关于「贝叶斯知识追踪:四个参数、掌握线与可识别性」,《零泄题苏格拉底考官》给出的定量结论是什么?
结论:第三、四章讲知识追踪的四参数更新、掌握线与可识别性,并说明五分之一的学员拿到答案时,假证书占全部证书的 0.2534,而事后拟合把它解释成生源更好。
数据口径:书中参数为示意;全部计算由书中程序生成,不调用任何模型接口,不含任何密钥或服务器地址,也不收录具体的套话提示词;本书是非个人化经营建议。 第一、二章量化叮嘱为何无效,在每题 3 次尝试、40 道题的示意课程里,p 等于 0.001 时一个 30 人的小班至少漏掉一道题的概率是 0.9727;再画出四种考官架构的信息流图,计算判分预言机的比特上限与尝试上限的代价。
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。
| step resp | P(L|obs) | P(L next) | P(correct next) | mastered | |
| 0 | - | - | 0.3000 | 0.4100 | |
| 1 | 0 | 0.0508 | 0.1458 | 0.3020 | no |
| 2 | 1 | 0.4343 | 0.4909 | 0.5436 | no |
| 3 | 1 | 0.8127 | 0.8314 | 0.7820 | no |
| 4 | 0 | 0.3814 | 0.4433 | 0.5103 | no |
| 5 | 1 | 0.7818 | 0.8036 | 0.7625 | no |
| 6 | 1 | 0.9485 | 0.9536 | 0.8675 | yes |
| 7 | 1 | 0.9893 | 0.9904 | 0.8933 | yes |
| 8 | 1 | 0.9978 | 0.9981 | 0.8986 | yes |
| 9 | 1 | 0.9996 | 0.9996 | 0.8997 | yes |
| mastery 0.95 first reached at step 6 |
查看完整验算程序
# Chapter 3. p31: Bayesian knowledge tracing, one student, one skill.
# The four parameters of Corbett and Anderson's model (values are
# illustrative, of the size often reported):
# L0 probability the skill is known before the first item 0.30
# T probability of learning at each opportunity 0.10
# G probability of a correct answer without the skill 0.20
# S probability of a wrong answer despite the skill 0.10
# After each graded response:
# correct: L|obs = L(1-S) / (L(1-S) + (1-L)G)
# wrong: L|obs = LS / (LS + (1-L)(1-G))
# then L' = L|obs + (1 - L|obs) T
# Only the verdict bit enters the update; the answer itself never does.
L0, T, G, S = 0.30, 0.10, 0.20, 0.10
MASTERY = 0.95
def update(p, correct):
if correct:
post = p * (1 - S) / (p * (1 - S) + (1 - p) * G)
else:
post = p * S / (p * S + (1 - p) * (1 - G))
return post, post + (1 - post) * T
def p_correct(p):
return p * (1 - S) + (1 - p) * G
seq = (0, 1, 1, 0, 1, 1, 1, 1, 1)
p = L0
print("step resp P(L|obs) P(L next) P(correct next) mastered")
print("0 - - %.4f %.4f" % (p, p_correct(p)))
done = None
for i, r in enumerate(seq, 1):
post, p = update(p, r)
flag = "yes" if p >= MASTERY else "no"
if p >= MASTERY and done is None:
done = i
print("%-4d %-5d %-9.4f %-10.4f %-16.4f %s"
% (i, r, post, p, p_correct(p), flag))
print("mastery %.2f first reached at step %s" % (MASTERY, done))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《零泄题苏格拉底闯关考官、服务端答案物理隔离与掌握度贝叶斯追问树母典》第 3 章「贝叶斯知识追踪:四个参数、掌握线与可识别性」(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-262 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026262,
author = {Fan, Yafu},
title = {A Zero-Leak Socratic Examiner: Server-Side Isolation of Answers and Bayesian Mastery Probing Trees},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}