c***z
发帖数: 6348 | 1 My boss asked me a question, just want to verify that I have the correct
answer and won't lose face before his bosses.:)
If a person visits one websites where the gender ratio is 5:5, and one
websites where the gender ratio is 7:3, and assume that the visits are
independent of each other (the Naive Bayes assumption), then the chance that
the person is male is 70%.
Is that correct? |
g******2
发帖数: 234 | 2 it seems to be P(M|S1)*P(M|S2)/P(M). If you assume P(M) = 0.5, then it's 0.7
. |
c***z
发帖数: 6348 | |
I*****a
发帖数: 5425 | 4 does naive bayes assumption mean that ?
that
【在 c***z 的大作中提到】
: My boss asked me a question, just want to verify that I have the correct
: answer and won't lose face before his bosses.:)
: If a person visits one websites where the gender ratio is 5:5, and one
: websites where the gender ratio is 7:3, and assume that the visits are
: independent of each other (the Naive Bayes assumption), then the chance that
: the person is male is 70%.
: Is that correct?
|
c***z
发帖数: 6348 | 5 yeah, I am pretty confident that is why the method has "Naive" in its name :
) |
j********7
发帖数: 15 | 6 I think Naive Bayes means, like conditioning on y, all xi's are indedendent?
:
【在 c***z 的大作中提到】
: yeah, I am pretty confident that is why the method has "Naive" in its name :
: )
|
c***z
发帖数: 6348 | 7 Oh, yes, you are right
conditionally, not absolutely independent
indedendent?
【在 j********7 的大作中提到】
: I think Naive Bayes means, like conditioning on y, all xi's are indedendent?
:
: :
|
z******n
发帖数: 397 | 8 貌似除了需要P(S1, S2)=P(S1)P(S2)，还需要P(S1, S2|M) = P(S1|M)P(S2|M)
.7
【在 g******2 的大作中提到】
: it seems to be P(M|S1)*P(M|S2)/P(M). If you assume P(M) = 0.5, then it's 0.7
: .
|
j********7
发帖数: 15 | 9 How do you get this formula?
.7
【在 g******2 的大作中提到】
: it seems to be P(M|S1)*P(M|S2)/P(M). If you assume P(M) = 0.5, then it's 0.7
: .
|
I*****a
发帖数: 5425 | 10 and I don't quite understand the question itself...
:
【在 c***z 的大作中提到】
: yeah, I am pretty confident that is why the method has "Naive" in its name :
: )
|
|
|
l*******r
发帖数: 28 | 11 信息不足够，没法给出解。
that
【在 c***z 的大作中提到】
: My boss asked me a question, just want to verify that I have the correct
: answer and won't lose face before his bosses.:)
: If a person visits one websites where the gender ratio is 5:5, and one
: websites where the gender ratio is 7:3, and assume that the visits are
: independent of each other (the Naive Bayes assumption), then the chance that
: the person is male is 70%.
: Is that correct?
|
u**r
发帖数: 160 | |
r*****d
发帖数: 346 | 13 直接套贝叶斯公式，得0.7; 唯一需要的是P(S1, S2|Male) = P(S1|Male)P(S2|Male)
and P(S1, S2|Female) = P(S1|Female)P(S2|Female), which is the 'Naive Bayes
Assumption' you are referring to ：）
businessman/IT guy/stat guy which categories does your boss belong? |
l*******r
发帖数: 28 | 14 P(S1|Male)=?
【在 r*****d 的大作中提到】
: 直接套贝叶斯公式，得0.7; 唯一需要的是P(S1, S2|Male) = P(S1|Male)P(S2|Male)
: and P(S1, S2|Female) = P(S1|Female)P(S2|Female), which is the 'Naive Bayes
: Assumption' you are referring to ：）
: businessman/IT guy/stat guy which categories does your boss belong?
|
c****t
发帖数: 19049 | 15 这跟s1,s2 independent与否关系不大，假设independent也无所谓。关键是假设了U(s1
,s2)->1。不过搞biz有个0.7也就行了 |
c***z
发帖数: 6348 | 16 Can you explain U(s1,s2)->1 ?
Thanks!
s1
【在 c****t 的大作中提到】
: 这跟s1,s2 independent与否关系不大，假设independent也无所谓。关键是假设了U(s1
: ,s2)->1。不过搞biz有个0.7也就行了
|
c***z
发帖数: 6348 | 17 前者可以推出后者
【在 z******n 的大作中提到】
: 貌似除了需要P(S1, S2)=P(S1)P(S2)，还需要P(S1, S2|M) = P(S1|M)P(S2|M)
:
: .7
|
c***z
发帖数: 6348 | 18 Bayes inference :)
【在 I*****a 的大作中提到】
: and I don't quite understand the question itself...
:
: :
|
c****t
发帖数: 19049 | 19 就是说用这两个网站就可以给你在此命题下接近完备的信息啊
【在 c***z 的大作中提到】
: Can you explain U(s1,s2)->1 ?
: Thanks!
:
: s1
|
l*******r
发帖数: 28 | 20 推不出来吧
【在 c***z 的大作中提到】
: 前者可以推出后者
|
|
|
c***z
发帖数: 6348 | 21 businessman :)
economics major from Yale
【在 r*****d 的大作中提到】
: 直接套贝叶斯公式，得0.7; 唯一需要的是P(S1, S2|Male) = P(S1|Male)P(S2|Male)
: and P(S1, S2|Female) = P(S1|Female)P(S2|Female), which is the 'Naive Bayes
: Assumption' you are referring to ：）
: businessman/IT guy/stat guy which categories does your boss belong?
|
I*****a
发帖数: 5425 | 22 I guess you need a lot more information in order to get what you want,
including P(s1) P(s2) and P(s1 and s2).
Here you don't even have P(s1).
【在 c***z 的大作中提到】
: Bayes inference :)
|
I*****a
发帖数: 5425 | 23 that's not necessary. that condition is too harsh.
In reality P(not s1 or s2) should be large.
s1
【在 c****t 的大作中提到】
: 这跟s1,s2 independent与否关系不大，假设independent也无所谓。关键是假设了U(s1
: ,s2)->1。不过搞biz有个0.7也就行了
|
c***z
发帖数: 6348 | 24 yes, you are right, hehe
I really need to read my textbook again
【在 l*******r 的大作中提到】
: 推不出来吧
|
c***z
发帖数: 6348 | 25 Yeah, we kind of assumed a lot. In business world, we have to. :) |
c****t
发帖数: 19049 | 26 是说“此命题下”
【在 I*****a 的大作中提到】
: that's not necessary. that condition is too harsh.
: In reality P(not s1 or s2) should be large.
:
: s1
|
I*****a
发帖数: 5425 | 27 nor sufficient
【在 c****t 的大作中提到】
: 是说“此命题下”
|
l*******r
发帖数: 28 | 28 如果只知道 P(M|S1=1) = 0.5, P(M|S2=1) = 0.7, and P(S1, S2) = P(S1)P(S2)， 是
没法得到P(M|S1=1, S2=1)
that
【在 c***z 的大作中提到】
: My boss asked me a question, just want to verify that I have the correct
: answer and won't lose face before his bosses.:)
: If a person visits one websites where the gender ratio is 5:5, and one
: websites where the gender ratio is 7:3, and assume that the visits are
: independent of each other (the Naive Bayes assumption), then the chance that
: the person is male is 70%.
: Is that correct?
|
c***z
发帖数: 6348 | 29 加上P(S1, S2|M) = P(S1|M)P(S2|M)
大胆假设小心求证，呵呵
【在 l*******r 的大作中提到】
: 如果只知道 P(M|S1=1) = 0.5, P(M|S2=1) = 0.7, and P(S1, S2) = P(S1)P(S2)， 是
: 没法得到P(M|S1=1, S2=1)
:
: that
|
z******n
发帖数: 397 | 30 从我有限的经验看，这个假设在某些情境下蛮强的。设想S1和S2是对男性吸引力超大的
网站...
【在 c***z 的大作中提到】
: 加上P(S1, S2|M) = P(S1|M)P(S2|M)
: 大胆假设小心求证，呵呵
|
|
|
g****l
发帖数: 213 | 31 能不能这样理解，第一个网站的几率是50/50, 等于没有信息，第二个网站7:3 ， 所以
70% 是男? |
I*****a
发帖数: 5425 | 32 To be accurate, assuming conditionally independent and unconditionally
independent, it is "P(M|S1) = P(M) = 1/2" that makes the 1st website "not
informative" as you said.
【在 g****l 的大作中提到】
: 能不能这样理解，第一个网站的几率是50/50, 等于没有信息，第二个网站7:3 ， 所以
: 70% 是男?
|
I*****a
发帖数: 5425 | 33 2nd this.
This is one sufficient condition given that we know P(M)(=1/2 for example).
【在 z******n 的大作中提到】
: 貌似除了需要P(S1, S2)=P(S1)P(S2)，还需要P(S1, S2|M) = P(S1|M)P(S2|M)
:
: .7
|
