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。

English version

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 respP(L|obs)P(L next)P(correct next)mastered
0--0.30000.4100
100.05080.14580.3020no
210.43430.49090.5436no
310.81270.83140.7820no
400.38140.44330.5103no
510.78180.80360.7625no
610.94850.95360.8675yes
710.98930.99040.8933yes
810.99780.99810.8986yes
910.99960.99960.8997yes
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}
}

本书其他章节结论