环境：python 3, scikit-learn 0.18

KNN可以看成：有一堆已经知道分类的数据，当一个新数据进入的时候，就开始跟训练数据里的每个点求距离，然后挑离这个训练数据最近的K个点看看这几个点属于什么类型，然后利用少数服从多数的原则，给新数据归类。

#coding:utf-8 """ sklearn 0.18 python 3 """ from sklearn.model_selection import GridSearchCV from sklearn.model_selection import train_test_split from sklearn.neighbors import KNeighborsClassifier from sklearn.metrics import accuracy_score,confusion_matrix,classification_report import input_data import numpy as np import pickle mnist = input_data.read_data_sets('mnist/',one_hot=False) x = mnist.train.images y = mnist.train.labels #train_test_split设置一部分数据作为验证集 train_data,validation_data,train_labels,validation_labels = train_test_split(x,y,test_size=0.1) #clf相当于一个采用KNN算法进行分类 m_neighbors=4表示k=4的分类器 clf = KNeighborsClassifier(n_neighbors=4,algorithm='auto',weights='distance') clf.fit(train_data,train_labels) predictions=[] for i in range(1000): if i % 100 == 0: print('= = = = = = > > > > > >','epoch :',int(i/100)) #满足输入规范要求clf.predict([]) output = clf.predict([mnist.test.images[i]]) predictions.append(output) #混淆矩阵 print (confusion_matrix(mnist.test.labels[0:1000],predictions)) #f1-score,precision,recall print (classification_report(mnist.test.labels[0:1000],np.array(predictions))) #计算准确度 print ('test accuracy is :',accuracy_score(mnist.test.labels[0:1000],predictions)) #将训练好的分类器保存 with open('mnist_knn.pickle','wb') as f: pickle.dump(clf,f)

结果

分析统计结果中的混淆矩阵,recall,precision,F1-score KNN算法对mnist数据集检测精度达到96%