标签:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 | #基于最后一列的分类标签,计算给定数据集的香农熵 def calcShannonEnt(dataset): num_of_entries = len (dataset) label_counts = {} for feat_vec in dataset: current_lebel = feat_vec[ - 1 ] if current_lebel not in label_counts.keys(): label_counts[current_lebel] = 0 label_counts[current_lebel] + = 1 shannonEnt = 0.0 for value in label_counts.values(): prob = float (value) / num_of_entries shannonEnt - = prob * log(prob, 2 ) return shannonEnt |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 | # ================================= # 按照给定特征划分数据集 # 输入:dataset数据集; # axis指定特征,用下标表示; # value需要返回的特征的值 # 返回:数据集中特征值等于value的子集 # ================================= def splitDataset(dataset, axis, value): retDataset = [] for featVec in dataset: if featVec[axis] = = value: reducedFeatVec = featVec[ 0 :axis] reducedFeatVec.extend(featVec[axis + 1 :]) retDataset.append(reducedFeatVec) return retDataset |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 | # =============================================== # 输入: # dataSet: 数据集 # 输出: # bestFeature: 和原数据集熵差最大划分对应的特征的列号 # =============================================== def chooseBestFeatureToSplit(dataSet): # 最后一列用于标签,剩下的才是特征 numFeatures = len (dataSet[ 0 ]) - 1 # 根据标签计算的熵 baseEntropy = calcShannonEnt(dataSet) bestInfoGain = 0.0 ; bestFeature = - 1 # iterate over all the features for i in range (numFeatures): # 取出某个特征列的所有值 featList = [example[i] for example in dataSet] # 去重 uniqueVals = set (featList) newEntropy = 0.0 for value in uniqueVals: subDataSet = splitDataset(dataSet, i, value) prob = len (subDataSet) / float ( len (dataSet)) newEntropy + = prob * calcShannonEnt(subDataSet) # calculate the info gain,计算信息增益 infoGain = baseEntropy - newEntropy # 和目前最佳信息增益比较,如果更大则替换掉 if (infoGain > bestInfoGain): bestInfoGain = infoGain bestFeature = i # 返回代表某个特征的下标 return bestFeature |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 | #用于生成数据集,测试计算熵的函数 def testDataset(): dataset1 = [[ 1 , 1 , ‘yes‘ ], [ 1 , 1 , ‘yes‘ ], [ 1 , 0 , ‘no‘ ], [ 0 , 1 , ‘no‘ ], [ 0 , 1 , ‘no‘ ]] labels = [ ‘no surfacing‘ , ‘flippers‘ ] return dataset1, labels # 用于测试的函数 def test(): mydata, labels = testDataset() print chooseBestFeatureToSplit(mydata) |
1 2 3 4 5 6 7 8 9 10 11 12 | # 传入分类名称组成的列表,返回出现次数最多的分类名称 import operator def majorityCnt(class_list): classCount = {} for vote in class_list: if vote not in classCount: classCount[vote] = 0 classCount[vote] + = 1 sorted_class_list = sorted (classCount.iteritems(), key = operator.itemgetter( 1 ), reverse = True ) return sorted_class_list[ 0 ][ 0 ] |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 | # =============================================== # 本函数用于创建决策树 # 输入: # dataSet: 数据集 # labels: 划分特征标签集 # 输出: # myTree: 生成的决策树 # =============================================== def createTree(dataSet, labels): # 获得类别标签列表 classList = [example[ - 1 ] for example in dataSet] # 递归终止条件一:如果数据集内所有分类一致 if classList.count(classList[ 0 ]) = = len (classList): return classList[ 0 ] # 递归终止条件二:如果所有特征都划分完毕,任然不能将数据集划分成仅仅包含唯一类别的分组 if len (dataSet[ 0 ]) = = 1 : # 只剩下一列为类别列 return majorityCnt(classList) # 返回出现次数最多的类别 # 选择最佳划分特征,返回的时候特征的下标 best_feature = chooseBestFeatureToSplit(dataSet) best_feat_label = labels[best_feature] # 创建空树 myTree = {best_feat_label:{}} # 删除划分后的特征标签 del (labels[best_feature]) # 获取最佳划分特征中全部的特征值 featValues = [example[best_feature] for example in dataSet] # 去重 uniqueVals = set (featValues) for value in uniqueVals: subLabels = labels[:] # 保存用于下一次递归 myTree[best_feat_label][value] = createTree(splitDataset(dataSet, best_feature, value), subLabels) return myTree |
1 2 3 4 5 6 7 8 9 10 11 12 | # 把传入的树序列化之后存入文件 def storeTree(inputTree, filename): import pickle # 用于序列化的模块 fw = open (filename, ‘w‘ ) pickle.dump(inputTree, fw) fw.close() # 从文件中把存好的树反序列化出来 def grabTree(filename): import pickle fr = open (filename) return pickle.load(filename) |
标签:
原文地址:http://www.cnblogs.com/mooba/p/5413980.html