[Python] 决策树

《机器学习实战》第三章 决策树

-------------------------------------

#1 trees.py　　计算给定数据集的香农熵

-------------------------------------

 1 from math import log
 2
 3 # 计算给定数据集的香农熵
 4 def calcShannonEnt(dataSet):
 5     numEnres = len(dataSet)
 6     labelCoounts = {}
 7     for featVec in dataSet:
 8         #为所有可能分类创建字典
 9         currentLabel = featVec[-1]
10         if currentLabel not in labelCoounts.keys():
11             labelCoounts[currentLabel] = 0
12         labelCoounts[currentLabel] += 1
13     shannonEnt = 0.0
14     for key in labelCoounts:
15         prob = float(labelCoounts[key]) / numEnres
16         shannonEnt -= prob * log(prob, 2)               #以2为底求对数
17     return shannonEnt
18
19 #用来 得到简单鱼类鉴定数据集
20 def createDataSet():
21     dataSet = [[1, 1, ‘yes‘],
22                [1, 1, ‘yes‘],
23                [1, 0, ‘no‘],
24                [0, 1, ‘no‘],
25                [0, 1, ‘no‘]]
26     labels = [‘no surfacing‘, ‘flippers‘]
27     return dataSet, labels

-------------------------------------

#2 trees.py　　划分数据集 待划分的数据集、划分数据集的待征、需要返回的特征的值

-------------------------------------

1 # 划分数据集   待划分的数据集、划分数据集的待征、需要返回的特征的值
2 def splitDataSet(dataSet, axis, value):
3     retDataSet = []
4     for featVec in dataSet:
5         if featVec[axis] == value:
6             reducedFeatVec = featVec[:axis]
7             reducedFeatVec.extend(featVec[axis + 1:])
8             retDataSet.append(reducedFeatVec)
9     return retDataSet

-------------------------------------

#3 trees.py　　选择最好的数据集划分方式

-------------------------------------

 1 # 划分数据集   待划分的数据集、划分数据集的待征、需要返回的特征的值
 2 def splitDataSet(dataSet, axis, value):
 3     retDataSet = []
 4     for featVec in dataSet:
 5         if featVec[axis] == value:
 6             reducedFeatVec = featVec[:axis]
 7             reducedFeatVec.extend(featVec[axis + 1:])
 8             retDataSet.append(reducedFeatVec)
 9     return retDataSet
10
11
12 # 选择最好的数据集划分方式
13 def chooseBestFeatureToSplit(dataSet):
14     numFeatures = len(dataSet[0]) - 1
15     baseEntropy = calcShannonEnt(dataSet)
16     bestInfoGain = 0.0;
17     bestFeature = -1;
18     for i in range(numFeatures):
19         featList = [example[i] for example in dataSet]
20         uniqueVals = set(featList)
21         newEntropy = 0.0;
22
23         for value in uniqueVals:
24             subDataSet = splitDataSet(dataSet, i, value)
25             prob = len(subDataSet) / float(len(dataSet))
26             newEntropy += prob * calcShannonEnt(subDataSet)
27
28         infoGain = baseEntropy - newEntropy
29
30         if (infoGain > bestInfoGain):
31             bestInfoGain = infoGain
32             bestFeature = i
33
34     return bestFeature

-------------------------------------

#4 trees.py　　创建树的函数代码   两个参数：数据集、标签列表

-------------------------------------

 1 import operator
 2
 3 # 创建树的函数代码 两个参数：数据集、标签列表
 4 def createTree(dataSet, labels):
 5     classList = [example[-1] for example in dataSet]
 6
 7     # 类别完全相同则停止继续划分
 8     if classList.count(classList[0]) == len(classList):
 9         return classList[0]
10
11     # 遍历完所有特征时返回出现次数最多的
12     if len(dataSet[0]) == 1:
13         return majorityCnt(classList)
14
15     bestFeat = chooseBestFeatureToSplit(dataSet)
16     bestFeatLabel = labels[bestFeat]
17     myTree = {bestFeatLabel: {}}
18     del (labels[bestFeat])
19
20     # 得到列表包含的所有属性值
21     featValues = [example[bestFeat] for example in dataSet]
22     uniqueVals = set(featValues)
23
24     # 遍历当前选择特征包含的所有属性值，在每个数据集划分上递归调用函数createTree()
25     for value in uniqueVals:
26         subLabels = labels[:]
27         myTree[bestFeatLabel][value] = createTree(splitDataSet(dataSet, bestFeat, value), subLabels)
28
29     return myTree