import operator
import csv
import numpy as np
def readDataset(filename):
'''
读取数据
:param filename: 数据文件名，CSV格式
:return: 以列表形式返回数据列表和特征列表
'''
with open(filename) as f:
reader = csv.reader(f)
header_row = next(reader)
labels = header_row[1:7]
dataset = []
for line in reader:
tempVect = line[1:]
dataset.append(tempVect)
trainIndex = [1, 2, 3, 6, 7, 10, 14, 15, 16, 17]
trainDataset = []
testDataset = []
for i in range(1, 18):
if (i in trainIndex):
trainDataset.append(dataset[i - 1])
else:
testDataset.append(dataset[i - 1])
trainDataset.append(dataset[3]) # 为保持和书中结果相同，训练集中增加第四条数据
return dataset, labels, trainDataset, testDataset
def Gini(dataset):
'''
计算gini基尼值
:param dataset: 输入数据集
:return: 返回基尼值gini
'''
numdata = len(dataset)
labels = {}
for featVec in dataset:
label = featVec[-1]
if label not in labels.keys():
labels[label] = 0
labels[label] += 1
gini = 1
for lab in labels.keys():
prop = float(labels[lab]) / numdata
gini -= prop ** 2
return gini
def splitDataset(dataset, axis, value):
'''
对某个特征进行划分后的数据集
:param dataset: 数据集
:param axis: 划分属性的下标
:param value: 划分属性值
:return: 返回剩余数据集
'''
restDataset = []
for featVec in dataset:
if featVec[axis] == value:
restFeatVec = featVec[:axis]
restFeatVec.extend(featVec[axis + 1:])
restDataset.append(restFeatVec)
return restDataset
def bestFeatureSplit(dataset):
'''
最优属性划分
:param dataset: 输入需要划分的数据集
:return: 返回最优划分属性的下标
'''
numFeature = len(dataset[0]) - 1
bestGiniIndex = 10000
bestFeature = -1
for i in range(numFeature):
featList = [example[i] for example in dataset]
uniqueValue = set(featList)
giniIndex = 0
for value in uniqueValue:
subDataset = splitDataset(dataset, i, value)
prop = len(subDataset) / float(len(dataset))
giniIndex += prop * Gini(subDataset)
if (giniIndex < bestGiniIndex):
bestGiniIndex = giniIndex
bestFeature = i
return bestFeature
def majorClass(classList):
'''
对叶节点的分类结果进行划分，投票原则
:param classList: 叶节点上的样本数量
:return: 返回叶节点划分结果
'''
classCount = {}
for vote in classList:
if vote not in classCount.keys():
classCount[vote] = 0
classCount[vote] += 1
sortedClassCount = sorted(classCount.items(), key=operator.itemgetter(1), reverse=True) # 返回数组
return sortedClassCount[0][0]
def decideTreePredict(decideTree, testData, labelsFull):
'''
决策树对测试数据进行结果预测
:param decideTree: 决策树模型
:param testData: 测试数据
:param labelsFull: 特征列表
:return: 返回预测结果
'''
firstFeat = list(decideTree.keys())[0]
secDict = decideTree[firstFeat]
featIndex = labelsFull.index(firstFeat)
classLabel = None
for value in secDict.keys():
if testData[featIndex] == value:
if type(secDict[value]).__name__ == 'dict':
classLabel = decideTreePredict(secDict[value], testData, labelsFull)
else:
classLabel = secDict[value]
return classLabel
def prevReduceBranch(bestFeatLabel, trainDataset, testDataset, labelsFull):
classList = [example[-1] for example in trainDataset]
bestFeatIndex = labelsFull.index(bestFeatLabel)
trainDataValues = [example[bestFeatIndex] for example in trainDataset]
uniqueValues = set(trainDataValues)
error = 0
for value in uniqueValues:
partClassList = [classList[i] for i in range(len(classList)) if trainDataValues[i] == value]
major = majorClass(partClassList)
for data in testDataset:
if data[bestFeatIndex] == value and data[-1] != major:
error += 1
# print('预剪枝继续展开错误数：' + str(error))
return error
def majorTest(major, testData):
error = 0
for i in range(len(testData)):
if major != testData[i][-1]:
error += 1
# print('当前节点为结节点错误数： ' + str(error))
return error
def postReduceBranch(subTree, testData, labelsFull):
error = 0
for i in range(len(testData)):
if decideTreePredict(subTree, testData[i], labelsFull) != testData[i][-1]:
error += 1
# print('后剪枝保留子树错误数： ' + str(error))
return error
def createTree(trainDataset, labels, datasetFull, labelsFull, testDataset):
'''
递归创建决策树
:param dataset: 数据集列表
:param labels: 标签集列表
:param datasetFull: 数据集列表，再传一次
:param labelsFull: 标签集列表，再传一次
:param testData: 测试数据集列表
:return: 返回决策树字典
'''
classList = [example[-1] for example in trainDataset]
if classList.count(classList[0]) == len(classList):
return classList[0]
if len(dataset[0]) == 1:
return (majorClass(classList))
bestFeat = bestFeatureSplit(trainDataset)
bestFeatLabel = labels[bestFeat]
# 预剪枝
# if prevReduceBranch(bestFeatLabel, trainDataset, testDataset, labelsFull) < majorTest(
# majorClass(classList),
# testDataset):
# myTree = {bestFeatLabel: {}}
# else:
# return majorClass(classList)
myTree = {bestFeatLabel: {}}
del (labels[bestFeat])
featValues = [example[bestFeat] for example in trainDataset]
uniqueVal = set(featValues)
# 创建所有属性标签的所有值，以防漏掉某些取值,例如西瓜数据集2.0中的 色泽：浅白
bestFeatIndex = labelsFull.index(bestFeatLabel)
featValuesFull = [example[bestFeatIndex] for example in datasetFull]
uniqueValFull = set(featValuesFull)
if uniqueVal == uniqueValFull:
for value in uniqueVal:
subLabels = labels[:] # 递归回退过程需要继续使用标签，所以前行过程标签副本
myTree[bestFeatLabel][value] = createTree(splitDataset(trainDataset,
bestFeat, value),subLabels, datasetFull,labelsFull,splitDataset(testDataset,
bestFeat, value))
else:
for value in uniqueVal:
subLabels = labels[:] # 递归回退过程需要继续使用标签，所以前行过程标签副本
myTree[bestFeatLabel][value] = createTree(splitDataset(trainDataset,
bestFeat, value),subLabels, datasetFull,labelsFull,splitDataset(testDataset,
bestFeat, value))
uniqueValFull.remove(value)
for value in uniqueValFull:
myTree[bestFeatLabel][value] = majorClass(classList)
return myTree
# 后剪枝
# print(myTree)
# if postReduceBranch(myTree, testDataset, labelsFull) <=
majorTest(majorClass(classList), testDataset):
# return myTree
# else:
# return majorClass(classList)
if __name__ == '__main__':
filename = 'C:\\Users\\14399\\Desktop\\西瓜2.0.csv'
dataset, labels, trainDataset, testDataset = readDataset(filename)
datasetFull = trainDataset[:]
labelsFull = labels[:]
myTree = createTree(trainDataset, labels, datasetFull, labelsFull, testDataset)
print(myTree)
未剪枝：{'脐部': {'凹陷': {'色泽': {'浅白': '否', '青绿': '是', '乌黑': '是'}}, '稍凹': {'根蒂': {'蜷缩': '否', '稍蜷': {'色泽': {'青绿': '是', '乌黑': {'纹理': {'稍糊': '是', '清晰': '否', '模糊': '是'}}, '浅白': '是'}}, '硬挺': '是'}}, '平坦': '否'}}
预剪枝：{'脐部': {'稍凹': '是', '平坦': '否', '凹陷': '是'}}
后剪枝: {'脐部': {'稍凹': {'根蒂': {'蜷缩': '否', '稍蜷': {'色泽': {'乌黑': '是', '青绿': '是', '浅白': '是'}}, '硬挺': '是'}}, '凹陷': '是', '平坦': '否'}}
西瓜2.0数据集：链接：https://pan.baidu.com/s/12aVngexje2RdizgOg1Fr0A 提取码：uywy
参考：https://blog.csdn.net/sysu_cis/article/details/51874229