用python实现KNN算法

knn是最简单的分类算法，基于经验，利用各种距离进行分类

下面给出一种简单的实现：

'''
手动实现KNN算法
'''
import  numpy as np

class KNN:

    def __init__(self,exp_x,exp_y,weights=1,distance_func=None):
        '''
        :param exp_x: 样本的特征向量集
        :param exp_y: 对应的标签集
        :param weights: 每个特征对应的权重
        :param distance_func: 用于计算距离的函数,distance_func(simple_feature_vector,target_feature_vector)
        '''
        self.__weights=weights
        if not isinstance(exp_x,np.ndarray):
            self.__x=np.array(exp_x)
        self.__y=exp_y
        if distance_func is None:
            self.__distance_func = self.default_dist
        else:
            self.__distance_func = distance_func

    def default_dist(self,vector,target_vector,weights):
        return np.sum(((target_vector - vector)*weights) ** 2) ** 0.5

    def predict(self,target_vector,k=5):
        '''
        :param target_vector: 需要预测的特征向量
        :param k: 结果取前k个
        :return: 返回结果标签
        '''
        assert k<len(self.__y),'the k must be less the length of exp_y'
        if not isinstance(target_vector,np.ndarray):
            target_vector=np.array(target_vector)
        all_dist=[]#将所有距离的结果放入一个列表中，如果数据量比较大则需要使用数据库进行存储
        for vector,tag in zip(self.__x,self.__y):
            ds=self.__distance_func(vector,target_vector,self.__weights)
            all_dist.append((ds,tag))
        all_dist.sort(key=lambda x:x[0])
        result_list=all_dist[:k+1]
        freq={} #统计标签频率
        for ds,tag in result_list:
            freq[tag] = freq.get(tag,0) + 1
        return max(freq.items(),key=lambda x:x[1])[0] #返回频率高的标签

if __name__ == '__main__':
    x=[[0,0],
       [1,1],
       [-1,-1],
       [10,0],
       [11,0],
       [12,0],]
    y=['A','A','A','B','B','B']
    knn=KNN(x,y)
    r=knn.predict([7,6],2)
    print(r)

测试结果：

"D:\Program Files\Python\Python36\python.exe" D:/Workspace/测试空间/机器学习/实现knn算法.py
B

Process finished with exit code 0