概述
Target output is discrete(i.e. binary, or multiple classes)...
What decision trees:
The expression is:
How to learn? Learn from the table below.
The ID3 Algorithm
At each node:
select the feature that results in the largest expected reduction in entorpy for the target label.
select the feature with largest information gain.
D = THE TRAINING DATA
T = the random variable corresponding to play tennis.
p(T = yes) = 9/14
p(T = no) = 5/14
The definition of entropy:
From:(http://zhidao.baidu.com/question/454628)
假设在不改变宏观物质的表现的情况下,微观分子的可能排列的总数为W, 则熵S=lgW
譬如,一个很乱的书架,共有3层,上面随机放着十本书。那么,W=10^3,S=lgW=3
实际上熵是一个表明系统混沌程度的状态量.
From: (http://tieba.baidu.com/f?kz=66299285)
只有当你所使用的那个特定系统中的能量密度参差不齐的时候,能量才能够转化为功,这时,能量倾向于从密度较高的地方流向密度较低的地方,直到一切都达到均匀为止。正是依靠能量的这种流动,你才能从能量得到功。 |
In this case, the definition of entropy is:
For example:
The ID3
Algorithm (Training data D, Features F):
if all examples in D have the same label:
return a leaf node with that label
let x belogn to F be the feature with the largest information gain
let T be a tree root lableled with feature X
let D1, D2,...Dk be the partition produced by splitting D on feature X
for each Di belong to {D1,D2,...Dk}
let Ti = ID3(Di, F-{X})
add Ti as a new branch of T
return T
How to erazor the tree?(reduced error pruning)
Will follow up on next Tuesday(May 11)
Refereneces:
[1] http://ir.hit.edu.cn/qinbing_01.ppt[2]http://zhidao.baidu.com/question/454628
[3]http://tieba.baidu.com/f?kz=66299285
转载于:https://www.cnblogs.com/Qing_Zhu/archive/2010/05/07/1729366.html
最后
以上就是动听咖啡为你收集整理的Decision Tree learning的全部内容,希望文章能够帮你解决Decision Tree learning所遇到的程序开发问题。
如果觉得靠谱客网站的内容还不错,欢迎将靠谱客网站推荐给程序员好友。
发表评论 取消回复