决策树Python实现

Python实现一

在这里我们先调用sklearn算法包中的接口，看一下算法的效果。

实验数据（可能你并不陌生~~~）：

1.5 50 thin
1.5 60 fat
1.6 40 thin
1.6 60 fat
1.7 60 thin
1.7 80 fat
1.8 60 thin
1.8 90 fat
1.9 70 thin
1.9 80 fat
1.9 92 fat
1.6 50 fat
1.65 50 thin
1.68 48 thin
1.7 60 thin
1.7 50 thin
1.7 65 fat

算法调用：

# _*_ encoding:utf-8 _*_
import numpy as np
import scipy as sp
from sklearn import tree
from sklearn.metrics import precision_recall_curve
from sklearn.metrics import classification_report
from sklearn.cross_validation import train_test_split

#####载入数据######
data=[]
labels=[]
f=open('./data/1.txt')
for line in f:
#print line
tokens=line.strip().split(' ')
#print tokens
data.append([float(tk) for tk in tokens[:-1] ])
#print data
labels.append(tokens[-1])
#print labels

data=np.array(data)
# print data
# print data.shape#(10,2)

labels=np.array(labels)
# print labels
# print labels.shape#(10,)

#########转换标签为0/1###########
label=np.zeros(labels.shape)
label[labels=='fat']=1
# print label

##########拆分成训练数据集合测试数据集############
data_train,data_test,label_train,label_test=train_test_split(data,label,test_size=0.2,random_state=0)

####使用信息熵作为划分标准，对决策树进行训练#######
clf=tree.DecisionTreeClassifier(criterion='entropy')
'''
criterion : string, optional (default="gini")
The function to measure the quality of a split. Supported criteria are
"gini" for the Gini impurity and "entropy" for the information gain.
'''
print clf
clf.fit(data_train,label_train)

##########把决策树结构写入文件###########
f=open('tree.dot','w')
f=tree.export_graphviz(clf,out_file=f)

####系数反映每个特征的影响力，越大表示该特征在分类中起到的作用越大####
print clf.feature_importances_

#########测试结果的打印##########
answer=clf.predict(data_test)
print data_test
print  answer
print label_test
print np.mean(answer==label_test)

#############准确率与召回率##########
precision,recall,threshold=precision_recall_curve(label_test,clf.predict(data_test))
# print precision
# print recall
# print threshold
answer=clf.predict(data)#将所有的数据进行测试
print label
print answer
print classification_report(label,answer,target_names=['thin','fat'])

Python 实现二

以下使用Python实现的比较简单的ID3算法，整个过程还是比较好理解的，在该代码中没有考虑标签和属性值缺失的情况，目的只是为了大家更好的理解算法的基本思想。

# -*- encoding:utf-8 -*-
import operator
import math

#构造数据集
def createDataSet():
dataSet=[[1,1,'yes'],[1,1,'yes'],[1,0,'no'],[0,1,'no'],[0,1,'no']]
features = ['A','B']
return dataSet,features

#构造决策树
def treeGrowth(dataSet,features):
classList=[example[-1] for example in dataSet]
if classList.count(classList[0])==len(classList):
return classList[0]
if len(dataSet[0])==1:#没有更多的特征
return classify(classList)

bestFeat=findBestSplit(dataSet)#bestFeat is the index of best feature
bestFeatLabel=features[bestFeat]
myTree={bestFeatLabel:{}}
featValues=[example[bestFeat] for example in dataSet]#选择出的最好的属性的取值
uniqueFeatValues=set(featValues)
del(features[bestFeat])#删除掉该属性
for values in uniqueFeatValues:
subDataSet=splitDataSet(dataSet,bestFeat,values)
myTree[bestFeatLabel][values]=treeGrowth(subDataSet,features)
return myTree

#当没有多余的feature，并且剩下的样本不完全是一样的类别时，采用多数取决
def classify(classList):
classCount={}
for vote in classList:
if vote not in classCount.keys():
classCount[vote]=0
classCount[vote]+=1
sortedClassCount=sorted(classCount.iteritems(),key=operator.itemgetter(1),reverse=True)
return sortedClassCount[0][0]

#寻找用于分裂的最佳属性（遍历所有属性，计算信息增益，最大的为最好的分裂属性）
def findBestSplit(dataSet):
numFeatures=len(dataSet[0])-1
baseEntropy=calcShannonEnt(dataSet)
bestInfoGain=0.0
bestFeat=-1
for i in range(numFeatures):
featValues=[example[i] for example in dataSet]
uniqueFeatValues=set(featValues)
newEntropy=0.0
for val in uniqueFeatValues:
subDataSet=splitDataSet(dataSet,i,val)
prob=len(subDataSet)/float(len(dataSet))
newEntropy+=prob*calcShannonEnt(subDataSet)
if (baseEntropy-newEntropy)>bestInfoGain:
bestInfoGain=baseEntropy-newEntropy
bestFeat=i
return bestFeat

#选择完分裂属性以后，就进行数据集的分裂
def splitDataSet(dataSet,feat,values):
retDataSet=[]
for featVec in dataSet:
if featVec[feat]==values:
reduceFeatVec=featVec[:feat]
reduceFeatVec.extend(featVec[feat+1:])
retDataSet.append(reduceFeatVec)
return retDataSet

#计算数据集的熵
def calcShannonEnt(dataSet):
numEntries=len(dataSet)
labelCounts={}
for featVec in dataSet:
currentLabel=featVec[-1]
if currentLabel not in labelCounts.keys():
labelCounts[currentLabel]=0
labelCounts[currentLabel]+=1
shannoEnt=0.0

for key in labelCounts:
prob=float(labelCounts[key])/numEntries
if prob!=0:
shannoEnt-=prob*math.log(prob,2)
return shannoEnt

#根据上面构造的决策树进行数据分类
def predict(tree,newObject):
while isinstance(tree,dict):
key=tree.keys()[0]
tree=tree[key][newObject[key]]
return tree

if __name__ == '__main__':
dataSet,features=createDataSet()
tree=treeGrowth(dataSet,features)
print tree
print predict(tree,{'A':1,'B':1})

实验结果为:

最终构造的决策树为：

{'A': {0: 'no', 1: {'B': {0: 'no', 1: 'yes'}}}}

测试结果为：

yes

Python 实现三

以下是用Python实现的二叉决策树（分支只能是两个）方法，在该算法中你可以选择使用信息熵或者选择使用基尼指数来建立决策树，并且实现了可视化。

# -*- encoding:utf-8 -*-

from math import log
from PIL import Image,ImageDraw
import zlib

my_data=[['slashdot','USA','yes',18,'None'],
['google','France','yes',23,'Premium'],
['digg','USA','yes',24,'Basic'],
['kiwitobes','France','yes',23,'Basic'],
['google','UK','no',21,'Premium'],
['(direct)','New Zealand','no',12,'None'],
['(direct)','UK','no',21,'Basic'],
['google','USA','no',24,'Premium'],
['slashdot','France','yes',19,'None'],
['digg','USA','no',18,'None'],
['google','UK','no',18,'None'],
['kiwitobes','UK','no',19,'None'],
['digg','New Zealand','yes',12,'Basic'],
['slashdot','UK','no',21,'None'],
['google','UK','yes',18,'Basic'],
['kiwitobes','France','yes',19,'Basic']]

#创建决策节点
class decidenode():
def __init__(self,col=-1,value=None,result=None,tb=None,fb=None):
self.col=col         #待检验的判断条件所对应的列索引值
self.value=value     #为了使结果为true，当前列要匹配的值
self.result=result   #叶子节点的值
self.tb=tb           #true下的节点
self.fb=fb           #false下的节点

#对数值型和离散型数据进行分类
def DivideSet(rows,column,value):
splitfunction=None
if isinstance(value,int) or isinstance(value,float):
splitfunction=lambda x :x>=value
else:
splitfunction=lambda x :x==value

set1=[row for row in rows if splitfunction(row[column])]
set2=[row for row in rows if not splitfunction(row[column])]
return (set1,set2)

#计算数据所包含的实例个数
def UniqueCount(rows):
result={}
for row in rows:
r=row[len(row)-1]
result.setdefault(r,0)
result[r]+=1
return result

#计算Gini impurity
def GiniImpurity(rows):
total=len(rows)
counts=UniqueCount(rows)
imp=0
for k1 in counts:
p1=float(counts[k1])/total
for k2 in counts:
if k1==k2: continue
p2=float(counts[k2])/total
imp+=p1*p2
return imp

#计算信息熵Entropy
def entropy(rows):
log2=lambda x:log(x)/log(2)
results=UniqueCount(rows)
# Now calculate the entropy
ent=0.0
for r in results.keys( ):
p=float(results[r])/len(rows)
ent=ent-p*log2(p)
return ent

#计算方差(当输出为连续型的时候，用方差来判断分类的好或坏，决策树两边分别是比较大的数和比较小的数)
#可以通过后修剪来合并叶子节点
def variance(rows):
if len(rows)==0:return 0
data=[row[len(rows)-1] for row in rows]
mean=sum(data)/len(data)
variance=sum([(d-mean)**2 for d in data])/len(data)
return variance

###############################################################33
#创建决策树递归
def BuildTree(rows,judge=entropy):
if len(rows)==0:return decidenode()

#初始化值
best_gain=0
best_value=None
best_sets=None
best_col=None
S=judge(rows)

#获得最好的gain
for col in range(len(rows[0])-1):
total_value={}
for row in rows:
total_value[row[col]]=1
for value in total_value.keys():
(set1,set2)=DivideSet(rows,col,value)

#计算信息增益，将最好的保存下来
s1=float(len(set1))/len(rows)
s2=float(len(set2))/len(rows)
gain=S-s1*judge(set1)-s2*judge(set2)
if gain > best_gain:
best_gain=gain
best_value=value
best_col=col
best_sets=(set1,set2)
#创建节点
if best_gain>0:
truebranch=BuildTree(best_sets[0])
falsebranch=BuildTree(best_sets[1])
return decidenode(col=best_col,value=best_value,tb=truebranch,fb=falsebranch)
else:
return decidenode(result=UniqueCount(rows))

#打印文本形式的tree
def PrintTree(tree,indent=''):
if tree.result!=None:
print str(tree.result)
else:
print '%s:%s?' % (tree.col,tree.value)
print indent,'T->',
PrintTree(tree.tb,indent+'  ')
print indent,'F->',
PrintTree(tree.fb,indent+'  ')

def getwidth(tree):
if tree.tb==None and tree.fb==None: return 1
return getwidth(tree.tb)+getwidth(tree.fb)

def getdepth(tree):
if tree.tb==None and tree.fb==None: return 0
return max(getdepth(tree.tb),getdepth(tree.fb))+1

#打印图表形式的tree
def drawtree(tree,jpeg='tree.jpg'):
w=getwidth(tree)*100
h=getdepth(tree)*100+120
img=Image.new('RGB',(w,h),(255,255,255))
draw=ImageDraw.Draw(img)
drawnode(draw,tree,w/2,20)
img.save(jpeg,'JPEG')

def drawnode(draw,tree,x,y):
if tree.result==None:
# Get the width of each branch
w1=getwidth(tree.fb)*100
w2=getwidth(tree.tb)*100
# Determine the total space required by this node
left=x-(w1+w2)/2
right=x+(w1+w2)/2
# Draw the condition string
draw.text((x-20,y-10),str(tree.col)+':'+str(tree.value),(0,0,0))
# Draw links to the branches
draw.line((x,y,left+w1/2,y+100),fill=(255,0,0))
draw.line((x,y,right-w2/2,y+100),fill=(255,0,0))
# Draw the branch nodes
drawnode(draw,tree.fb,left+w1/2,y+100)
drawnode(draw,tree.tb,right-w2/2,y+100)
else:
txt=' \n'.join(['%s:%d'%v for v in tree.result.items( )])
draw.text((x-20,y),txt,(0,0,0))

#对新实例进行查询
def classify(observation,tree):
if tree.result!=None: return tree.result
else:
v=observation[tree.col]
branch=None
if isinstance(v,int) or isinstance(v,float):
if v>=tree.value:
branch=tree.tb
else:
branch=tree.fb
else:
if v==tree.value:
branch=tree.tb
else:
branch=tree.fb
return classify(observation,branch)

#后剪枝,设定一个阈值mingain来后剪枝，当合并后熵增加的值小于原来的值，就合并
def prune(tree,mingain):
if tree.tb.result==None:
prune(tree.tb,mingain)
if tree.fb.result==None:
prune(tree.fb,mingain)

if tree.tb.result!=None and tree.fb.result!=None:
tb1,fb1=[],[]
for v,c in tree.tb.result.items():
tb1+=[[v]]*c    #这里是为了跟row保持一样的格式，因为UniqueCount就是对这种进行的计算

for v,c in tree.fb.result.items():
fb1+=[[v]]*c

delta=entropy(tb1+fb1)-(entropy(tb1)+entropy(fb1)/2)
if delta<mingain:
tree.tb,tree.fb=None,None
tree.result=UniqueCount(tb1+fb1)

#对缺失属性的数据进行查询
def mdclassify(observation,tree):
if tree.result!=None:
return tree.result

if observation[tree.col]==None:
tb,fb=mdclassify(observation,tree.tb),mdclassify(observation,tree.fb)        #这里的tr跟fr实际是这个函数返回的字典
tbcount=sum(tb.values())
fbcount=sum(fb.values())
tw=float(tbcount)/(tbcount+fbcount)
fw=float(fbcount)/(tbcount+fbcount)

result={}
for k,v in tb.items():
result.setdefault(k,0)
result[k]=v*tw
for k,v in fb.items():
result.setdefault(k,0)
result[k]=v*fw
return result

else:
v=observation[tree.col]
branch=None
if isinstance(v,int) or isinstance(v,float):
if v>=tree.value:
branch=tree.tb
else:
branch=tree.fb
else:
if v==tree.value:
branch=tree.tb
else:
branch=tree.fb
return mdclassify(observation,branch)

def main():                  #以下内容为我测试决策树的代码
a=BuildTree(my_data)
print "生成的决策树为：\n"

PrintTree(a)
print '----'*15
drawtree(a,jpeg='treeview.jpg')
print "利用阈值0.1进行剪枝后，得到：\n"
prune(a,0.1)
PrintTree(a)
print '-----'*15
prune(a,1)
print "利用阈值1进行剪枝后，得到：\n"
PrintTree(a)
print '----'*15
mdclassify(['google','France',None,None],a)
print mdclassify(['google','France',None,None],a)

mdclassify(['google',None,'yes',None],a)
print mdclassify(['google',None,'yes',None],a)

if __name__=='__main__':
main()

实验结果为：



《完》

所谓的不平凡就是平凡的N次幂。
-------By Ada