手写数字识别

载入数据

KNN分类

from numpy import *
import operator
import os

# KNN分类核心方法
def kNNClassify(newInput, dataSet, labels, k):  #(test_x[i],train_x,train_y,3)
    numSamples = dataSet.shape[0]  # shape[0]代表数据集的行数

    #欧式距离
    # tile(A, reps): 将A重复reps次来构造一个矩阵
    diff = tile(newInput, (numSamples, 1)) - dataSet  # 差
    squaredDiff = diff ** 2  # 差值的平方
    squaredDist = sum(squaredDiff, axis = 1)   # 差值的平方和
    distance = squaredDist ** 0.5 #1/2次方

    #街市距离
    #diff = tile(newInput,(numSamples,1)) - dataSet
    #absDiff = fabs(diff)
    #absDist = sum(absDiff,axis = 1)
    #distance = absDist

    # # step 2: 对距离排序
    # argsort()返回排序后的索引
    sortedDistIndices = argsort(distance)

    classCount = {}  # 定义一个空的字典
    for i in range(k):
        # # step 3: 选择k个最小距离
        voteLabel = labels[sortedDistIndices[i]]

        # # step 4: 计算类别的出现次数
        # when the key voteLabel is not in dictionary classCount, get()
        # will return 0
        classCount[voteLabel] = classCount.get(voteLabel, 0) + 1

    # # step 5: 返回出现次数最多的类别作为分类结果
    maxCount = 0
    for key, value in classCount.items():
        if value > maxCount:
            maxCount = value
            maxIndex = key

    return maxIndex

图片转换为向量

# 将图片转换为向量
def  img2vector(filename):
    rows = 32
    cols = 32
    imgVector = zeros((1, rows * cols))
    fileIn = open(filename)
    for row in range(rows):
        lineStr = fileIn.readline()
        for col in range(cols):
            imgVector[0, row * 32 + col] = int(lineStr[col])

    return imgVector

加载数据集

# 加载数据集
def loadDataSet():
    # # step 1: 读取训练数据集
    print ("---Getting training set...")
    dataSetDir = 'D:/practice/digits/'
    trainingFileList = os.listdir(dataSetDir + 'trainingDigits')  # 加载测试数据
    numSamples = len(trainingFileList)  #training文件夹里有多少文件

    train_x = zeros((numSamples, 1024))  #train_x是一个文件数x1024的数组
    train_y = []
    for i in range(numSamples):
        filename = trainingFileList[i]  #training文件夹里的文件名

        # get train_x
        train_x[i, :] = img2vector(dataSetDir + 'trainingDigits/%s' % filename) #将一个txt文件转化为向量，存入train_x的第i行

        # get label from file name such as "1_18.txt"
        label = int(filename.split('_')[0]) # 用'_'分隔开文件名，取前面的部分为标签，如"1_18.txt"，返回1
        train_y.append(label)  #train_y加入取得的label

    # # step 2:读取测试数据集
    print("---Getting testing set...")
    testingFileList = os.listdir(dataSetDir + 'testDigits') # load the testing set
    numSamples = len(testingFileList)
    test_x = zeros((numSamples, 1024))
    test_y = []
    for i in range(numSamples):
        filename = testingFileList[i]

        # get train_x
        test_x[i, :] = img2vector(dataSetDir + 'testDigits/%s' % filename)

        # get label from file name such as "1_18.txt"
        label = int(filename.split('_')[0]) # return 1
        test_y.append(label)

    return train_x, train_y, test_x, test_y

主实现函数

def testHandWritingClass():
    # # step 1: 加载数据
    print("step 1: load data...")
    train_x, train_y, test_x, test_y = loadDataSet()

    # # step 2: 模型训练.
    print("step 2: training...")
    pass

    # # step 3: 测试
    print("step 3: testing...")
    numTestSamples = test_x.shape[0]  #测试向量个数
    matchCount = 0 #匹配数
    for i in range(numTestSamples):
        predict = kNNClassify(test_x[i], train_x, train_y, 7)  #对test_x的每行向量，train_x作为数据集，train_y作为标签集，进行K=3的knn分类
        if predict == test_y[i]:
            matchCount += 1
        else:
            print("The predict is:",predict," but the test is ",test_y[i])
    accuracy = float(matchCount) / numTestSamples

    # # step 4: 输出结果
    print("step 4: show the result...")
    print(('The classify accuracy is: %.2f%%')% (accuracy * 100))

结果比较

K=7

K=5

K=3

可见当K值越小，由K个最近邻类别投票的结果就越精确