记录一下联合概率分布,以及概率推理。
概率公理
对于任意的命题A,B
- 0 ≤ P ( A ) ≤ 1 0 \le P(A) \le 1 0≤P(A)≤1
- P ( t r u e ) = 1 P(true) = 1 P(true)=1 and P ( f a l s e ) = 0 P(false) = 0 P(false)=0
- P ( A ∩ B ) = P ( A ) + P ( B ) − P ( A ∩ B ) P(A \cap B) = P(A) + P(B) - P(A \cap B) P(A∩B)=P(A)+P(B)−P(A∩B)
先验概率
命题的先验概率/无条件概率:没有任何其他信息存在的情况下关于命题的信度。
先验概率分布:对所有可能的情况分配概率值(是一个分布,有多个值),如:
P
(
W
e
a
t
h
e
r
)
=
<
0.72
,
0.1
,
0.08
,
0.1
>
(
需
要
归
一
化
)
P(Weather) = <0.72, 0.1, 0.08, 0.1> (需要归一化)
P(Weather)=<0.72,0.1,0.08,0.1>(需要归一化)。
联合概率分布:对随机变量集合给出所有可能的条件取值
P
(
W
e
a
t
h
e
r
,
C
a
v
i
t
y
)
=
一
个
4
∗
2
的
矩
阵
P(Weather, Cavity) = 一个4*2 的矩阵
P(Weather,Cavity)=一个4∗2的矩阵:
任何一个概率查询都能够从全联合概率分布得到解答
条件概率
- 条件概率/后验概率: 得到关于随机变量的某些证据(分母),即在某些情况下,所求命题的概率:如$P(cavity | toothache) = 0.8 ,当toothache已知的前提下有cavity的概率 $
- 条件概率分布的表示:$P(Cavity | Toothache) = 2维向量)
- 若我们有额外的 证据,若cavity已知,那 p ( c a v i t y ∣ t o o t h a c h e , c a v i t y ) = 1 p(cavity | toothache, cavity) = 1 p(cavity∣toothache,cavity)=1
- 若两个时间间不相关则概率式子可以简化, 如 p ( c a v i t y ∣ t o o t h a c h e , s u n n y ) = P ( c a v i t y ∣ t o o t h a c h e ) = 0.8 p(cavity | toothache, sunny) = P(cavity | toothache) = 0.8 p(cavity∣toothache,sunny)=P(cavity∣toothache)=0.8
- 以上的这些判断都是依赖于领域知识的,就是针对具体事件的。
条件概率: P ( a ∣ b ) = P ( a ∩ b ) / P ( b ) P(a | b) = P(a \cap b)/ P(b) P(a∣b)=P(a∩b)/P(b) if P ( b ) > 0 P(b) > 0 P(b)>0乘法规则: P ( a ∩ b ) = P ( a ∣ b ) P ( b ) = P ( b ∣ a ) P ( a ) P(a \cap b) = P(a | b)P(b) = P(b | a)P(a) P(a∩b)=P(a∣b)P(b)=P(b∣a)P(a)。联合概率分布可以写成条件分布的形式 P ( W e a t h e r , C a v i t y ) = P ( W e a t h e r ∣ C a v i t y ) P ( C a v i t y ) P(Weather , Cavity) = P(Weather | Cavity)P(Cavity) P(Weather,Cavity)=P(Weather∣Cavity)P(Cavity)
联合概率分布可以使用链式法则:
利用全联合分布进行推理
全联合分布:
对于任意命题 ψ \psi ψ, 计算所有为真的原子命题: p ( ψ ) = ∑ ω ψ p ( ω ) p(\psi) = \sum_\omega \psi p(\omega) p(ψ)=∑ωψp(ω)
如上, P ( t o o t h a c h e ) = 0.108 + 0.012 = 0.016 + 0.064 = 0.2 P(toothache) = 0.108+0.012=0.016+0.064 = 0.2 P(toothache)=0.108+0.012=0.016+0.064=0.2
全联合概率分布:
P
(
c
a
v
i
t
y
∨
t
o
o
t
h
a
c
h
e
)
=
0.018
+
0.012
+
0.016
+
0.064
+
0.072
+
0.008
=
0.28
P(cavity \vee toothache) = 0.018+0.012+0.016+0.064+0.072+0.008 = 0.28
P(cavity∨toothache)=0.018+0.012+0.016+0.064+0.072+0.008=0.28
计算条件概率
p ( ¬ c a v i t y ∩ t o o t h a c h e ) = P ( ¬ c a v i t y ∩ t o o t h a c h e ) / P ( t o o t h a c h e ) = ( 0.016 + 0.064 ) / ( 0.108 + 0.012 + 0.016 + 0.064 ) = 0.4 p(\neg cavity \cap toothache) = P(\neg cavity \cap toothache) /P(toothache) = {(0.016 + 0.064)}/{(0.108 + 0.012 + 0.016 + 0.064)} = 0.4 p(¬cavity∩toothache)=P(¬cavity∩toothache)/P(toothache)=(0.016+0.064)/(0.108+0.012+0.016+0.064)=0.4
归一化
分母可以看成一个归一化常数a,因为分母是一个常数,所以直接用a表示:
计算询问变量的概率分布,是通过固定证据变量(分母),然后对隐变量求和计算得到的。
一般的,我们求的是已知证据变量E情况下的查询变量Y的后验概率分布。隐变量为:H = X - Y - E。
证据变量和查询变量的联合分布是通过对隐变量求和得到:
P
(
Y
∣
E
=
e
)
=
a
P
(
Y
,
E
=
e
)
=
a
∑
h
P
(
Y
,
E
=
e
,
H
=
h
)
P(Y | E = e) = aP(Y,E = e) = a\sum hP(Y,E = e, H = h)
P(Y∣E=e)=aP(Y,E=e)=a∑hP(Y,E=e,H=h)
独立性
A,B独立当且仅当:
P
(
A
∣
B
)
=
P
(
A
)
P(A | B) = P(A)
P(A∣B)=P(A) or
P
(
B
∣
A
)
=
P
(
B
)
P(B | A) = P(B)
P(B∣A)=P(B) or
P
(
A
,
B
)
=
P
(
A
)
P
(
B
)
P(A, B) = P(A)P(B)
P(A,B)=P(A)P(B)
通过条件独立,可以在一定程度上减少概率分布的表的一些不必要的数据。
P
(
c
a
t
c
h
∣
t
o
o
t
h
a
c
h
e
,
c
a
v
i
t
y
)
=
P
(
c
a
t
c
h
∣
c
a
v
i
t
y
)
P(catch | toothache, cavity) = P(catch | cavity)
P(catch∣toothache,cavity)=P(catch∣cavity)
贝叶斯规则
利用贝叶斯规则和条件独立性解下题:
P
(
C
a
v
i
t
y
∣
t
o
o
t
h
a
c
h
e
∨
c
a
t
c
h
)
=
P
(
t
o
o
t
h
a
c
h
e
∨
c
a
t
c
h
∣
C
a
v
i
t
y
)
P
(
C
a
v
i
t
y
)
/
P
(
t
o
o
t
h
a
c
h
e
∨
c
a
t
c
h
)
=
a
P
(
t
o
o
t
h
a
c
h
e
∨
c
a
t
c
h
∣
C
a
v
i
t
y
)
P
(
C
a
v
i
t
y
)
=
a
P
(
t
o
o
t
h
a
c
h
e
∣
C
a
v
i
t
y
)
P
(
c
a
t
c
h
∣
C
a
v
i
t
y
)
P
(
C
a
v
i
t
y
)
=
a
<
0.108
,
0.016
>
≈
<
0.871
,
0.129
>
P(Cavity | toothache \vee catch) = P(toothache \vee catch | Cavity)P(Cavity) / P(toothache \vee catch) = aP(toothache \vee catch | Cavity)P(Cavity) = aP(toothache|Cavity)P(catch|Cavity)P(Cavity) = a<0.108, 0.016> \approx <0.871,0.129>
P(Cavity∣toothache∨catch)=P(toothache∨catch∣Cavity)P(Cavity)/P(toothache∨catch)=aP(toothache∨catch∣Cavity)P(Cavity)=aP(toothache∣Cavity)P(catch∣Cavity)P(Cavity)=a<0.108,0.016>≈<0.871,0.129>
上面式子倒数第二步直接分开,是因为toothache和catch是条件独立的。
P ( t o o t h a c h e ) = 0.108 + 0.012 + 0.016 + 0.064 = 0.2 P(toothache)=0.108+0.012+0.016+0.064 = 0.2 P(toothache)=0.108+0.012+0.016+0.064=0.2
P ( C a v i t y ) = < 0.108 + 0.012 + 0.072 + 0.008 , 0.016 + 0.064 + 0.144 + 0.576 > = < 0.2 , 0.8 > P(Cavity) = <0.108+0.012+0.072+0.008, 0.016+0.064+0.144+0.576> = <0.2,0.8> P(Cavity)=<0.108+0.012+0.072+0.008,0.016+0.064+0.144+0.576>=<0.2,0.8>
P ( T o o t h a c h e ∣ C a v i t y ) = α P ( T o o t h a c h e , C a v i t y , C a t c h ) = α < [ P ( t o o t h a c h e , c a v i t y , c a t c h ) + P ( t o o t h a c h e , ┐ c a v i t y , c a t c h ) + P ( t o o t h a c h e , c a v i t y , ┐ c a t c h ) + P ( t o o t h a c h e , ┐ c a v i t y , ┐ c a t c h ) ] , P ( ┐ t o o t h a c h e , c a v i t y , c a t c h ) + P ( ┐ t o o t h a c h e , ┐ c a v i t y , c a t c h ) + P ( ┐ t o o t h a c h e , c a v i t y , ┐ c a t c h ) + P ( ┐ t o o t h a c h e , ┐ c a v i t y , ┐ c a t c h ) ] > P(Toothache|Cavity)\\ =\alpha P(Toothache,Cavity,Catch)\\ =\alpha < [P(toothache,cavity,catch)+P(toothache,\urcorner cavity,catch)+P(toothache,cavity,\urcorner catch)+P(toothache,\urcorner cavity,\urcorner catch)], \\ P(\urcorner toothache,cavity,catch)+P(\urcorner toothache,\urcorner cavity,catch)+P(\urcorner toothache,cavity,\urcorner catch)+P(\urcorner toothache,\urcorner cavity,\urcorner catch)] > P(Toothache∣Cavity)=αP(Toothache,Cavity,Catch)=α<[P(toothache,cavity,catch)+P(toothache,┐cavity,catch)+P(toothache,cavity,┐catch)+P(toothache,┐cavity,┐catch)],P(┐toothache,cavity,catch)+P(┐toothache,┐cavity,catch)+P(┐toothache,cavity,┐catch)+P(┐toothache,┐cavity,┐catch)]>
P ( C a v i t y ∣ t o o t h a c h e ∨ c a t c h ) = α P ( C a v i t y , t o o t h a c h e ∨ c a t c h ) = α < ( 0.108 + 0.012 + 0.072 ) , ( 0.016 + 0.064 + 0.144 ) > = < 0.4615 , 0.5384 > P(Cavity|toothache \vee catch)\\ =\alpha P(Cavity,toothache \vee catch)\\ =\alpha <(0.108+0.012+0.072), (0.016+0.064+0.144)>\\ =<0.4615,0.5384> P(Cavity∣toothache∨catch)=αP(Cavity,toothache∨catch)=α<(0.108+0.012+0.072),(0.016+0.064+0.144)>=<0.4615,0.5384>