关于「说真话与短名单:提议方的诚实与接受方的赌注」,《稳定匹配与无私奔配对》给出的定量结论是什么?
结论:第四章说明提议方说真话不吃亏,接受方中有 0.233 能靠谎报得益,但缩短名单是一场可能落空的赌注。
数据口径:书中的人物、档期与偏好均为示意;全部计算由书中程序生成;本书是非个人化经营建议,只使用当事人明示提供的偏好,不做未经同意的画像,配对结果只是建议,双方都可以拒绝。 第一章在示意扫描中发现口味越一致,先到先得越糟,平均私奔对从 4.14 个升到 14.12 个;评分优先只在所有买手看法完全一致时才稳定;罗斯对英国八个地区的比较单看计数,单侧费希尔检验 p 值为 0.2143,须与定理和实验复现一起才站得住。
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: 0.233
| markets 500 of 4x4, every agent tries all 65 reported lists | |
| creators (proposers) who can gain by lying | 0 of 2000 |
| brands (receivers) who can gain by lying | 465 of 2000 (0.233) |
| markets where at least one brand can gain | 188 of 500 (0.376) |
| mean gain of a gaining brand | 1.64 places |
| brands whose best lie beats their brand-optimal stable partner | 0 |
查看完整验算程序
# Chapter 4. p41: who can gain by misstating a ranking? Exhaustive check.
# Model: 4 creators propose to 4 brands, uniformly random true rankings,
# fixed seed, 500 illustrative markets. For every agent we try every
# possible reported list (each ordered subset of the other side: 65
# lists, truncations included) with everyone else truthful, and ask
# whether the outcome under the TRUE ranking beats truth-telling.
# Theory: proposers can never gain (Dubins and Freedman 1981; Roth
# 1982); receivers sometimes can (Roth 1982).
import itertools
import math
import random
N, MARKETS = 4, 500
rng = random.Random(20261113)
LISTS = [list(p) for k in range(N + 1)
for p in itertools.permutations(range(N), k)]
def deferred(pp, rp):
rank = [{p: i for i, p in enumerate(v)} for v in rp]
nxt, held, free = [0] * N, [None] * N, list(range(N))
while free:
p = free.pop()
while nxt[p] < len(pp[p]) and p not in rank[pp[p][nxt[p]]]:
nxt[p] += 1
if nxt[p] >= len(pp[p]):
continue
r = pp[p][nxt[p]]
nxt[p] += 1
cur = held[r]
if cur is None:
held[r] = p
elif rank[r][p] < rank[r][cur]:
held[r] = p
free.append(cur)
else:
free.append(p)
return held
def value(true, partner):
return N - true.index(partner) if partner is not None else 0
prop_gain = rec_gain = rec_any = beyond = 0
rec_size = []
for _ in range(MARKETS):
kp = [rng.sample(range(N), N) for _ in range(N)]
bp = [rng.sample(range(N), N) for _ in range(N)]
held = deferred(kp, bp)
part_k = {k: b for b, k in enumerate(held) if k is not None}
for k in range(N):
base = value(kp[k], part_k.get(k))
for lst in LISTS:
h = deferred(kp[:k] + [lst] + kp[k + 1:], bp)
got = {kk: b for b, kk in enumerate(h) if kk is not None}
if value(kp[k], got.get(k)) > base:
prop_gain += 1
break
bh = deferred(bp, kp) # 品牌提议:品牌最优稳定配对
bopt = [None] * N
for k, b in enumerate(bh):
if b is not None:
bopt[b] = k
any_here = False
for b in range(N):
base = value(bp[b], held[b])
best = base
for lst in LISTS:
h = deferred(kp, bp[:b] + [lst] + bp[b + 1:])
best = max(best, value(bp[b], h[b]))
if best > base:
rec_gain += 1
any_here = True
rec_size.append(best - base)
if best > value(bp[b], bopt[b]):
beyond += 1
if any_here:
rec_any += 1
print("markets %d of %dx%d, every agent tries all %d reported lists"
% (MARKETS, N, N, len(LISTS)))
print("creators (proposers) who can gain by lying: %d of %d"
% (prop_gain, MARKETS * N))
print("brands (receivers) who can gain by lying: %d of %d (%.3f)"
% (rec_gain, MARKETS * N, rec_gain / (MARKETS * N)))
print("markets where at least one brand can gain: %d of %d (%.3f)"
% (rec_any, MARKETS, rec_any / MARKETS))
print("mean gain of a gaining brand: %.2f places"
% (math.fsum(rec_size) / len(rec_size)))
print("brands whose best lie beats their brand-optimal stable partner: %d"
% beyond)唯一数据源与引用
范雅福(Fan, Y.). (2026). 《稳定匹配盖尔沙普利算法、买手客户达人品牌摄影师档期的无私奔配对母典》第 4 章「说真话与短名单:提议方的诚实与接受方的赌注」(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-299 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026299,
author = {Fan, Yafu},
title = {Gale-Shapley Stable Matching: Pairings without Blocking Pairs for Buyers and Clients, Creators and Brands, Photographers and Slots},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}