Slide 1 — Lecture 9: Probability
(第1页——第9讲:概率)
Knowledge Points · 要点
- Conditional probability(条件概率)
- Bayes’ theorem and applications(贝叶斯定理与应用)
Explanation · 解释
- Conditional probability measures the likelihood of an event given that another event has occurred:
- Bayes’ theorem inverts conditioning:
Example · 例子
- If 30% of students are finance majors and 80% of finance majors pass, while overall pass rate is 60%, then
Extension · 拓展
- Independence: (A) and (B) independent iff (P(A\mid B)=P(A))(等价于 (P(A\cap B)=P(A)P(B)))。
- Law of total probability partitions the sample space to compute marginals.
Image/Data Analysis · 图像/数据
- This slide frames the chapter roadmap: definitions → conditional probability → Bayes.
Summary · 小结
- 核心任务:学会条件概率的计算、识别独立性,并能用贝叶斯定理进行“反向推断”。
Slide 2 — Conditional Probability
(第2页——条件概率)
Knowledge Points · 要点
- Definition and notation (P(A\mid B))
- Relationship with joint and marginal probabilities
Explanation · 解释
- Definition: Given (B) has occurred, restrict the sample space to (B). Then
- Symmetry:
- Product rule(乘法法则的另一种写法):
Example · 例子
- Draw a card: (A={\text{Ace}}), (B={\text{Spade}})。
Extension · 拓展
- Chain rule for multiple events:
- When events are independent, (P(A\mid B)=P(A)).
Image/Data Analysis · 图像/数据
- Slide states definition and two formulas; emphasize that the denominator must be nonzero.
Summary · 小结
- 计算条件概率的步骤:找交集 → 找被条件事件概率 → 相除。注意独立性可简化计算。
Slide 3 — Example: Bradley Investments (Setup)
(第3页——示例:布拉德利的投资〈情境〉)
Knowledge Points · 要点
-
Two stocks: Markley Oil (M) and Collins Mining (C)
-
Payoff outcomes in $000 after 3 months:
- (M:{10,5,0,-20})
- (C:{8,-2})
Explanation · 解释
- We will model events like “(M) profitable” (payoff (>0)) and “(C) profitable”。
- Joint outcomes ((M,C)) each have an assigned probability (given next slide).
Example · 例子
- Define events: (M={\text{10, 5}}), (C={\text{8}})。
Extension · 拓展
- Profit thresholds can be changed (e.g., ( \ge 0 ) vs (>0)); be explicit because results depend on this definition.
Image/Data Analysis · 图像/数据
- A payoff table lists discrete outcomes for each stock; combine with a probability table to compute joints/marginals.
Summary · 小结
- 先定义事件,再从联合分布求边际与条件概率。
Slide 4 — Example: Bradley Investments (Computation)
(第4页——示例:布拉德利的投资〈计算〉)
Knowledge Points · 要点
- Events: (M={\text{M profitable}}), (C={\text{C profitable}})
- Given joint outcome probabilities(见右表)
Explanation · 解释
- From the table (highlighted rows are profitable–profitable):
- Marginal of (M): sum all rows with (M>0):
- Conditional of (C) given (M):
Example · 例子
- 求 (P(M\mid C)): 先算 再得
Extension · 拓展
- Independence test: compare 因为 (0.36\neq 0.336),两只股票利润并不独立(正相关)。
Image/Data Analysis · 图像/数据
- 表格列出了8个联合结果及其概率;边际通过“按行/列求和”,条件通过“交集 ÷ 边际”得到。
Summary · 小结
-
计算流程:
- 从联合表求交集概率;2) 求边际和;3) 用公式 (P(A\mid B)=P(A\cap B)/P(B));
- 对比 (P(A\cap B)) 与 (P(A)P(B)) 判断独立性。
Slide 5 — Multiplication Law
(第5页——乘法法则)
Knowledge Points · 要点
- Multiplication Law connects joint probability and conditional probability。
- Formula:
- Independence condition: only if
Explanation · 解释
- The multiplication law provides a method to compute the probability of two events happening together。
- Example: In Bradley Investments,
- It shows the relationship between conditional and joint probabilities.
📖 点击查看中文解释
乘法法则用于计算两个事件同时发生的概率。公式为:
它体现了联合概率、条件概率和边际概率之间的关系。若事件独立,则有
Example · 例子
For Markley Oil and Collins Mining:
Then
📖 点击查看中文例子
已知 (P(M)=0.70),(P(C\mid M)=0.5143),则
这说明两家公司同时盈利的概率为 36%。
Extension · 拓展
- Reverse formula:
- Symmetric for any two events (A,B)。
- Forms the foundation of Bayes’ theorem。
Image/Data Analysis · 图像/数据
- The formula allows deriving one probability type (joint, conditional, or marginal) from the others。
- This is visualized through the contingency table of Bradley Investments。
Summary · 小结
- 乘法法则是连接联合概率与条件概率的桥梁。
- 若独立则直接相乘;若非独立则需乘以条件概率修正。
Slide 6 — Example: Bradley Investments (Using Multiplication Law)
(第6页——布拉德利投资例:乘法法则计算)
Knowledge Points · 要点
- Compute (P(M \cap C)) using the multiplication law。
- Given (P(M)=0.70,\ P(C\mid M)=0.5143)。
Explanation · 解释
This result matches the table from the previous slide。
📖 点击查看中文解释
用乘法法则计算两事件同时发生的概率:
表示两家公司同时盈利的概率为 36%。
Example · 例子
If (P(M)=0.5) and (P(C\mid M)=0.3),then
Extension · 拓展
- If (P(M\cap C)) differs from (P(M)P(C)),then the two are not independent。
- This computation is the reverse of the conditional probability formula。
Image/Data Analysis · 图像/数据
- The table shows joint probabilities in the center ((.36)) and marginals ((.70,\ .48)) in the margins。
- Helps visualize intersection and dependence。
Summary · 小结
- 通过 (P(A\cap B)=P(A)P(B\mid A)) 可由边际和条件概率求出联合概率。
- 用于判断两个事件的相关性或独立性。
Slide 7 — Joint vs Marginal Probabilities
(第7页——联合概率与边际概率)
Knowledge Points · 要点
- Joint probability: probability that both events occur。
- Marginal probability: total probability of a single event, obtained by summing across the other variable。
Explanation · 解释
In the Bradley table:
- Joint probabilities (e.g., (P(M\cap C)=0.36)) are inside the table。
- Marginals (e.g., (P(M)=0.36+0.34=0.70), (P(C)=0.36+0.12=0.48)) are sums of row or column totals。
📖 点击查看中文解释
联合概率:两个事件同时发生的概率(表格中的中间部分)。 边际概率:一个事件发生的总概率(通过对另一维度求和获得)。
Example · 例子
From the table:
Extension · 拓展
- Law of total probability (for (C))
Image/Data Analysis · 图像/数据
- Red arrow → joint probabilities。
- Blue arrow → marginal probabilities。
- These form the foundation for calculating conditional probabilities and verifying Bayes’ theorem。
Summary · 小结
- 联合概率描述两个事件的共同发生;
- 边际概率是对一个事件的整体分布;
- 二者通过条件概率公式相互联系。
Slide 8 — Independent Events
(第8页——独立事件)
Knowledge Points · 要点
- Two events are independent if one’s occurrence does not change the probability of the other。
- Formula:
Explanation · 解释
- Independence equivalences
- For dependent events, we must use
📖 点击查看中文解释
若事件 (A) 的发生不会改变事件 (B) 的概率,则 (A) 与 (B) 独立:
否则为依赖事件。
Example · 例子
From Bradley:
Since (0.36\neq0.336), (M) and (C) are not independent。
Extension · 拓展
- For multiple events: independence requires all pairwise and group intersections to satisfy the product rule。
- Independence implies no predictive relationship。
Image/Data Analysis · 图像/数据
- Diagram: overlapping circles show independence when intersection equals product area。
- Deviation from equality shows correlation。
Summary · 小结
- 独立事件满足 (P(A\cap B)=P(A)P(B))。
- 若不满足,则为相关事件。
- 本例中两只股票正相关,非独立。
Slide 9 — Independent Events (continued)
(第9页——独立事件 补充)
Knowledge Points · 要点
- Using table data to test independence。
- Relationship between joint and marginal probabilities。
Explanation · 解释
Given
then
Since
Markley Oil (M) and Collins Mining (C) are not independent。
📖 点击查看中文解释
已知
而
二者不相等,因此 M 与 C 不独立。
Example · 例子
If events were independent, we should have
But observed data give 0.36 ⇒ slight positive correlation。
Extension · 拓展
- Independence implies knowledge of one event does not change the probability of the other。
- Dependence often indicates positive or negative association。
Image/Data Analysis · 图像/数据
- The table shows (P(M\cap C)=0.36), (P(M)=0.70), (P(C)=0.48)。
- Highlighted cells (.36, .70, .48) visualize marginal vs. joint probabilities。
Summary · 小结
- 检验独立性的方法是比较 (P(A\cap B)) 与 (P(A)P(B))。
- 若不相等,则事件存在相关性(正或负)。
Slide 10 — Mutual Exclusiveness vs. Independence
(第10页——互斥与独立的区别)
Knowledge Points · 要点
- Mutual exclusiveness (互斥) and independence (独立) are two different relationships。
- Mutually exclusive ⇢ cannot occur together。
- Independent ⇢ one’s occurrence does not affect the other。
Explanation · 解释
-
Two events with nonzero probabilities cannot be both mutually exclusive and independent。
- If (A) and (B) are mutually exclusive, then (P(A\cap B)=0)。
- Independence requires (P(A\cap B)=P(A)P(B)>0)(for nonzero probabilities)。 ⇒ Contradiction。
-
Therefore, mutual exclusiveness implies dependence。
Example · 例子
If event A = “roll 1”, event B = “roll 2” on a die, then (P(A\cap B)=0)。 → Mutually exclusive, but not independent。
Extension · 拓展
- If one mutually exclusive event occurs, the other cannot occur ⇒ they are dependent。
- If events are not mutually exclusive, they may or may not be independent—depends on data。
Summary · 小结
- 互斥 ⇒ 绝不独立;
- 不互斥 ⇒ 可能独立也可能不独立。
Slide 11 — Bayes’ Theorem
(第11页——贝叶斯定理)
Knowledge Points · 要点
- Connects prior, likelihood, and posterior probabilities。
- Updates belief in event A after observing event B。
Explanation · 解释
- Start with a prior probability of A: (P(A))。
- Collect new evidence B and know (P(B\mid A))。
- Compute the posterior using Bayes’ theorem:
- Denominator expanded via law of total probability:
📖 点击查看中文解释
贝叶斯定理通过“先验 → 似然 → 后验”更新概率:
它提供了一种利用新信息修正原有判断的方法。
Example · 例子
If (P(A)=0.3,\ P(B\mid A)=0.8,\ P(B)=0.6):
Extension · 拓展
- Common applications: diagnostic tests, spam filtering, investment forecasting。
- Posterior probability becomes the new prior when new data arrive。
Summary · 小结
- 贝叶斯定理帮助我们在获得新信息后动态更新信念。
- 后验 =(似然 × 先验)÷ 证据。
Slide 12 — Example: L.S. Clothiers
(第12页——案例:L.S. Clothiers)
Knowledge Points · 要点
- Illustrates application of Bayes’ theorem to decision making。
- Context: approval process for a shopping center project。
Explanation · 解释
A proposed shopping center may affect local business L.S. Clothiers。 If built, L.S. plans to relocate。 The planning board recommends approval or rejection to the council。 We evaluate probabilities of “approval” and “project built” using Bayes’ framework after new recommendation information arrives。
📖 点击查看中文解释
L.S. Clothiers 案例展示贝叶斯定理在决策分析中的应用: 在收到规划委员会的新建议(批准或不批准)后,更新“项目会被批准”的概率。
Example · 例子
- Suppose prior (P(\text{Approved})=0.6)。
- Board recommends approval with (P(\text{Recommend}\mid \text{Approved})=0.9)。
- Overall recommendation rate (P(\text{Recommend})=0.7)。 Then
Extension · 拓展
- Real decision makers use posterior probabilities to revise strategies。
- Example: L.S. may prepare relocation plans only if posterior > threshold (e.g., 0.75)。
Summary · 小结
- 案例体现了贝叶斯更新在商业与城市规划决策中的实际意义。
- 新证据 → 修正判断 → 更新决策。
Slide 13 — Example: L.S. Clothiers (Negative Recommendation)
(第13页——L.S. Clothiers 案例:负面推荐)
Knowledge Points · 要点
- The board recommended against building the shopping center.
- Define (B): event of a negative recommendation.
- Need to update belief about whether the town council will approve (A₁) or disapprove (A₂) the project.
Explanation · 解释
Past history gives:
Since event (B) occurred, L.S. should revise its probability assessments for (A_1) and (A_2).
📖 点击查看中文解释
规划委员会对建造购物中心给出了反对意见(事件 (B))。 过去经验显示:
已知 (B) 发生后,应当使用贝叶斯定理修正城镇委员会批准((A_1))或不批准((A_2))的概率。
Example · 例子
Given:
We will compute joint and posterior probabilities in the next slide.
Summary · 小结
- 定义事件:(A_1=\text{批准}),(A_2=\text{不批准}),(B=\text{反对推荐})。
- 目标:求 (P(A_1\mid B))、(P(A_2\mid B))。
Slide 14 — Example: L.S. Clothiers (Tree Diagram)
(第14页——L.S. Clothiers 树形图)
Knowledge Points · 要点
- Bayes tree representation of the relationship between (A_i) and (B)。
- Compute joint probabilities via multiplication law:
Explanation · 解释
From the diagram:
📖 点击查看中文解释
由树状图可得:
可看出委员会的推荐(B)与市政会的决策(A₁/A₂)之间存在明显依赖。
Example · 例子
Using the joint results, total probability of (B):
Summary · 小结
- 树形图帮助可视化条件概率和联合概率的关系。
- 下一步:用贝叶斯定理求后验 (P(A_1\mid B))。
Slide 15 — Bayes’ Theorem (Concept)
(第15页——贝叶斯定理:概念)
Knowledge Points · 要点
- Bayes’ theorem finds the posterior probability of event (A_i) given that (B) occurred。
- Applies when (A_1, A_2, \dots, A_n) form a mutually exclusive and exhaustive partition of the sample space.
Explanation · 解释
For events (A_1, A_2):
📖 点击查看中文解释
贝叶斯定理用于求事件 (A_i) 在 (B) 发生后出现的后验概率:
它要求各 (A_i) 相互互斥且穷尽样本空间(如 YES/NO)。
Example · 例子
Here,
Then posterior probabilities can be found using the next slide.
Summary · 小结
- 贝叶斯定理核心思想:新信息修正旧信念。
- (A_i) 必须是互斥且完备的事件集合。
Slide 16 — Bayes’ Theorem (Formula and Substitution)
(第16页——贝叶斯定理:公式与代入)
Knowledge Points · 要点
- Bayes’ formula (two-event version):
Explanation · 解释
Given:
Compute:
📖 点击查看中文解释
已知
则
即在委员会给出反对意见的情况下,项目被批准的概率降至约 34.1%。
Example · 例子
Interpretation: the new evidence (negative recommendation) reduces the chance of approval.
Extension · 拓展
- The denominator (P(B)) acts as a normalizing factor, ensuring total probability = 1.
- Posterior probabilities always sum to one:
Summary · 小结
- 通过贝叶斯定理,L.S. Clothiers 将“反对建议”纳入考虑后,重新估计批准概率。
- 新信息 → 更新后验概率 → 改变决策判断。
Slide 17 — Bayes’ Theorem (Calculation Example)
(第17页——贝叶斯定理:计算示例)
Knowledge Points · 要点
- Using Bayes’ formula to calculate ( P(A_1 \mid B) )。
- Formula for two mutually exclusive events (A_1, A_2):
Explanation · 解释
Substitute the values:
This is the posterior probability that the town council will approve the shopping center given the negative recommendation.
📖 点击查看中文解释
根据贝叶斯定理:
将题目数据代入:
即在规划委员会提出反对意见的情况下,市政会批准项目的概率为 34%。
Example · 例子
Compare posterior with prior:
→ Evidence (B) reduces confidence in project approval.
Summary · 小结
- 后验概率 (0.34) 明显低于先验概率 (0.70)。
- 规划委员会的意见显著影响市政会的决策。
Slide 18 — Example: L.S. Clothiers (Conclusion)
(第18页——L.S. Clothiers 案例结论)
Knowledge Points · 要点
- Interpreting the meaning of the posterior probability.
- Assessing how new information changes decisions.
Explanation · 解释
Conclusion: Since the board recommended against the new shopping center, the posterior probability that the council will approve the project is
reduced from the prior probability
This indicates that the board’s recommendation strongly influences the council’s final decision.
📖 点击查看中文解释
结论:由于规划委员会反对建设新的购物中心, 项目被批准的后验概率为
从原先的先验概率
显著下降。说明委员会的意见对市政会的决策具有重要影响。
Summary · 小结
- 贝叶斯更新后的结果显示:新证据(反对意见)使批准概率降低约一半。
- 说明决策过程对外部信号高度敏感。
Slide 19 — Example: L.S. Clothiers (Excel Table Representation)
(第19页——L.S. Clothiers 案例:Excel 表格计算)
Knowledge Points · 要点
- Excel formulas for computing prior, conditional, joint, and posterior probabilities.
- Demonstrates how Bayes’ theorem can be implemented using spreadsheet formulas.
| Events | Prior Probabilities | Conditional Probabilities | Joint Probabilities | Posterior Probabilities |
|---|---|---|---|---|
| (A_1) | ||||
| (A_2) | ||||
| SUM | — |
📖 点击查看中文解释
- 联合概率:。
- 后验概率:,其中 。
- 上表等价于 Excel 的计算流程,只是把单元格内容改为了数学公式的呈现方式。
Slide 20 — Example: L.S. Clothiers (Full Probability Table)
(第20页——L.S. Clothiers 案例:全概率表)
Knowledge Points · 要点
Relationships among probabilities (概率关系式)
📖 点击查看中文解释
全概率表中的关系式说明如下:
这些关系式表明:
- 第一式展示了事件 的全概率来源于 与 各自的交集;
- 第二与第三式分别对应 、 的全概率分解;
- 第四式体现了补事件 的全概率关系;
- 整体与贝叶斯树(Bayes Tree)结果一致,说明所有联合、边际与补事件概率相互平衡。
Summary · 小结
- 通过表格形式可一目了然地验证全概率定律与贝叶斯关系。
- 该案例完整体现了“先验 → 似然 → 后验”的推理过程。