m*****e 发帖数: 207 | 1 this (likelihood principle example ) is a nice one from James Berger's book
"Statistical Decision Theory and Bayesian Analysis, 2nd ed"
as chenhm said, there are a lot of other Bayesian/Frequentist disputes
which he didn't mention.
I happened to be reading Berger's book these days since I've been thinking
about a paradox at http://www.math.washington.edu/~burdzy/Bayes/prisoner.shtml
and http://www.math.washington.edu/~burdzy/Bayes/ethics.shtml It might be a
slap on the faces of Bayesian peopl | c****m 发帖数: 91 | 2 本来只是业余玩票性质的,被你这么一搞只好去那个网站看看了,不过偶基本没
搞明白那个例子是怎么反对使用Bayes inference的,偶倒设想这个问题可以转化
为用game theory来解释,用Berger书上的minimax,也不太sure,等专业人士来教导吧.
What I do care is the principle in inductive/transductive learning (i.e.,
learning how to learn for those CLT people majoring CS): Bayesian is one
way of doing it (in model selection), VC theory and MDL are very attractive
also. 偶不习惯以公理体系那套东东思考概率,觉得有意义的是algorithmic randomness.
When facing a decision problem (e.g., just multiple composite hypothesis
testing,见4536,
【在 m*****e 的大作中提到】 : this (likelihood principle example ) is a nice one from James Berger's book : "Statistical Decision Theory and Bayesian Analysis, 2nd ed" : as chenhm said, there are a lot of other Bayesian/Frequentist disputes : which he didn't mention. : I happened to be reading Berger's book these days since I've been thinking : about a paradox at http://www.math.washington.edu/~burdzy/Bayes/prisoner.shtml : and http://www.math.washington.edu/~burdzy/Bayes/ethics.shtml It might be a : slap on the faces of Bayesian peopl
| m*****e 发帖数: 207 | 3
hoho, 大家都是业余,都是业余。。。你那个GLRT的偶看了,不太熟悉,不过偶
记得总是可以用chisqaure的,去查查google或者参考书吧。
那个paradox偶是这么理解的(加上了偶自己推的一些结果):
有两个瓮,分别都装了一千个球,白色或者黑色。
第一个瓮的球是一个一个装的,每次随机放一个黑球或白球进去,各占1/2的概率
。(所以白球数满足二项分布)。第二个瓮的颜色分布是已知的:490个白球,510
个黑球。
现在让你从第一个瓮里任选999个,然后给你看它们的颜色,假如你看到479个白球。
然后给你两个选择:要么选第一个瓮里剩的那个球,要么从第二个瓮里随便拿一个球。
这个球是白色你就赢(生存),是黑色你就输(死亡)。
同时,这个试验(游戏)还要给其他99个人做,但是互相之间不许交流。另外,很重
要的一点:第一个瓮最开始放满以后就不再拿出来重新sample了(就是说每个游戏者
面对的是同样的第一个瓮)。
问题来了:
1)你是个“被噎死派”(Bayesian),假如你的目的就是你自己赢,那么应该选第一个
瓮的最后一个球,因为你赢的可能性是一半;而从第二个瓮里拿球只有49%的
【在 c****m 的大作中提到】 : 本来只是业余玩票性质的,被你这么一搞只好去那个网站看看了,不过偶基本没 : 搞明白那个例子是怎么反对使用Bayes inference的,偶倒设想这个问题可以转化 : 为用game theory来解释,用Berger书上的minimax,也不太sure,等专业人士来教导吧. : What I do care is the principle in inductive/transductive learning (i.e., : learning how to learn for those CLT people majoring CS): Bayesian is one : way of doing it (in model selection), VC theory and MDL are very attractive : also. 偶不习惯以公理体系那套东东思考概率,觉得有意义的是algorithmic randomness. : When facing a decision problem (e.g., just multiple composite hypothesis : testing,见4536,
|
|