尝试一些用KNN来做数字识别,测试数据来自:
MNIST handwritten digit database, Yann LeCun, Corinna Cortes and Chris Burges
http://yann.lecun.com/exdb/mnist/
1、数据
将位图转为向量(数组),k尝试取值3-15,距离计算采用欧式距离。
d(x,y)=sqrt{sum_{i=1}^{n}(x_i-y_i)^2}
2、测试
调整k的取值和基础样本数量,测试得出k取值对识别正确率的影响,以及分类识别的耗时。
如何用python解析mnist图片 - 海上扬凡的博客 - 博客频道 - CSDN.NET
http://blog.csdn.net/u014046170/article/details/47445919
# -*- coding: utf-8 -*-
"""
Created on Wed Mar 08 14:38:15 2017
@author: zapline<278998871@qq.com>
"""
import struct
import os
import numpy
def read_file_data(filename):
f = open(filename, 'rb')
buf = f.read()
f.close()
return buf
def loadImageDataSet(filename):
index = 0
buf = read_file_data(filename)
magic, images, rows, columns = struct.unpack_from('>IIII' , buf , index)
index += struct.calcsize('>IIII')
data = numpy.zeros((images, rows * columns))
for i in xrange(images):
imgVector = numpy.zeros((1, rows * columns))
for x in xrange(rows):
for y in xrange(columns):
imgVector[0, x * columns + y] = int(struct.unpack_from('>B', buf, index)[0])
index += struct.calcsize('>B')
data[i, :] = imgVector
return data
def loadLableDataSet(filename):
index = 0
buf = read_file_data(filename)
magic, images = struct.unpack_from('>II' , buf , index)
index += struct.calcsize('>II')
data = []
for i in xrange(images):
lable = int(struct.unpack_from('>B', buf, index)[0])
index += struct.calcsize('>B')
data.append(lable)
return data
def loadDataSet():
path = "D:\kingsoft\ml\dataset\"
trainingImageFile = path + "train-images.idx3-ubyte"
trainingLableFile = path + "train-labels.idx1-ubyte"
testingImageFile = path + "t10k-images.idx3-ubyte"
testingLableFile = path + "t10k-labels.idx1-ubyte"
train_x = loadImageDataSet(trainingImageFile)
train_y = loadLableDataSet(trainingLableFile)
test_x = loadImageDataSet(testingImageFile)
test_y = loadLableDataSet(testingLableFile)
return train_x, train_y, test_x, test_y
# -*- coding: utf-8 -*-
"""
Created on Wed Mar 08 14:35:55 2017
@author: zapline<278998871@qq.com>
"""
import numpy
def kNNClassify(newInput, dataSet, labels, k):
numSamples = dataSet.shape[0]
diff = numpy.tile(newInput, (numSamples, 1)) - dataSet
squaredDiff = diff ** 2
squaredDist = numpy.sum(squaredDiff, axis = 1)
distance = squaredDist ** 0.5
sortedDistIndices = numpy.argsort(distance)
classCount = {}
for i in xrange(k):
voteLabel = labels[sortedDistIndices[i]]
classCount[voteLabel] = classCount.get(voteLabel, 0) + 1
maxCount = 0
for key, value in classCount.items():
if value > maxCount:
maxCount = value
maxIndex = key
return maxIndex
# -*- coding: utf-8 -*-
"""
Created on Wed Mar 08 14:39:21 2017
@author: zapline<278998871@qq.com>
"""
import dataset
import knn
def testHandWritingClass():
print "step 1: load data..."
train_x, train_y, test_x, test_y = dataset.loadDataSet()
print "step 2: training..."
pass
print "step 3: testing..."
numTestSamples = test_x.shape[0]
matchCount = 0
for i in xrange(numTestSamples):
predict = knn.kNNClassify(test_x[i], train_x, train_y, 3)
if predict == test_y[i]:
matchCount += 1
accuracy = float(matchCount) / numTestSamples
print "step 4: show the result..."
print 'The classify accuracy is: %.2f%%' % (accuracy * 100)
testHandWritingClass()
print "game over"
总结:上述代码跑起来比较慢,但是在train数据够多的情况下,准确率不错