KNN算法示例

# -*- 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
                    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
        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(correct)
        accuracy = self.getAccuracy(testSet,predictions)
        print('Accuracy: ' + repr(accuracy) + '%')


if __name__ == '__main__':
    a = KNearestNeighbor()
    a.Run()