关于「证据权重:单一特征的似然比、分贝班与四种特征的相加」,《奢品多特征似然比初筛》给出的定量结论是什么?
结论:一个特征给出一个似然比,取对数后得到可加的证据权重,单位可以用分贝班。示意参数下,走线针距的中性点是 0.60 个离散单位,单件期望权重 3.127 分贝班,标准差 5.212,单一特征很少能单独定案。四个特征朴素相加时,与规格不符的货平均给出 11.94 分贝班。相加的前提是四个读数在每个假设之内彼此独立。下一章检验这个前提:货品的磨损、拍照光线、同一位操作员,都会让几个特征一起偏,这时朴素相加就会把同一份证据记上好几次。
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: 0.60, 3.127, 5.212
| prior P(N) = 0.10 | prior odds = 0.1111 | ||
| z | LR | deciban | P(N | z) |
| -1.0 | 0.147 | -8.34 | 0.0160 |
| -0.5 | 0.267 | -5.73 | 0.0288 |
| 0.0 | 0.487 | -3.13 | 0.0513 |
| 0.6 | 1.000 | 0.00 | 0.1000 |
| 1.0 | 1.616 | 2.08 | 0.1522 |
| 1.5 | 2.945 | 4.69 | 0.2465 |
| 2.0 | 5.366 | 7.30 | 0.3735 |
| 2.5 | 9.777 | 9.90 | 0.5207 |
| 3.0 | 17.814 | 12.51 | 0.6644 |
| neutral score (LR = 1): z = 0.60 | |||
| score needed to reach P(N) = 0.5 from 0.10: z = 2.431 | |||
| expected weight per item under N: 0.720 nats = 3.127 deciban | |||
| spread of the weight (sd, either class): 5.212 deciban |
查看完整验算程序
# Chapter 1. p11: one feature, one likelihood ratio, one weight of evidence.
# Stitch-pitch deviation score z, in units of the store's own reference
# spread for brand A (illustrative). Under G (consistent with the brand A
# reference) z ~ N(0, 1); under N (not consistent with it) z ~ N(d, 1),
# d = 1.2. Both distributions and the prior are illustrative only.
# Weight of evidence in decibans (Turing and Good): 10 * log10(LR).
# 外观概率参考,不构成真伪结论。
import math
D = 1.2
PRIOR_N = 0.10 # 示意:个人寄卖渠道里与参考规格不符的比例
DB = 10.0 / math.log(10.0) # 1 自然对数单位 = 4.343 分贝班
def llr(z, d=D):
# ln LR = ln f(z | N) - ln f(z | G),等方差正态
return d * z - d * d / 2.0
def post(prior, ln_lr):
o = prior / (1.0 - prior) * math.exp(ln_lr)
return o / (1.0 + o)
print("prior P(N) = %.2f prior odds = %.4f"
% (PRIOR_N, PRIOR_N / (1.0 - PRIOR_N)))
print(" z LR deciban P(N | z)")
for z in [-1.0, -0.5, 0.0, 0.6, 1.0, 1.5, 2.0, 2.5, 3.0]:
w = llr(z)
print("%5.1f %9.3f %9.2f %10.4f"
% (z, math.exp(w), DB * w, post(PRIOR_N, w)))
print("neutral score (LR = 1): z = %.2f" % (D / 2.0))
z_even = (math.log(9.0) + D * D / 2.0) / D
print("score needed to reach P(N) = 0.5 from 0.10: z = %.3f" % z_even)
mean_w = D * D / 2.0
print("expected weight per item under N: %.3f nats = %.3f deciban"
% (mean_w, DB * mean_w))
print("spread of the weight (sd, either class): %.3f deciban" % (DB * D))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 1 章「证据权重:单一特征的似然比、分贝班与四种特征的相加」(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}
}