k近邻算法--手写识别系统

下面的例子来源为《机器学习实战》，例子只能识别0-9。

首先需要将图像二进制数据转化为测试向量：

def imgTransformVector(filename):  # 将 32x32 二进制图像矩阵转化为 1x1024 向量
    returnVector = np.zeros((1,1024))
    fr = open(filename)
    for i in range(32):
        lineStr = fr.readline()
        for j in range(32):
            returnVector[0,32*i+j] = int(lineStr[j])
    return returnVector

接着是算法的实现代码：

def handWritingTextTest():
    handWritingLabels = []
    # listdir 返回指定的文件夹包含的文件或文件夹的名字的列表
    trainingFileList = os.listdir(‘/Users/Desktop/trainingDigits‘)
    trainingDataLen = len(trainingFileList)                # 获取训练数据集的大小
    trainingMatrix = np.zeros((trainingDataLen,1024))
    for i in range(trainingDataLen -1):
        fileNameString = trainingFileList[i + 1]          # 第i个训练样本的文件名
        fileString = fileNameString.split(‘.‘)[0]    # 截去.txt部分
        classNumberString = int(fileString.split(‘_‘)[0])   #获得分类数字
        handWritingLabels.append(classNumberString)
        trainingMatrix[i,:] = imgTransformVector(‘/Users/Desktop/trainingDigits/%s‘%fileNameString)
    testFileList = os.listdir(‘/Users/Desktop/testDigits‘)
    errorCount = 0.0
    testDataLen = len(testFileList)
    for i in range(testDataLen - 1):
        fileNameString = testFileList[i +1]
        fileString = fileNameString.split(‘.‘)[0]
        classNumberString = int(fileString.split(‘_‘)[0])
        testDataVector = imgTransformVector(‘/Users/Desktop/testDigits/%s‘%fileNameString)
        classifierResult = classifyPerson(testDataVector,trainingMatrix,handWritingLabels,3)

        if (classifierResult != classNumberString):
            errorCount += 1
            print(‘the classifier:%d, the real answer:%d‘ % (classifierResult, classNumberString))
    print(‘\nthe total errorCount:%d‘%errorCount)
    print(‘\nthe total errorRate:%.d‘%(errorCount/float(testDataLen)))