What does Likelihood-Ratio Screening of Luxury Goods conclude, in numbers, about “Correcting for Correlation: the Effective Number of Independent Features, Exact Weights and a One-Number Slope”?
Answer: Chapter 3 gives the effective number of independent features (four features are worth only 2.007 under the illustrative matrix) and a one-number slope of 0.571.
Scope of the numbers: Brands are named only generically, the parameters are illustrative, all calculations are generated by the book's programs, and the book is not individual business advice. Chapter 3 gives the effective number of independent features (four features are worth only 2.007 under the illustrative matrix) and a one-number slope of 0.571.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.571, 2.007
| illustrative R: n | 4, n_eff = 1' R^-1 1 = 2.007 |
| feature naive weight d exact weight (R^-1 d) | |
| stitch | 1.200 0.849 |
| hardware | 1.000 -0.111 |
| spectrum | 1.500 1.351 |
| grain | 0.900 0.227 |
| claimed D^2 | 5.500 true D^2 = 3.140 slope k = 0.571 |
| expected weight under N: naive claims 11.94 deciban, true 6.82 | |
| equal correlation: n_eff = n / (1 + (n - 1) rho) | |
| n rho | 0.2 rho=0.4 rho=0.6 rho=0.8 |
| 2 | 1.67 1.43 1.25 1.11 |
| 3 | 2.14 1.67 1.36 1.15 |
| 4 | 2.50 1.82 1.43 1.18 |
| 6 | 3.00 2.00 1.50 1.20 |
| 8 | 3.33 2.11 1.54 1.21 |
| limit as n grows | 1 / rho = 5.00, 2.50, 1.67, 1.25 |
Show the full program
# Chapter 3. p31: the effective number of independent features. For a
# within-class correlation matrix R, n_eff = 1' R^-1 1 (Bayley and
# Hammersley's effective number, written for a mean of correlated
# scores); the exact separation is d' R^-1 d and the exact weights are
# R^-1 d. The 4 x 4 matrix below is illustrative: hardware weight and
# hardware spectrum share a supplier factor (0.6), stitch and grain share
# wear and lighting (0.5).
import math
NAMES = ["stitch", "hardware", "spectrum", "grain"]
D = [1.2, 1.0, 1.5, 0.9]
R = [[1.0, 0.3, 0.2, 0.5],
[0.3, 1.0, 0.6, 0.2],
[0.2, 0.6, 1.0, 0.2],
[0.5, 0.2, 0.2, 1.0]]
def solve(a, b):
# 高斯消元(部分主元),只用于小矩阵
n = len(b)
m = [row[:] + [b[i]] for i, row in enumerate(a)]
for c in range(n):
p = max(range(c, n), key=lambda r: abs(m[r][c]))
m[c], m[p] = m[p], m[c]
for r in range(c + 1, n):
f = m[r][c] / m[c][c]
m[r] = [x - f * y for x, y in zip(m[r], m[c])]
x = [0.0] * n
for r in range(n - 1, -1, -1):
x[r] = (m[r][n] - math.fsum(m[r][k] * x[k]
for k in range(r + 1, n))) / m[r][r]
return x
def dot(x, y):
return math.fsum(a * b for a, b in zip(x, y))
ones = [1.0] * 4
neff = dot(ones, solve(R, ones))
w = solve(R, D)
claim2, true2 = dot(D, D), dot(D, w)
print("illustrative R: n = 4, n_eff = 1' R^-1 1 = %.3f" % neff)
print("feature naive weight d exact weight (R^-1 d)")
for k in range(4):
print("%-9s %14.3f %20.3f" % (NAMES[k], D[k], w[k]))
print("claimed D^2 = %.3f true D^2 = %.3f slope k = %.3f"
% (claim2, true2, true2 / claim2))
print("expected weight under N: naive claims %.2f deciban, true %.2f"
% (10.0 / math.log(10.0) * claim2 / 2.0,
10.0 / math.log(10.0) * true2 / 2.0))
print("equal correlation: n_eff = n / (1 + (n - 1) rho)")
print(" n rho=0.2 rho=0.4 rho=0.6 rho=0.8")
for n in [2, 3, 4, 6, 8]:
print("%4d " % n + "".join("%9.2f" % (n / (1.0 + (n - 1) * r))
for r in (0.2, 0.4, 0.6, 0.8)))
print("limit as n grows: 1 / rho = 5.00, 2.50, 1.67, 1.25")Single source and citation
Fan, Y. (2026). 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 (Version 1.0.0), Chapter 3: Correcting for Correlation: the Effective Number of Independent Features, Exact Weights and a One-Number Slope. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 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}
}Other chapter answers in this book
- Chapter 1 · Weight of Evidence: Single-Feature Likelihood Ratios, Decibans and the Sum of Four Features
- Chapter 2 · Double Counting: How Correlated Features Make Naive Bayes Overconfident
- Chapter 4 · A Loss Matrix and Three Zones: Instant Buy, Referral and Decline
- Chapter 5 · Measuring More or Stopping: Cheap Evidence Is Worth Most at the Boundaries
- Chapter 6 · Channel Priors, Audits and Calibration: What Gets Referred Decides What a Store Can Learn
- Chapter 7 · East and West in the Sources: Suspecting the Neighbour of the Axe, and Peirce's Thermometer of Belief
- Chapter 8 · Closing the Loop: a Three-Zone Intake Procedure, the Refer-or-Buy Screener and a Pre-Registered Test