机器学习实战 --应用实例（k-近邻算法）-- 手写数字识别

机器学习实战 --应用实例（k-近邻算法）-- 手写数字识别

硬件需求：PC + 复印机 or 相机

软件：PS + Python

1. 原始照片：

2. PS处理图片：

将图片转为灰度图，且调整对比度，将背底白色与黑字区分开，避免出现灰色，为下一步二值图做好准备

从众多数字中用手动手段截取数字图片，注意尽量截取图片大小保持最小，例如：

用色阶工具，将图片的对比度调整为黑白二色，具体操作细节见链接：

http://jingyan.baidu.com/article/922554468bf465851748f459.html

示例：

将图片大小调整为高度32色素。

新建图片32*32色素，将原图片复制到该新建图片中，并放置到图片正中。

3. 图片转化为矩阵

用Python将图片由灰度图转化为二值图，且将二值图转化为1和0的矩阵

• 先安装JPEG和FreeType库

pip install libjpeg-devel freetype-devel

• 再安装PIL 或 pillow库

对于Linux、Mac、win32系统，采用PIL库

对于win64系统，采用pillow库：

非官方的64位库：

http://www.lfd.uci.edu/~gohlke/pythonlibs/

叫做Pillow，下载下来，是个 .whl 结尾的文件，这个其实就是python使用的一种压缩文件，后缀名改成zip，可以打开。

这个需要用 pip 安装。

pip install Pillow-2.7.0-cp27-none-win_amd64.whl 

• 将灰度图转化为二值图

from PIL import Image

from numpy import *

import numpy as np

img = Image.open('C:/Users/310118430/OneDrive/Machine_Learning/Program/2/4.png')

new_img = img.convert('1')

new_img.save('2/4_1.png')

• 将二值图对应的1,0矩阵储存为数字矩阵

pixels = new_img.load()

arr = zeros([new_img.width, new_img.height],int8)

for x in range(new_img.height):

for y in range(new_img.width):

if pixels[x,y]>126:

arr[x,y]=0

else: arr[x,y]=1

xm = np.matrix(arr)

#np.savetxt('2.txt',arr,fmt='%s',newline='\n')

np.savetxt('2/txt/4_1.txt',xm.T,fmt='%s')

得到数字矩阵4_1.txt，矩阵长宽为32*32

4. 运用kNN算法，识别出数字

原始程序

KNN算法：

# -*- coding: utf-8 -*-

'''

Created on Sep 16, 2010

kNN: k Nearest Neighbors

Input: inX: vector to compare to existing dataset (1xN)

dataSet: size m data set of known vectors (NxM)

labels: data set labels (1xM vector)

k: number of neighbors to use for comparison (should be an odd number)

Output: the most popular class label

@author: pbharrin

'''

from numpy import *

import operator

from os import listdir

def createDataSet():

group = array([[1.0,1.1],[1.0,1.0],[0,0],[0,0.1]])

labels = ['A','A','B','B']

return group, labels

def classify0(inX, dataSet, labels, k): #type of inX is [x,y], dataSet type is array, labels type is list, k type is int

dataSetSize = dataSet.shape[0]

diffMat = tile(inX, (dataSetSize,1)) - dataSet #tile函数位于python模块 numpy.lib.shape_base中，他的功能是重复某个数组

sqDiffMat = diffMat**2 #**2 平方

sqDistances = sqDiffMat.sum(axis=1)

distances = sqDistances**0.5

sortedDistIndicies = distances.argsort()

classCount={}

for i in range(k):

voteIlabel = labels[sortedDistIndicies[i]]

classCount[voteIlabel] = classCount.get(voteIlabel,0) + 1

sortedClassCount = sorted(classCount.iteritems(), key=operator.itemgetter(1), reverse=True)

return sortedClassCount[0][0]

def file2matrix(filename):

fr = open(filename)

numberOfLines = len(fr.readlines()) #get the number of lines in the file

returnMat = zeros((numberOfLines,3)) #prepare matrix to return

classLabelVector = [] #prepare labels return

fr = open(filename)

index = 0

for line in fr.readlines():

line = line.strip()

listFromLine = line.split('\t')

returnMat[index,:] = listFromLine[0:3]

classLabelVector.append(int(listFromLine[-1]))

index += 1

return returnMat,classLabelVector

def autoNorm(dataSet):

minVals = dataSet.min(0)

maxVals = dataSet.max(0)

ranges = maxVals - minVals

normDataSet = zeros(shape(dataSet))

m = dataSet.shape[0]

normDataSet = dataSet - tile(minVals, (m,1))

normDataSet = normDataSet/tile(ranges, (m,1)) #element wise divide

return normDataSet, ranges, minVals

def datingClassTest():

hoRatio = 0.10 #hold out 10%

datingDataMat, datingLabels = file2matrix('data/Ch02/datingTestSet2.txt')

normMat, ranges, minVals = autoNorm(datingDataMat)

m = normMat.shape[0]

numTestVecs = int(m*hoRatio)

errorCount = 0.0

for i in range(numTestVecs):

classifierResult = classify0(normMat[i,:],normMat[numTestVecs:m,:],datingLabels[numTestVecs:m],3)

print "the classifier came back with: %d, the real answer is: %d" % (classifierResult, datingLabels[i])

if (classifierResult != datingLabels[i]): errorCount += 1.0

print "the total error rate is: %f" % (errorCount/float(numTestVecs))

print errorCount

def classifyPerson():

resultList = ['not at all', 'in small doses', 'in large doses']

percentTats = float(raw_input("percentage of time spent playing video games?"))

ffMiles = float(raw_input("frequent filer miles earned per year?"))

iceCream = float(raw_input("liters of ice cream cunsumed per year?"))

datingDataMat, datingLabels = file2matrix('data/Ch02/datingTestSet2.txt')

normMat, ranges, minVals = autoNorm(datingDataMat)

inArr = array([ffMiles, percentTats, iceCream])

classifierResult = classify0((inArr-minVals)/ranges, normMat, datingLabels,3)

print "You will probably like this person: ", resultList[classifierResult -1]

def img2vector(filename):

returnVect = zeros((1,1024))

fr = open(filename)

for i in range(32):

lineStr = fr.readline()

for j in range(32):

returnVect[0,32*i+j] = int(lineStr[j])

return returnVect

def handwritingClassTest():

hwLabels = []

trainingFileList = listdir('data/Ch02/digits/trainingDigits') #load the training set

m = len(trainingFileList)

trainingMat = zeros((m,1024))

for i in range(m):

fileNameStr = trainingFileList[i]

fileStr = fileNameStr.split('.')[0] #take off .txt

classNumStr = int(fileStr.split('_')[0])

hwLabels.append(classNumStr)

trainingMat[i,:] = img2vector('data/Ch02/digits/trainingDigits/%s' % fileNameStr)

testFileList = listdir('data/Ch02/digits/testDigits') #iterate through the test set

errorCount = 0.0

mTest = len(testFileList)

for i in range(mTest):

fileNameStr = testFileList[i]

fileStr = fileNameStr.split('.')[0] #take off .txt

classNumStr = int(fileStr.split('_')[0])

vectorUnderTest = img2vector('data/Ch02/digits/testDigits/%s' % fileNameStr)

classifierResult = classify0(vectorUnderTest, trainingMat, hwLabels, 3)

print "the classifier came back with: %d, the real answer is: %d" % (classifierResult, classNumStr)

if (classifierResult != classNumStr): errorCount += 1.0

print "\nthe total number of errors is: %d" % errorCount

print "\nthe total error rate is: %f" % (errorCount/float(mTest))

def handwriting():

hwLabels = []

trainingFileList = listdir('data/Ch02/digits/trainingDigits') #load the training set

m = len(trainingFileList)

trainingMat = zeros((m,1024))

for i in range(m):

fileNameStr = trainingFileList[i]

fileStr = fileNameStr.split('.')[0] #take off .txt

classNumStr = int(fileStr.split('_')[0])

hwLabels.append(classNumStr)

trainingMat[i,:] = img2vector('data/Ch02/digits/trainingDigits/%s' % fileNameStr)

testFileList = listdir('2/txt') #iterate through the test set

errorCount = 0.0

mTest = len(testFileList)

for i in range(mTest):

fileNameStr = testFileList[i]

fileStr = fileNameStr.split('.')[0] #take off .txt

classNumStr = int(fileStr.split('_')[0])

vectorUnderTest = img2vector('2/txt/%s' % fileNameStr)

classifierResult = classify0(vectorUnderTest, trainingMat, hwLabels, 3)

print "the classifier came back with: %d, the real answer is: %d" % (classifierResult, classNumStr)

if (classifierResult != classNumStr): errorCount += 1.0

print "\nthe total number of errors is: %d" % errorCount

print "\nthe total error rate is: %f" % (errorCount/float(mTest))

kNN算法的个人感悟：

import kNN

from numpy import *

import operator

from os import listdir

group, labels = kNN.createDataSet() #group type is array, labels type is list

'''

group

Out[9]:

array([[ 1. , 1.1],

[ 1. , 1. ],

[ 0. , 0. ],

[ 0. , 0.1]])

labels

Out[10]: ['A', 'A', 'B', 'B']

'''

kNN.classify0([0,0],group,labels,3) #kNN算法的核心，给出矩阵数据，再给一个list（每行元素的类型），然后给出目标点的数据list，求出与之最近点，并用以命名其类型

'''

inX = [0,0]

dataSet = group

labels

k=3

dataSetSize = dataSet.shape[0] #4L

diffMat = tile(inX, (dataSetSize,1)) - dataSet #求出间距,

sqDiffMat = diffMat**2 #间距的平方

sqDistances = sqDiffMat.sum(axis=1) #平方和

distances = sqDistances**0.5 #间距

sortedDistIndicies = distances.argsort() #从小到大的值对应的排名

classCount={}

for i in range(k):

voteIlabel = labels[sortedDistIndicies[i]] #获取距离目标点最近的点的前k个点的序列

classCount[voteIlabel] = classCount.get(voteIlabel,0) + 1 #统计前k个点中，类型为'A'和'B'的数量分别为多少

sortedClassCount = sorted(classCount.iteritems(),key=operator.itemgetter(1), reverse=True) ##operator, 使用元组的第二个元素进行排序, reverse=True -> 降序

sortedClassCount[0][0]

'''

datingDataMat, datingLabels = kNN.file2matrix('data/Ch02/datingTestSet2.txt') #将文本转化为数据矩阵和类型list

datingDataMat #数据矩阵

datingLabels #类型list

'''

filename = 'data/Ch02/datingTestSet2.txt'

fr = open(filename) #type is fr

numberOfLines = len(fr.readlines()) #get the number of lines in the file ???

returnMat = zeros((numberOfLines,3))#prepare matrix to return, type of returnMat is array

classLabelVector = [] #prepare labels return

fr = open(filename)

index = 0

for line in fr.readlines():

line = line.strip() #type of line is str, 去除首末的空格字符

listFromLine = line.split('\t') #按照'\t'将str分割为list

returnMat[index,:] = listFromLine[0:3] #第index行的内容为listFromLine[0:3]，如果不用index，那么变为每一行元素内容都为该list

classLabelVector.append(int(listFromLine[-1])) #list 元素为单个数字，故直接用list.append()最简单

index += 1

'''

import matplotlib

import matplotlib.pyplot as plt

from matplotlib.font_manager import FontProperties

import numpy.random

font = FontProperties(fname='C:\Windows\Fonts\simfang.ttf',size=14)

fig = plt.figure()

ax = fig.add_subplot(111) #参数111的意思是：将画布分割成1行1列，图像画在从左到右从上到下的第1块

#################画出数据矩阵中所有的点####################

ax.scatter(datingDataMat[:,1], datingDataMat[:,2]) #每一行，第2列和第3列的数据，datingDataMat[:,2] 类型为 array，1000行，每行元素为一个数字

# datingDataMat[:,0]为获得的飞行常客里程数；datingDataMat[:,1]为'玩视频游戏所耗时间百分比'；datingDataMat[:,2]为'每周所消费的冰淇淋公升数'

plt.xlabel(u'玩视频游戏所耗时间百分比',fontproperties=font)

plt.ylabel(u'每周所消费的冰淇淋公升数',fontproperties=font)

#################画出数据矩阵中所有的点，并用颜色按照类型list进行分类####################

#ax.scatter(datingDataMat[:,1], datingDataMat[:,2], 150*array(random.randn(len(datingLabels))),15*array(datingLabels))

#其中按照datingLabels分类，150*array(random.randn(len(datingLabels)))描述不同类的点中心彩色部分的大小，15*array(datingLabels)描述点的颜色

#ax.scatter(datingDataMat[:,1], datingDataMat[:,2], 15*array(datingLabels),150*array(random.randn(len(datingLabels))))

ax.scatter(datingDataMat[:,1], datingDataMat[:,2], 15*array(datingLabels),15*array(datingLabels))

ax2 = fig.add_subplot(111) #参数111的意思是：将画布分割成1行1列，图像画在从左到右从上到下的第1块

ax2.scatter(datingDataMat[:,0], datingDataMat[:,1],15*array(datingLabels),15*array(datingLabels))

#####带图例的图像####

f2 = plt.figure(2)

idx_1=[]

idx_2=[]

idx_3 = []

i=0

while(i<len(datingLabels)):

if datingLabels[i]==1:

idx_1.append(i)

if datingLabels[i]==2:

idx_2.append(i)

if datingLabels[i]==3:

idx_3.append(i)

i=i+1

f2 = plt.figure(2)

p1 = plt.scatter(datingDataMat[idx_1,0], datingDataMat[idx_1,1], marker = 'x', color = 'm', label=u'不喜欢', s = 30)

p2 = plt.scatter(datingDataMat[idx_2,0], datingDataMat[idx_2,1], marker = '+', color = 'c', label=u'魅力一般', s = 50)

p3 = plt.scatter(datingDataMat[idx_3,0], datingDataMat[idx_3,1], marker = 'o', color = 'r', label=u'极具魅力', s = 15)

plt.legend(prop=font, loc = 'upper left') #prop = font显示出中文字，否则会显示出乱码

########对数据矩阵中的每列，进行归一化 newValue = (oldValue-min)/(max-min)#######

normMat, ranges, minVals=kNN.autoNorm(datingDataMat)

normMat

ranges

minVals

'''

dataSet = datingDataMat

minVals = dataSet.min(0) #沿着axis=0的方向取最小值，minVals type is list

maxVals = dataSet.max(0)

ranges = maxVals - minVals

normDataSet = zeros(shape(dataSet))

m = dataSet.shape[0] #沿着axis=0的方向，shape值是多少，本例中即行数

normDataSet = dataSet - tile(minVals, (m,1)) #堆叠，minVals(list)按行，将整个矩阵重复m次；按列，重复1次

normDataSet = normDataSet/tile(ranges, (m,1)) #element wise divide

'''

######测试算法：用剩余的10%数据测算，之前建立的模型，算出某点的类型，是否与实际情况相吻合###

#kNN.datingClassTest()

'''

hoRatio = 0.10 #hold out 10%

datingDataMat, datingLabels = kNN.file2matrix('data/Ch02/datingTestSet2.txt') # datingDataMat: array; datingLabels: list

normMat, ranges, minVals = kNN.autoNorm(datingDataMat) #normMat: array; ranges: array; minVals: array

m = normMat.shape[0] #m: long，数据矩阵总行数

numTestVecs = int(m*hoRatio)

errorCount = 0.0

for i in range(numTestVecs):

classifierResult = kNN.classify0(normMat[i,:],normMat[numTestVecs:m,:],datingLabels[numTestVecs:m],10) #classifierResult: int;前100个用来检验，后900个用来建立模型

print "the classifier came back with: %d, the real answer is: %d" % (classifierResult, datingLabels[i])

if (classifierResult != datingLabels[i]): errorCount += 1.0 #统计错误总数

print "the total error rate is: %f" % (errorCount/float(numTestVecs)) #统计概率

print errorCount

'''

####### 用raw_input 输入某种情况，用模型算出结果，给出结论 ######

kNN.classifyPerson()

############ 将 32x32的矩阵，转化为1x1024的矩阵 ##############

testVector = kNN.img2vector('data/Ch02/digits/testDigits/0_13.txt')

testVector[0,0:31]

testVector[0,32:63]

######## 识别手写数字 #########

#先将训练集中的数字矩阵，由32x32，转化为1x1024；

#再导入进array，用每个训练文件建立一行，m个训练文件，建立一列，建立数据矩阵 m*1024

#从每个训练文件的文件名中提取label，建立label的list

#再用几个检验矩阵，带进模型中计算，验证下准确率

kNN.handwritingClassTest()

------

手写数字识别

import kNN

from numpy import *

import operator

from os import listdir

######## 识别手写数字 #########

#先将训练集中的数字矩阵，由32x32，转化为1x1024；

#再导入进array，用每个训练文件建立一行，m个训练文件，建立一列，建立数据矩阵 m*1024

#从每个训练文件的文件名中提取label，建立label的list

#再用几个检验矩阵，带进模型中计算，验证下准确率

#kNN.handwritingClassTest()

'''

hwLabels = [] #作为label的list

trainingFileList = listdir('data/Ch02/digits/trainingDigits') #load the training set, type of trainingFileList is list

m = len(trainingFileList)

trainingMat = zeros((m,1024)) #数据矩阵

for i in range(m):

fileNameStr = trainingFileList[i]

fileStr = fileNameStr.split('.')[0] #take off .txt

classNumStr = int(fileStr.split('_')[0])

hwLabels.append(classNumStr)

trainingMat[i,:] = kNN.img2vector('data/Ch02/digits/trainingDigits/%s' % fileNameStr) #将32*32的矩阵转化为1*1024，再写入大的矩阵中，作为一行

testFileList = listdir('data/Ch02/digits/testDigits')#iterate through the test set

errorCount = 0.0

mTest = len(testFileList)

for i in range(mTest):

fileNameStr = testFileList[i]

fileStr = fileNameStr.split('.')[0] #take off .txt

classNumStr = int(fileStr.split('_')[0])

vectorUnderTest = kNN.img2vector('data/Ch02/digits/testDigits/%s' % fileNameStr) #用做识别的样本

classifierResult = kNN.classify0(vectorUnderTest, trainingMat, hwLabels, 3)

print "the classifier came back with: %d, the real answer is: %d" % (classifierResult, classNumStr)

if (classifierResult != classNumStr): errorCount += 1.0

print "\nthe total number of errors is: %d" % errorCount

print "\nthe total error rate is: %f" % (errorCount/float(mTest))

'''

kNN.handwriting()

-------------

图像与数字矩阵的转换

from PIL import Image

from numpy import *

import numpy as np

img = Image.open('2/4.png')

new_img = img.convert('1')

new_img.save('2/4_1.png')

pixels = new_img.load()

arr = zeros([new_img.width, new_img.height],int8)

for x in range(new_img.height):

for y in range(new_img.width):

if pixels[x,y]>126:

arr[x,y]=0

else: arr[x,y]=1

xm = np.matrix(arr)

#np.savetxt('2.txt',arr,fmt='%s',newline='\n')

np.savetxt('2/txt/4_1.txt',xm.T,fmt='%s')

'''

l1 = arr.tolist()

file=open('data.txt','w')

file.write(str(l1));

file.close()

'''