机器学习实战--KNN手写数字识别

数据集：https://github.com/pbharrin/machinelearninginaction/blob/master/Ch02/digits.zip

代码：

import numpy as np
import operator
import matplotlib
import matplotlib.pyplot as plt
import os
def classfy0KNN(intX,dataset,labels,K):
    newX = np.tile(intX,(dataset.shape[0],1))
    diff = newX - dataset
    sqrDiff = diff**2
    sumSqrDiff = sqrDiff.sum(axis=1)
    distance = sumSqrDiff**0.5
    sortIndex = distance.argsort()
    LabelDir = {}
    for i in range(K):
        labelName = labels[sortIndex[i]]
        LabelDir[labelName] = LabelDir.get(labelName,0) + 1
    sortDir = sorted(LabelDir.items(),key=operator.itemgetter(1),reverse=True)
    return sortDir[0][0]

def img2vector(filename):
    oneImg = np.zeros((1,1024))
    with open(filename) as f:
        for i in range(32):
            oneline = f.readline()
            for j in range(32):
                oneImg[0,32*i+j] = int(oneline[j])
    return oneImg

def handwritingClassTest():
    hwLables = []
    trainingFileList = os.listdir('trainingDigits')
    m = len(trainingFileList)
    trainDataset = np.zeros((m,1024))
    for i in range(m):
        fileNameStr = trainingFileList[i]
        #获取该图像对应数字标签
        fileStr = fileNameStr.split('.')[0]
        classNumLable = int(fileStr.split('_')[0])
        hwLables.append(classNumLable)
        trainDataset[i,:] = img2vector('trainingDigits/'+fileNameStr)
    testFileList = os.listdir('testDigits')
    m = len(testFileList)
    errorNum = 0
    for i in range(m):
        fileNameStr = testFileList[i]
        # 获取该图像对应数字标签
        fileStr = fileNameStr.split('.')[0]
        classNumLable = int(fileStr.split('_')[0])
        testVect = img2vector('testDigits/'+fileNameStr)
        predict = classfy0KNN(testVect,trainDataset,hwLables,5)
        print('the real number is : ',classNumLable,' predict is : ',predict)
        if predict != classNumLable:
            errorNum += 1
    print('the error rate is : ',(errorNum/m))



if __name__ == '__main__':
    # testVect = img2vector('testDigits/0_13.txt')
    # print(testVect[0,:31])
    handwritingClassTest()

 

运行结果：