今天,我们介绍机器学习里比较常用的一种分类算法,决策树。决策树是对人类认知识别的一种模拟,给你一堆看似杂乱无章的数据,如何用尽可能少的特征,对这些数据进行有效的分类。
决策树借助了一种层级分类的概念,每一次都选择一个区分性最好的特征进行分类,对于可以直接给出标签 label 的数据,可能最初选择的几个特征就能很好地进行区分,有些数据可能需要更多的特征,所以决策树的深度也就表示了你需要选择的几种特征。
在进行特征选择的时候,常常需要借助信息论的概念,利用最大熵原则。
决策树一般是用来对离散数据进行分类的,对于连续数据,可以事先对其离散化。
在介绍决策树之前,我们先简单的介绍一下信息熵,我们知道,熵的定义为:
我们先构造一些简单的数据:
from sklearn import datasets
import numpy as np
import matplotlib.pyplot as plt
import math
import operator
def Create_data():
dataset = [[1,1,‘yes’],
[1, 1,‘yes’],
[1, 0, ‘no’],
[0, 1, ‘no’],
[0, 1, ‘no’],
[3, 0, ‘maybe’]]
feat_name = [‘no surfacing’, ‘flippers’]
return dataset, feat_name
然后定义一个计算熵的函数:
def Cal_entrpy(dataset):
n_sample = len(dataset)
n_label = {}
for featvec in dataset:
current_label = featvec[-1]
if current_label not in n_label.keys():
n_label[current_label] = 0
n_label[current_label] += 1
shannonEnt = 0.0
for key in n_label:
prob = float(n_label[key]) / n_sample
shannonEnt -= prob * math.log(prob, 2)
return shannonEnt
要注意的是,熵越大,说明数据的类别越分散,越呈现某种无序的状态。
下面再定义一个拆分数据集的函数:
def Split_dataset(dataset, axis, value):
retDataSet = []
for featVec in dataset:
if featVec[axis] == value:
reducedFeatVec = featVec[:axis]
reducedFeatVec.extend(featVec[axis+1 :])
retDataSet.append(reducedFeatVec)
return retDataSet
结合前面的几个函数,我们可以构造一个特征选择的函数:
def Choose_feature(dataset):
num_sample = len(dataset)
num_feature = len(dataset[0]) - 1
baseEntrpy = Cal_entrpy(dataset)
best_Infogain = 0.0
bestFeat = -1
for i in range (num_feature):
featlist = [example[i] for example in dataset]
uniquValus = set(featlist)
newEntrpy = 0.0
for value in uniquValus:
subData = Split_dataset(dataset, i, value)
prob = len(subData) / float(num_sample)
newEntrpy += prob * Cal_entrpy(subData)
info_gain = baseEntrpy - newEntrpy
if (info_gain 》 best_Infogain):
best_Infogain = info_gain
bestFeat = i
return bestFeat
然后再构造一个投票及计票的函数
def Major_cnt(classlist):
class_num = {}
for vote in classlist:
if vote not in class_num.keys():
class_num[vote] = 0
class_num[vote] += 1
Sort_K = sorted(class_num.iteritems(),
key = operator.itemgetter(1), reverse=True)
return Sort_K[0][0]
有了这些,就可以构造我们需要的决策树了:
def Create_tree(dataset, featName):
classlist = [example[-1] for example in dataset]
if classlist.count(classlist[0]) == len(classlist):
return classlist[0]
if len(dataset[0]) == 1:
return Major_cnt(classlist)
bestFeat = Choose_feature(dataset)
bestFeatName = featName[bestFeat]
myTree = {bestFeatName: {}}
del(featName[bestFeat])
featValues = [example[bestFeat] for example in dataset]
uniqueVals = set(featValues)
for value in uniqueVals:
subLabels = featName[:]
myTree[bestFeatName][value] = Create_tree(Split_dataset
(dataset, bestFeat, value), subLabels)
return myTree
def Get_numleafs(myTree):
numLeafs = 0
firstStr = myTree.keys()[0]
secondDict = myTree[firstStr]
for key in secondDict.keys():
if type(secondDict[key]).__name__ == ‘dict’ :
numLeafs += Get_numleafs(secondDict[key])
else:
numLeafs += 1
return numLeafs
def Get_treedepth(myTree):
max_depth = 0
firstStr = myTree.keys()[0]
secondDict = myTree[firstStr]
for key in secondDict.keys():
if type(secondDict[key]).__name__ == ‘dict’ :
this_depth = 1 + Get_treedepth(secondDict[key])
else:
this_depth = 1
if this_depth 》 max_depth:
max_depth = this_depth
return max_depth
评论
查看更多