p31; shares follow from the normal
# distribution function.
import math
from statistics import NormalDist
T2 = 3.140
Q_LO, Q_HI = 0.0539, 0.8362
DB = 10.0 / math.log(10.0)
PHI = NormalDist()
def logit(p):
return math.log(p / (1.0 - p))
print("channel prior buy needs decline needs | buy share N share")
print(" (deciban) (deciban) | of G in buy")
for name, prior in (("repeat", 0.02), ("first-time", 0.10),
("bulk lot", 0.30)):
need_buy = logit(Q_LO) - logit(prior)
need_dec = logit(Q_HI) - logit(prior)
sd = math.sqrt(T2)
g_buy = PHI.cdf((need_buy + T2 / 2.0) / sd)
n_buy = PHI.cdf((need_buy - T2 / 2.0) / sd)
nb = prior * n_buy
share = nb / (nb + (1.0 - prior) * g_buy)
print("%-10s %6.2f %10.2f %13.2f | %8.3f %9.4f"
% (name, prior, DB * need_buy, DB * need_dec, g_buy, share))
item = 4.0
print("one item with weight %.1f deciban toward N:" % item)
for name, prior in (("repeat", 0.02), ("first-time", 0.10),
("bulk lot", 0.30)):
q = 1.0 / (1.0 + math.exp(-(logit(prior) + item / DB)))
zone = "buy" if q < Q_LO else ("decline" if q > Q_HI else "refer")
print(" %-10s P(N) = %.4f zone %s" % (name, q, zone))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《奢品真伪多特征似然比鉴定、走线节距五金光谱皮质纹理融合与送检秒收贝叶斯边界母典》第 6 章「渠道先验、抽检与校准:谁被送检,决定店里能学到什么」(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-281 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026281,
author = {Fan, Yafu},
title = {Multi-Feature Likelihood-Ratio Screening of Pre-Owned Luxury Goods: Fusing Stitch Pitch, Hardware Spectra and Leather Grain, and the Bayesian Boundary between Lab Referral and Instant Purchase},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}