python之k近邻的手写数字识别

1.准备数据：

# downloadDatas.py
import numpy as np
def loadDatas(txtName):
    numberDatas = np.zeros((1,1024))
    f = open(txtName)
    for i in range(32):
        datasLines = f.readline()
        for j in range(32):
            numberDatas[0,32*i+j] = int(datasLines[j])
    return numberDatas
# txtName = '0_0.txt'
# print(loadDatas(txtName)[0,0:32])

2.k-近邻算法：

# knn.py
import numpy as np
import operator
def knn(test,trains,labels,k):
    trainsSizeHang = trains.shape[0]
    values = np.tile(test,(trainsSizeHang,1)) - trains
    sqValues = values ** 2
    sqDistances = sqValues.sum(axis = 1)
    distances = sqDistances ** 0.5
    sortedDists = np.argsort(distances)
    classCount = {}
    for i in range(k):
        voteLabel = labels[sortedDists[i]]
        classCount[voteLabel] = classCount.get(voteLabel,0) + 1
    classCount = classCount.items()
    sortClassCount = sorted(classCount,key=operator.itemgetter(1),reverse=True)
    return sortClassCount[0][0]

# trains = [[3 ,104],
#           [2 ,100],
#           [1 ,81],
#           [101 ,10],
#           [99 ,5],
#           [98 ,2]]
# # ‘1’表示爱情片，‘2’表示动作片
# labels = [1 , 1 , 1 , 2 , 2 ,2]
# test = [18 ,90]
# k = 3
#
# trains = np.array(trains)
# labels = np.array(labels)
# test = np.array(test)
# print('测试样本的类别：' + str(knn(test , trains , labels , k)))

3.测试算法：

import numpy as np
import os
from downloadDatas import loadDatas
from knn import knn
trainFileList = os.listdir('trainingDigits')
trainNum = len(trainFileList)
trainDatas = np.zeros((trainNum,1024))
labels = []
for i in range(trainNum):
    txtName = trainFileList[i]
    txt = txtName.split('.')[0]
    txtClass = int(txt.split('_')[0])
    labels.append(txtClass)
    trainDatas[i,:] = loadDatas('trainingDigits/%s' % txtName)
labels = np.array(labels)
print(labels)
testFileList = os.listdir('testDigits')
testNum = len(testFileList)
errorNum = 0
for i in range(testNum):
    testTxtName = testFileList[i]
    testTxt = testTxtName.split('.')[0]
    testTxtClass = int(testTxt.split('_')[0])
    testData = loadDatas('testDigits/%s' % testTxtName)
    checkResult = knn(testData,trainDatas,labels,3)
    if checkResult != testTxtClass:
        errorNum +=1
    print('测试结果：' + str(checkResult) + ',正确结果：' + str(testTxtClass))
error = errorNum / testNum
print('测试结果的错误率：' + '%.2f%%' % (error*100))