[PGM] Bayes Network and Conditional Independence
2 - 1 - Semantics & Factorization
2 - 2 - Reasoning Patterns
2 - 3 - Flow of Probabilistic Influence
2 - 4 - Conditional Independence
2 - 5 - Independencies in Bayesian Networks
2 - 6 - Naive Bayes
2 - 7 - Application Medical Diagnosis
2 - 8 - Knowledge Engineering Example-SAMIAM
感觉棒棒哒! Professor Daphne Koller
Chain Rule for Bayesian Networks
Calculate joint conditional probability
以下的过程,尤其是概率和为1这一点,对应上图中的某个node的table的一行。
Intercausal reasoning
其实就是先验后验推断的东西。
When can X influence Y?
理性的理解需要建立在一定的贝叶斯统计基础,如下:
Grade --> Difficulty,平均分高,则难度低。
Grade <-- Difficulty,难度低,则平均分高。
Difficulty --> Grade --> Letter,试题难度对Letter如何影响?
Difficulty <-- Grade <-- Letter,试题难度对Letter如何影响?
Grade通过后验作用于Difficulty;
Grade通过先验作用于Letter;
如果Grade确定,先验后验分布各自确定,俩分布当然独立。
如果Grade未确定,Difficulty的先验(Grade)会受到其先验(Letter)的影响。
Grade <-- Intelligence --> SAT,显而易见的影响关系。
作为两个分布的先验,
如果已知,俩分布当然各自独立;
如果未知,其中一个node的数据反推到的后验参数,自然会影响另一个分布。
Difficulty --> Grade <-- Intelligence,共同条件分布,V-structure。
如果已知,数据是由两个参数作用的,当然相互见有影响,例如:
- 参数1小一点,参数2大一点,可能也会是相同的data。
如果未知,两个参数的分布自然没什么关系。
实践:S - I - G - D, S 是否会影响 D?
如果I 已知,block,不能。
如果I 未知,G未知,block,不能。
如果I 未知,G已知,可以。
注:这是快捷的判断方式,针对局部简单的情况。
Conditional Independence
Notice: 思考是否受到了共同先验的影响。
举个例子:两个coin,1)正常的 2)不正常的0.9 vs 0.1
若已知上述信息,那么X1 X2 表示的“上” “下” 面就可直接确定。
若未知上述信息,那么X1 X2 表示的“上” “下” 面,比如:猜测下一次正面的概率,就会受到已知“数据”的影响。
具体地讲:已知两次投掷都是正面,是否更偏向于coin is biased?这边不独立了呢。
D-separation
X与Y是D分离的 given Z。
表示:d-sep(X,Y|Z)
本质就是先验后验相互影响。
好文一读:d-separation: How to determine which variables are independent in a Bayes net
注:这是系统的方法,针对复杂情况。
- “Ancestral graph": this is a reduced version of the original net, 即只考虑长辈。
- "Moralize": 伴侣两两连线。
- “Disorient": 转为无向图。
- “Delete the given and their edges":去除条件部分。
解读结果:
- 不连接,则独立。
- 若连接,不独立。
- If one or both of the variables are missing (because they were givens, and weretherefore deleted), they are independent.
两个经典例子,共赏
附加题
P(D|CEG) =? P(D|C)
Are D and E conditionally independent, given C? AND
Are D and G conditionally independent, given C?
可见,将EG对D的影响,转化为了两个独立问题。俩问题都满足,才是相等。
I-maps
独立图,什么东东? (后续章节有专题)
P满足与图G相关的局部独立性,那么图G是P的一个I-map,P可能有多个I-map。
If P factorizes over G, then G is an I-map for P.
G1 is an I-map for P1.
G2 is an I-map for P1 and P2
- I-map的因子分解
Theorem: If G is an I-map for P, then P factorizes over G.
- 最小I-map
/*...*/
CPCS Network
了解大规模BN的一些问题和重难点。
Ref: https://dslpitt.org/uai/papers/94/p484-pradhan.pdf
Ref: http://www.cs.cmu.edu/afs/cs/project/jair/pub/volume13/cheng00a-html/node15.html
The main network used in our tests is a subset of the CPCS (Computer-based Patient Case Study) model [Pradhan et al.1994], a large multiply-connected multi-layer network consisting of 422 multi-valued nodes and covering a subset of the domain of internal medicine.
Among the 422 nodes,
14 nodes describe diseases, 显眼的特征
33 nodes describe history and risk factors, and 相关指标特征
the remaining 375 nodes describe various findings related to the diseases. 不显眼的特征
To avoid complete table representation, 毕竟没人喜欢处理全连接网。
Knowledge Engineering Example
Samiam Download: http://reasoning.cs.ucla.edu/samiam/index.php?s=
[PGM] Bayes Network and Conditional Independence的更多相关文章
- 条件独立(conditional independence) 结合贝叶斯网络(Bayesian network) 概率有向图 (PRML8.2总结)
本文会利用到上篇,博客的分解定理,需要的可以查找上篇博客 D-separation对任何用有向图表示的概率模型都成立,无论随机变量是离散还是连续,还是两者的结合. 部分图为手写,由于本人字很丑,望见谅 ...
- 条件独立(conditional independence) 结合贝叶斯网络(Bayesian network) 概率有向图 (PRML8.2总结)
转:http://www.cnblogs.com/Dzhouqi/p/3204481.html本文会利用到上篇,博客的分解定理,需要的可以查找上篇博客 D-separation对任何用有向图表示的概率 ...
- [Machine Learning] Probabilistic Graphical Models:二、Bayes Network Fundamentals(1、Semantics & Factorization)
一.How to construct the dependency? 1.首字母即随机变量名称 2.I->G是更加复杂的模型,但Bayes里不考虑,因为Bayes只是无环图. 3.CPD = c ...
- [Bayesian] “我是bayesian我怕谁”系列 - Exact Inferences
要整理这部分内容,一开始我是拒绝的.欣赏贝叶斯的人本就不多,这部分过后恐怕就要成为“从入门到放弃”系列. 但,这部分是基础,不管是Professor Daphne Koller,还是统计学习经典,都有 ...
- 本人AI知识体系导航 - AI menu
Relevant Readable Links Name Interesting topic Comment Edwin Chen 非参贝叶斯 徐亦达老板 Dirichlet Process 学习 ...
- [Bayesian] “我是bayesian我怕谁”系列 - Exact Inference
要整理这部分内容,一开始我是拒绝的.欣赏贝叶斯的人本就不多,这部分过后恐怕就要成为“从入门到放弃”系列. 但,这部分是基础,不管是Professor Daphne Koller,还是统计学习经典,都有 ...
- PGM:概率图模型Graphical Model
http://blog.csdn.net/pipisorry/article/details/51461878 概率图模型Graphical Models简介 完全通过代数计算来对更加复杂的模型进行建 ...
- an introduction to conditional random fields
1.Structured prediction methods are essentially a combination of classification and graphical modeli ...
- 概率图论PGM的D-Separation(D分离)
目录[-] 本文大部分来自:http://www.zhujun.me/d-separation-separation-d.html 一.引言 二.三种情况分析 三.总结 四.应用例子 五.参考资料 其 ...
随机推荐
- 数据库复用代码【c3p0配置文件、数据库连接池】
前言 为了复用,记载一些以前写过的工具类.方法 c3p0配置文件[c3p0-config.xml] <?xml version="1.0" encoding="UT ...
- Jmeter之性能测试基础
1.概念:性能测试是通过自动化的测试工具模拟多种正常峰值及负载条件来对系统的各项性能指标进行测试.负载测试和压力测试都属于性能测试,两者可以结合进行.通过负载测试,确定在各种工作负载下系统的性能,目标 ...
- APUE 3 -- 信号 (signal)<II>: 可靠信号
一个事件可以事一个信号发送给一个进程,这个事件可以是硬件异常,可以是软件条件触发,可以是终端产生信号,也可以是一个kill函数调用.当信号产生后,内核通常会在进程表中设置某种形式的标志(flag).我 ...
- bat文件逐行读取txt
From_Ip='192.138.60.16'@echo offfor /f "tokens=1,2 delims='" %%a in (D:\ETL\bat\config.txt ...
- 静态页面如何实现 include 引入公用代码
一直以来,我司的前端都是用 php 的 include 函数来实现引入 header .footer 这些公用代码的,就像下面这样: <!-- index.php --> <!DOC ...
- [mysql] ERROR 1045 (28000): Access denied for user 'root'@'localhost' (using password: YES).
用mysql -u root -p显示ERROR 1045 (28000): Access denied for user 'root'@'localhost' (using password: YE ...
- Linux入门之常用命令(11) 系统监控 vmstat top
vmstat命令是最常见的Linux/Unix监控工具,可以展现给定时间间隔的服务器的状态值,包括服务器的CPU使用率,内存使用,虚拟内存交换情况,IO读写情况.这个命令是我查看Linux/Unix最 ...
- C++ Primer Plus 6 第二章
// myfirst.cpp--displays a message #include <iostream> // a PREPROCESSOR directive int main() ...
- Eight hdu 1043 八数码问题 双搜
Eight Time Limit: 10000/5000 MS (Java/Others) Memory Limit: 65536/32768 K (Java/Others)Total Subm ...
- hdu4686 Arc of Dream 2013 Multi-University Training Contest 9矩阵快速幂
Arc of Dream Time Limit: 2000/2000 MS (Java/Others) Memory Limit: 65535/65535 K (Java/Others) Tot ...