수리통계학 2강 - 확률집합함수 ~ 조건부 확률과 확률적 독립성

Mark·2022년 3월 18일
0

수리통계학

목록 보기
2/9

Thm 1.3.4

0P(C)1, cB0\le P(C)\le1, \ \forall c \in B
0=P(ϕ)P(c)P(C)=10=P(\phi) \le P(c) \le P(C) = 1

Thm 1.3.5

P(C1C2)=P(C1)+P(C2)P(C1C2)P(C_1 \cup C_2) = P(C_1) + P(C_2) - P(C_1 \cap C_2)

P(C1C2)=P(C1)+P(C1cC2)P(C_1 \cup C_2) = P(C_1) + P(C_1^c \cap C_2) (1)\cdot\cdot\cdot (1)

C2=(C1C2)(C1cC2)C_2 = (C_1 \cap C_2) \cup (C_1^c \cap C_2)

P(C2)=P(C1C2)+p(C1cC2)P(C_2) = P(C_1 \cap C_2) + p(C_1^c \cap C_2) (2)\cdot\cdot\cdot (2)
(1) - (2) -->
P(C1C2)P(c2)=P(C1)P(C1C2)P(C_1 \cup C_2) - P(c_2) = P(C_1) - P(C_1 \cap C_2)

Remark 1.3.1 (Inclusionexclusion formularInclusion - exclusion\ formular)

P(C1C2C3)=P1P2+P3P(C_1 \cup C_2 \cup C_3) = P_1 - P_2 + P_3
P1=P(C1)+P(C2)+P(C3)P_1 = P(C_1) + P(C_2) + P(C_3)
P2=P(C1C2)+P(C1C3)+P(C2C3)P_2 = P(C_1 \cap C_2) + P(C_1 \cap C_3) + P(C_2 \cap C_3)
P3=P(C1C2C3)P_3 = P(C_1 \cap C_2 \cap C_3)

In general,

P(C1C2C3Ck)=P1P2+P3+(1)k1PkP(C_1 \cup C_2 \cup C_3 \cup \cdot\cdot\cdot C_k) = P_1 - P_2 + P_3 - \cdot\cdot\cdot +(-1)^{k-1}P_k
pi:sum of all possible intersection of i setsp_i : sum\ of\ all\ possible\ intersection\ of\ i\ sets
*C2,C3:Mutually exclusiveC_2, C_3 \cdot\cdot\cdot: Mutually\ exclusive

Mutually exclusive sets C1,C2,are calledMutually\ exclusive\ sets\ C_1, C_2, \cdot\cdot\cdot are\ called Exhaaustive(만약 CkC_k의 합이 표본공간인 경우)

* limnCn\lim_{n\to\infin}C_n을 쓸 수 있는 특수한 경우

(1) Countable union and Increasing sets

n=1Cn, Cis:increasing set(C1C2C3 )\bigcup_{n=1}^\infin C_n,\ C_i's:increasing\ set(C_1 \subset C_2 \subset C_3\ \cdot\cdot\cdot\cdot)

(2) Countable union and Decreasing sets

n=1Cn, Cis:decreasing set(C1C2C3 )\bigcap_{n=1}^\infin C_n,\ C_i's:decreasing\ set(C_1 \supset C_2 \supset C_3\ \cdot\cdot\cdot\cdot)

ex) Cn=C_n = {x:0<x<11(n+1){x: 0 < x < 1- {1 \over (n+1)}}}
ex) Cn=C_n = {x:0<x<1+1(n+1){x: 0 < x < 1+ {1 \over (n+1)}}}

Thm 1.3.6

{CnC_n}: increasing set
limnP(Cn)\lim_{n \to \infin}P(C_n) := P(limnCn)P(\lim_{n \to \infin} C_n)
pf)

{CnC_n}: decreasing set
limnP(Cn)\lim_{n \to \infin}P(C_n) := P(limnCn)P(\lim_{n \to \infin} C_n)

Thm 1.3.7(Boole's inequality)

{CnC_n}: arbitrary seg of sets
P(n=1Cn)n=1P(Cn)P(\bigcup_{n=1}^ \infin C_n) \le \sum_{n=1}^ \infin P(C_n)

1.4 조건부확률과 독립성(Conditional Prob & Independence)

C1,C2CC_1, C_2 \in C,
The conditional probability of C2C_2 given C1C_1
P(C2C1)P(C_2 \mid C_1) := P(C2C1)P(C1)P(C_2 \cap C_1) \over P(C_1)
P(C1)ϕ*P(C_1) \ne \phi
(1) P(C2C1)0(nonnegativity)P(C_2 \mid C_1) \ge 0 (non-negativity)
(2) P(C2C2)=1(Normality)P(C_2 \mid C_2) = 1 (Normality)
(3) C2,C3,C4,:mutually exclusiveC_2, C_3, C_4, \cdot\cdot\cdot: mutually\ exclusive

-->P(i=1CiC1)=i=1P(CiC1)P(\bigcup_{i=1}^ \infin C_i \mid C_1)= \sum_{i=1}^ \infin P(C_i \mid C_1) <- countable additivity

ex1.4.1
포커 카드(52장)에서 비복원 추출 시, 6번째 뽑았던 것이 3번째 스페이드인 확률

--> 처음 5개 중 2개의 스페이드 출현
--> 마지막에 스페이드 출현

C1:two spades in the first five drawsC_1: two\ spades\ in\ the\ first\ five\ draws
C2:third spades in the sixth drawsC_2: third\ spades\ in\ the \ sixth\ draws
what is P(C1C2)?what\ is\ P(C_1 \cap C_2)?
--> 교집합을 구하기 쉽지 않으니, 조건부확률로 구하자

P(C2C1)P(C_2 \mid C_1) := P(C2C1)P(C1)P(C_2 \cap C_1) \over P(C_1) >>> P(C1C2)=P(C2C1)P(C_1 \cap C_2) = P(C_2 \mid C_1) *P(C1)P(C_1)

(1) P(C2C1)P(C_2 \mid C_1)는 52개의 카드 中 5개를 뽑았는데 2개의 스페이드가 나온 상황에서 또 6번째로 스페이드를 뽑을 확률

(P(C2C1)P(C_2 \mid C_1) = (132)(525)(13-2) \over (52-5) -> 47개의 패 중에서 스페이드를 뽑을 확률)

(2) P(C1)P(C_1)는 5개 뽑았는데 2개가 스페이드일 확률
P(C1)=P(C_1) = 13C252C5{}_{13}{\rm C}_{2} \over {}_{52}{\rm C}_{5}

(1) * (2) = 0.064

Joint Prob을 Marginal Prob과 Conditional Prob의 곱으로 나타내기

P(C1C2)=P(C1)P(C2C1P(C_1 \cap C_2) = P(C_1)*P(C_2 \mid C_1)

P(C3C1C2)=P(C_3 \mid C_1 \cap C_2) = P(C3C2C1)P(C1C2)P(C_3 \cap C_2 \cap C_1) \over P(C_1 \cap C_2)

P(C3C2C1)P(C_3 \cap C_2 \cap C_1) = P(C1C2)P(C3C1C2)P(C_1 \cap C_2)*P(C_3 \mid C_1 \cap C_2)

In general,
P(C1C2C3)P(C_1 \cap C_2 \cap C_3 \cdot\cdot\cdot)
= P(C1)P(C2C1)P(C3C1C2)P(C_1)*P(C_2 \mid C_1)*P(C_3 \mid C_1 \cap C_2) \cdot\cdot\cdot

Bayes Theorm

C1,C2,C3C_1, C_2, C_3 \cdot\cdot\cdot: mutually exclusive & exhaustive
P(Ci)>0,i=1,,kP(C_i) > 0, \forall i = 1, \cdot\cdot\cdot , k

P(CjC)P(C_j \mid C) = P(Cj)P(CCj)i=1kP(Ci)P(CCi){P(C_j)*P(C \mid C_j)} \over {\sum_{i=1}^k P(C_i)*P(C \mid C_i)}

  • 좌변이 구하기 어려운 경우이거나 비현실적인 경우, 우변을 통해 우회하여 구하는 것
    pf)

Remark 1.4.1 베이지안 기초 용어

  • P(Cj):Prior probP(C_j): Prior\ prob(사전확률)
  • P(CjC):Posterior probP(C_j \mid C): Posterior\ prob(사후확률)

Def 1.4.1 독립의 정의(왜 독립인 두사건의 교집합은 곱으로 나타내는가)

C1,C2:(Statistical) independent iffC_1, C_2: (Statistical)\ independent\ iff
P(C1C2)=P(C1)P(C_1 \mid C_2) = P(C_1)

In general,
C1,C2,independentC_1, C_2, \cdot\cdot\cdot independent
iff every collection of k event(2 kn\le k \le n)
P(Ci1Ci2Cik)P(C_{i1} \cap C_{i2} \cdot\cdot\cdot \cap C_{ik}) = j=1kP(Cij)\prod_{j=1}^k P(C_{ij})

cf)
Ci,Cj:pairwise independent ifC_i,C_j: pairwise\ independent\ if
P(CiCj)=P(Ci)P(Cj), ijP(C_i \cap C_j) = P(C_i)P(C_j),\ \forall i \ne j

0개의 댓글