1. 综述
1.1 Cover和Hart在1968年提出了最初的邻近算法
1.2 分类(classification)算法|回归算法,这里讨论的是分类算法
1.3 输入基于实例的学习(instance-based learning), 懒惰学习(lazy learning)
2.1 步骤
为了判断未知实例的类别,以所有已知类别的实例作为参照
选择参数K
计算未知实例与所有已知实例的距离
选择最近K个已知实例
根据少数服从多数的投票法则(majority-voting),让未知实例归类为K个最邻近样本中最多数的类别
2.2 细节:
关于K
关于距离的衡量方法:
2.2.1 Euclidean Distance 定义
其他距离衡量:余弦值(cos), 相关度 (correlation), 曼哈顿距离 (Manhattan distance)
2.3 举例
3. 例子
未知电影属于什么类型?
4、python代码实现
利用Python的机器学习库sklearn: SkLearnExample.py
import csv import random import math import operator def loadDataset(filename, split, trainingSet = [], testSet = []): ''' 导入数据 :param filename: :param split: 将数据总集以split为界限 分成训练集和测试集 :param trainingSet: :param testSet: :return: ''' with open(filename, 'rt') as csvfile: # 以逗号为分隔符 lines = csv.reader(csvfile) # 读取所有行 dataset = list(lines) for x in range(len(dataset)-1): for y in range(4): dataset[x][y] = float(dataset[x][y]) if random.random() < split: trainingSet.append(dataset[x]) else: testSet.append(dataset[x]) def euclideanDistance(instance1, instance2, length): ''' 计算euclideanDistance :param instance1: :param instance2: :param length: 维度 :return: ''' distance = 0 for x in range(length): distance += pow((instance1[x]-instance2[x]), 2) return math.sqrt(distance) def getNeighbors(trainingSet, testInstance, k): ''' 返回最近的k个邻居 :param trainingSet: 训练集 :param testInstance: 一个测试实例 :param k: 参数k :return: ''' distances = [] length = len(testInstance)-1 for x in range(len(trainingSet)): #testinstance dist = euclideanDistance(testInstance, trainingSet[x], length) distances.append((trainingSet[x], dist)) #distances.append(dist) distances.sort(key=operator.itemgetter(1)) neighbors = [] for x in range(k): neighbors.append(distances[x][0]) return neighbors def getResponse(neighbors): ''' 以距离排序,返回最近的几个点 :param neighbors: :return: ''' classVotes = {} for x in range(len(neighbors)): response = neighbors[x][-1] if response in classVotes: classVotes[response] += 1 else: classVotes[response] = 1 sortedVotes = sorted(classVotes.items(), key=operator.itemgetter(1), reverse=True) # python3 里的.items()返回的是列表,.iteritems()返回的是一个迭代器 return sortedVotes[0][0] def getAccuracy(testSet, predictions): ''' 预测值和实际值的准确率 :param testSet: :param predictions: :return: ''' correct = 0 for x in range(len(testSet)): if testSet[x][-1] == predictions[x]: correct += 1 return (correct/float(len(testSet)))*100.0 def main(): #prepare data trainingSet = [] testSet = [] split = 0.67 loadDataset(r'irisdata.txt', split, trainingSet, testSet) print('Train set: ' + repr(len(trainingSet))) print('Test set: ' + repr(len(testSet))) #generate predictions predictions = [] k = 3 for x in range(len(testSet)): # trainingsettrainingSet[x] neighbors = getNeighbors(trainingSet, testSet[x], k) result = getResponse(neighbors) predictions.append(result) print ('>predicted=' + repr(result) + ', actual=' + repr(testSet[x][-1])) accuracy = getAccuracy(testSet, predictions) print('Accuracy: ' + repr(accuracy)+ '%') if __name__ == '__main__': main()
5. 算法优缺点
5.1 算法优点
简单
易于理解
容易实现
通过对K的选择可具备丢噪音数据的健壮性
5.2 算法缺点
需要大量空间储存所有已知实例
算法复杂度高(需要比较所有已知实例与要分类的实例)
当其样本分布不平衡时,比如其中一类样本过大(实例数量过多)占主导的时候,新的未知实例容易被归类为这个主导样本,因为这类样本实例的数量过大,但这个新的未知实例实际并木接近目标样本