机器学习之KNN算法
本次随笔包括调用sklearn库函数和手撸KNN算法，代码如下：
算法步骤：
1、为了判断未知实例类别，以已知实例为参考
2、选择参数k（最近的k个已知实例）
3、计算未知实例与所有已知实例的距离
4、根据少数服从多数的方法，让未知实例归类为k个邻近样本中最多的类别
调用库函数：
1 from sklearn import neighbors 2 from sklearn import datasets 3 4 knn = neighbors.KNeighborsClassifier() 5 iris = datasets.load_iris() # 调用库里面的数据 6 7 knn.fit(iris.data,iris.target) 8 9 predictLable = knn.predict([[5,3,5,2]]) 10 print(predictLable)
KNN算法实现：
1 import random 2 import math 3 import operator 4 5 def loadData(filename,split,trainingSet=[],testSet=[]): 6 with open(filename,'r') as file: 7 lines = file.readlines() 8 dataset = [[] for i in range(len(lines)-1)] 9 for i in range(len(dataset)): 10 dataset[i][:] = (item for item in lines[i].strip().split(',')) # 逐行读取数据 11 for x in range(len(dataset)): 12 for y in range(len(dataset[0])-1): 13 dataset[x][y] = float(dataset[x][y]) # 将除最后一列的数据转化为浮点型 14 if random.random() < split: # 将数据集进行划分 15 trainingSet.append(dataset[x]) 16 else: 17 testSet.append(dataset[x]) 18 print("trainingSet",trainingSet) 19 print("testset",testSet) 20 return trainingSet,testSet 21 22 23 def conclu_distance(a,b,length): # 计算两个向量的欧氏距离 24 dis = 0 25 for i in range(length): 26 dis += pow(a[i]-b[i],2) 27 return math.sqrt(dis) 28 29 30 def getNeibors(trainingSet,testinstance,k): # 参数分别为：训练集，待计算邻居的向量，邻居数量k。 获取k个最近邻居的类别值 31 distance = [] 32 length = (len(testinstance)-1) 33 for x in range(len(trainingSet)): 34 dist = conclu_distance(testinstance,trainingSet[x],length) # 注意这里数据的添加格式 35 distance.append((trainingSet[x], dist)) 36 distance.sort(key=operator.itemgetter(1)) 37 neighbers = [] 38 for x in range(k): 39 neighbers.append(distance[x][0]) 40 return neighbers 41 42 43 def getResponse(neighbors): # 从得到的邻居中计算得到测试结果 44 classVotes = {} # 投票通常采用字典形式 45 for x in range(len(neighbors)): 46 response = neighbors[x][-1] 47 if response in classVotes: 48 classVotes[response]+=1 49 else: 50 classVotes[response] = 1 51 sortedVotes = sorted(classVotes,key=operator.itemgetter(1),reverse=True) 52 return sortedVotes[0] 53 54 55 def getAccuracy(testSet,predictions): # 计算测试值准确率 56 correct = 0 57 for x in range(len(testSet)): 58 if testSet[x][-1] == predictions[x]: 59 correct += 1 60 return (correct/float(len(testSet)))*100 61 62 def main(): 63 trainingSet = [] 64 testSet = [] 65 split = 0.67 66 loadData('irisdata.txt',split,trainingSet,testSet) 67 predictions = [] 68 k =3 69 for x in range(len(testSet)): 70 neighbers = getNeibors(trainingSet,testSet[x],k) 71 result = getResponse(neighbers) 72 predictions.append(result) 73 print('>predicted',repr(result),',actual=',repr(testSet[x][-1])) 74 accuracy = getAccuracy(testSet,predictions) 75 print('准确率为：',repr(accuracy),'%') 76 77 main()
