1、KNN 学习笔记

K Near Neighbor

优点：

缺点：计算复杂度高，空间复杂度高

一般算法流程：

1、收集、准备和分析数据

2、训练数据(不适用 KNN 算法)

3、测试数据

4、使用算法

准备： createDataSet()

KNN 分类算法的流程：（5）

classify0()

约会网站上使用 KNN 算法流程：

file2matrix() 准备数据 autoNorm() 归一化

matplotlib 通过散点图分析数据

datingClassTest() 测试代码

classifyPerson() 约会网站预测函数， 使用算法

fig.add_subplot(111)

ax.scatter(,)

A.min(0)

A.max(0)

测试代码：

datingClassTest()

完整系统

classifyPerson()

手写体识别

img2Vector

handwritingClassTest

#! /usr/bin/python
# -*- coding: utf8 -*-
'''
2017/8/16
KNN在约会网站和手写体识别系统中的应用
'''
from numpy import *
import operator
import matplotlib
import matplotlib.pyplot as plt
import os

#数据准备
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

# KNN算法
def classify0(inX,dataMat,labels,k):
length = dataMat.shape[0]
# diffData = dataMat - tile(inX,length) #代码错误****
diffData = dataMat - tile(inX,(length,1))
squareDiffData = diffData**2
addSquareData = squareDiffData.sum(axis = 1) #按照行相加
distances = addSquareData ** 0.5
sortedDistance = distances.argsort()
classCount = {}
for i in range(k):
votedLabel = labels[sortedDistance[i]]
classCount[votedLabel] = classCount.get(votedLabel,0) + 1
sortedClassCount = sorted(classCount.items(),key = operator.itemgetter(1),reverse = True)
return sortedClassCount[0][0]

# group,labels = createDataSet()
# print(classify0([0,0],group,labels,3))

#将文本结构化
def file2matrix(path):
fr = open(path)
lines = fr.readlines()
length = len(lines)
returnMat = zeros((length,3))
label = []
index = 0
for line in lines:
line = line.strip()
sqlitLine = line.split('\t')
returnMat[index,:] = sqlitLine[0:3]
label.append(int(sqlitLine[-1]))
index += 1
return returnMat,label

#数据归一化处理
def autoNorm(dataMat):
minVals = dataMat.min(0)
maxVals = dataMat.max(0)
ranges = maxVals - minVals
length = len(dataMat)
normalDataSet = (dataMat - tile(minVals,(length,1)))/tile(ranges,(length,1))
return minVals,maxVals,ranges,normalDataSet

# 分析数据（使用Matplotlib创建散点图）
# dataMat,labels = file2matrix('./../data/datingTestSet2.txt')
# fig = plt.figure()
# ax = fig.add_subplot(111)
# # ax.scatter(dataMat[:,1],dataMat[:,2])# 只有一种颜色
# ax.scatter(dataMat[:,1],dataMat[:,2],15.0*array(labels),15.0*array(labels))
# plt.xlabel('玩视频游戏所耗时间百分比')
# plt.ylabel('每周消费的冰淇淋公升数')
# plt.show()
# print('测试')

# 把集合分成两部分进行测试

# def datingClassTest(path):
# hoRatio = 0.10
# dataMat,labels = file2matrix(path)
# length = dataMat.shape[0]
# testLength =int(length * hoRatio)
# dataMin,dataMax,ranges,newDataMat = autoNorm(dataMat)
# errorCount = 0
# for i in range(testLength):
# classifierResult = classify0(newDataMat[i,:],newDataMat[testLength:length,:],labels[testLength:length],3)
# print("the classifier came back with %d,the real answer is %d" %(classifierResult,labels[i]))
# if (classifierResult == labels[i]):
# errorCount += 1
# print(" the error ratio is %f" %(errorCount/float(testLength)))
#
# datingClassTest('./../data/datingTestSet2.txt')

# #完整约会推检测系统
# def classifierPerson(path):
# resultList = ['not at all','in a small does','in large does']
# percentTats = float(input("percentage of time spent playing video games?"))
# ffMiles = float(input("frequent flier miles earned per year?"))
# iceCream = float(input("litters of icecream consumed per year?"))
# dataMat,labels = file2matrix(path)
# minEle,maxEle,ranges,newDataMat = autoNorm(dataMat)
# inX = ([percentTats,ffMiles,iceCream])
# newInX = (inX - minEle)/ranges
# resultLabel = classify0(newInX,newDataMat,labels,3)
# print(resultList[resultLabel])
# classifierPerson('./../data/datingTestSet2.txt')

# 使用KNN进行手写体识别

#首先将二进制文件转换成向量的结构化结果
def img2vector(filename):
returnMat = zeros((1,1024))
fr = open(filename)
for i in range(32):
line = fr.readline()
for j in range(32):
returnMat[0,i*32+j] = line[j]
return returnMat

def handwritingClassTest():
traindir = './../data//trainingDigits//'
testdir = './../data//testDigits//'
filelist = os.listdir(traindir)
length = len(filelist)
dataMat = zeros((length,1024))
labels = []
for i in range(length):
newpath = traindir + filelist[i]
singleLabel = filelist[i].split('_')[0]
labels.append(singleLabel)
dataMat[i,:] = img2vector(newpath)
testfilelist = os.listdir(testdir)
testlength = len(testfilelist)
testdataMat = zeros((testlength, 1024))
testlabels = []
errorRatio = 0.0
for i in range(testlength):
newpath = testdir + testfilelist[i]
singleLabel = int(testfilelist[i].split('_')[0])
testlabels.append(singleLabel)
testdataMat[i,:] = img2vector(newpath)
result = int(classify0(testdataMat[i],dataMat,labels,3))
print("real value: %d, predicted value:%d" %(testlabels[i],result))
if (testlabels[i] != result):
errorRatio += 1

print("errorRatio is %d , testlength is %d , ratio is %f" %(errorRatio,testlength,errorRatio/float(testlength)))

handwritingClassTest()