1.文件
5.0,3.5,1.6,0.6,apple
5.1,3.8,1.9,0.4,apple
4.8,3.0,1.4,0.3,apple
5.1,3.8,1.6,0.2,apple
4.6,3.2,1.4,0.2,apple
5.3,3.7,1.5,0.2,apple
5.0,3.3,1.4,0.2,apple
7.0,3.2,4.7,1.4,orange
6.4,3.2,4.5,1.5,orange
6.9,3.1,4.9,1.5,orange
5.5,2.3,4.0,1.3,orange
6.5,2.8,4.6,1.5,orange
5.7,2.8,4.5,1.3,orange
6.3,3.3,4.7,1.6,orange
7.3,2.9,6.3,1.8,banana
6.7,2.5,5.8,1.8,banana
7.2,3.6,6.1,2.5,banana
6.5,3.2,5.1,2.0,banana
6.4,2.7,5.3,1.9,banana
6.8,3.0,5.5,2.1,banana
5.7,2.5,5.0,2.0,banana
5.8,2.8,5.1,2.4,banana
2 代码
# -*- coding: UTF-8 -*- import math import csv import random import operator ''' @author:hunter @time:2017.03.31 ''' class KNearestNeighbor(object): def __init__(self): pass def loadDataset(self,filename, split, trainingSet, testSet): # 加载数据集 split以某个值为界限分类train和test with open(filename, 'r') as csvfile: lines = csv.reader(csvfile) #读取所有的行 dataset = list(lines) #转化成列表 for x in range(len(dataset)-1): for y in range(4): dataset[x][y] = float(dataset[x][y]) if random.random() < split: # 将所有数据加载到train和test中 生成0和1的随机浮点数 trainingSet.append(dataset[x]) else: testSet.append(dataset[x]) def calculateDistance(self,testdata, traindata, length): # 计算距离 distance = 0 # length表示维度 数据共有几维 for x in range(length): distance += pow((testdata[x]-traindata[x]), 2) return math.sqrt(distance) def getNeighbors(self,trainingSet, testInstance, k): # 返回最近的k个边距 distances = [] length = len(testInstance)-1 for x in range(len(trainingSet)): #对训练集的每一个数计算其到测试集的实际距离 dist = self.calculateDistance(testInstance, trainingSet[x], length) print('训练集:{}-距离:{}'.format(trainingSet[x], dist)) distances.append((trainingSet[x], dist)) distances.sort(key=operator.itemgetter(1)) # 把距离从小到大排列 neighbors = [] for x in range(k): #排序完成后取前k个距离 neighbors.append(distances[x][0]) print(neighbors) return neighbors def getResponse(self,neighbors): # 根据少数服从多数,决定归类到哪一类 classVotes = {} for x in range(len(neighbors)): response = neighbors[x][-1] # 统计每一个分类的多少 if response in classVotes: classVotes[response] += 1 else: classVotes[response] = 1 # 初始值为1 print(classVotes.items()) sortedVotes = sorted(classVotes.items(), key=operator.itemgetter(1), reverse=True) #reverse按降序的方式排列 return sortedVotes[0][0] def getAccuracy(self,testSet, predictions): # 准确率计算 correct = 0 for x in range(len(testSet)): if testSet[x][-1] == predictions[x]: #predictions是预测的和testset实际的比对 correct += 1 print('共有{}个预测正确,共有{}个测试数据'.format(correct,len(testSet))) return (correct/float(len(testSet)))*100.0 def Run(self): trainingSet = [] testSet = [] split = 0.75 self.loadDataset(r'testdata.txt', split, trainingSet, testSet) #数据划分 print('Train set: ' + str(len(trainingSet))) print('Test set: ' + str(len(testSet))) #generate predictions predictions = [] k = 3 # 取最近的3个数据 # correct = [] for x in range(len(testSet)): # 对所有的测试集进行测试 neighbors = self.getNeighbors(trainingSet, testSet[x], k) #找到3个最近的邻居 result = self.getResponse(neighbors) # 找这3个邻居归类到哪一类 predictions.append(result) # print('predictions: ' + repr(predictions)) 返回一个它在python中的描述 # print('>predicted=' + repr(result) + ', actual=' + repr(testSet[x][-1])) # print(correct) accuracy = self.getAccuracy(testSet,predictions) print('Accuracy: ' + repr(accuracy) + '%') if __name__ == '__main__': a = KNearestNeighbor() a.Run()