k_Nearest Neighbor算法

最新推荐文章于 2024-06-05 12:38:29 发布

原创最新推荐文章于 2024-06-05 12:38:29 发布 · 336 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#kNN算法 #机器学习 #heapq #numpy

机器学习专栏收录该内容

6 篇文章

订阅专栏

本文介绍了一个基于K近邻(KNN)算法的实现案例，通过Python代码展示了如何使用该算法进行分类预测，并给出了不同K值下的准确率表现。实验结果显示，在特定范围内选择合适的K值对于提高分类准确率至关重要。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

算法同之前的NN算法没什么太大的变化
直接上代码
程序输出不同k值下的准确率

#coding=utf-8
#author='HL'
'''
利用pickle模块存储对象
dump类似于write
load类似于read
完成对象的上传以及读取
'''

import cPickle as pickle
import numpy as np
import heapq



class NearestNeighbor(object):
    def __init__(self):
        pass

    def train(self,x,y):
        self.x_Tr = x
        self.y_Tr = y

    def predict(self,x,k):
        y_Pre = np.zeros(x.shape[0],dtype=type(self.y_Tr))

        for i in range(x.shape[0]):
            distance = np.sum(abs(self.x_Tr-x[i,]),axis=1)
            min_index = self.n_Labels_index_get(distance,k)
            dict1 = {}
            m = 0
            for index in min_index:
                dict1[self.y_Tr[index]] = dict1.get(self.y_Tr[index],0)+1
                if dict1[self.y_Tr[index]]>m:
                    m =  dict1[self.y_Tr[index]]
                    y_Pre[i] = self.y_Tr[index]

        print y_Pre
        return y_Pre

    def n_Labels_index_get(self,distance,n):
            return heapq.nsmallest(n,range(len(distance)),distance.take)




def un_P(path):
    f = file(path,'rb')
    dict1 = pickle.load(f)
    f.close()
    return dict1


data_Train = []
labels_Train = []
path = "A:/python_test/cifar-10-batches-py/"
for i in range(1,6):
    now_Path = path+'data_batch_'+str(i)
    dict1 = un_P(now_Path)
    it  = 0
    while it <len(dict1['data']):
        data_Train.append(dict1['data'][it])
        labels_Train.append(dict1['labels'][it])
        it+=10
    '''
     for item in dict1['data']:
        data_Train.append(item)
    for item in dict1['labels']:
        labels_Train.append(item)
    '''



data_Test = []
labels_Test = []
dict1 = un_P(path+'test_batch')
for item in dict1['data'][:1000]:
    data_Test.append(item)
for item in dict1['labels'][:1000]:
    labels_Test.append(item)



data_Tr = np.asarray(data_Train)
data_Te = np.asarray(data_Test)
labels_Tr = np.asarray(labels_Train)
labels_Te = np.asarray(labels_Test)
print data_Tr.shape
print data_Te.shape
print labels_Tr.shape
print labels_Te.shape




zz = NearestNeighbor()
zz.train(data_Tr,labels_Train)
res = []
for k in [1,3,5,7,9,15,20,40,60,100]:
    out = np.mean(zz.predict(data_Te,k)==labels_Te)
    res.append([k,out])
    print res
print res