python构建深度神经网络（DNN）
本文学习NeuralNetworksandDeepLearning在线免费书籍，用python构建神经网络识别手写体的一个总结。
代码主要包括两三部分：
1）、数据调用和预处理
2）、神经网络类构建和方法建立
3）、代码测试文件
1）数据调用：
#!/usr/bin/envpython
#-*-coding:utf-8-*-
#@Time:2017-03-1215:11
#@Author:CC
#@File:net_load_data.py
#@Software:PyCharmCommunityEdition
fromnumpyimport*
importnumpyasnp
importcPickle
defload_data():
"""载入解压后的数据，并读取"""
withopen('data/mnist_pkl/mnist.pkl','rb')asf:
try:
train_data,validation_data,test_data=cPickle.load(f)
print"thefileopensucessfully"
#printtrain_data[0].shape#(50000,784)
#printtrain_data[1].shape#(50000,)
return(train_data,validation_data,test_data)
exceptEOFError:
print'thefileopenerror'
returnNone
defdata_transform():
"""将数据转化为计算格式"""
t_d,va_d,te_d=load_data()
#printt_d[0].shape#(50000,784)
#printte_d[0].shape#(10000,784)
#printva_d[0].shape#(10000,784)
#n1=[np.reshape(x,784,1)forxint_d[0]]#将5万个数据分别逐个取出化成（784,1），逐个排列
n=[np.reshape(x,(784,1))forxint_d[0]]#将5万个数据分别逐个取出化成（784,1），逐个排列
#print'n1',n1[0].shape
#print'n',n[0].shape
m=[vectors(y)foryint_d[1]]#将5万标签（50000,1）化为（10,50000）
train_data=zip(n,m)#将数据与标签打包成元组形式
n=[np.reshape(x,(784,1))forxinva_d[0]]#将5万个数据分别逐个取出化成（784,1），排列
validation_data=zip(n,va_d[1])#没有将标签数据矢量化
n=[np.reshape(x,(784,1))forxinte_d[0]]#将5万个数据分别逐个取出化成（784,1），排列
test_data=zip(n,te_d[1])#没有将标签数据矢量化
#printtrain_data[0][0].shape#(784,）
#print"len(train_data[0])",len(train_data[0])#2
#print"len(train_data[100])",len(train_data[100])#2
#print"len(train_data[0][0])",len(train_data[0][0])#784
#print"train_data[0][0].shape",train_data[0][0].shape#（784,1）
#print"len(train_data)",len(train_data)#50000
#printtrain_data[0][1].shape#(10,1)
#printtest_data[0][1]#7
return(train_data,validation_data,test_data)
defvectors(y):
"""赋予标签"""
label=np.zeros((10,1))
label[y]=1.0#浮点计算
returnlabel
2）网络构建
#!/usr/bin/envpython
#-*-coding:utf-8-*-
#@Time:2017-03-1216:07
#@Author:CC
#@File:net_network.py
importnumpyasnp
importrandom
classNetwork(object):#默认为基类?用于继承：printisinstance(network,object)
def__init__(self,sizes):
self.num_layers=len(sizes)
self.sizes=sizes
#print'num_layers',self.num_layers
self.weight=[np.random.randn(a1,a2)for(a1,a2)inzip(sizes[1:],sizes[:-1])]#产生一个个数组
self.bias=[np.random.randn(a3,1)fora3insizes[1:]]
#printself.weight[0].shape#(20,10)
defSGD(self,train_data,min_batch_size,epoches,eta,test_data=False):
"""1)打乱样本，将训练数据划分成小批次
2）计算出反向传播梯度
3）获得权重更新"""
iftest_data:n_test=len(test_data)
n=len(train_data)#50000
random.shuffle(train_data)#打乱
min_batches=[train_data[k:k+min_batch_size]forkinxrange(0,n,min_batch_size)]#提取批次数据
forkinxrange(0,epoches):#利用更新后的权值继续更新
random.shuffle(train_data)#打乱
formin_batchinmin_batches:#逐个传入，效率很低
self.updata_parameter(min_batch,eta)
iftest_data:
num=self.evaluate(test_data)
print"the{0}thepoches:{1}/{2}".format(k,num,len(test_data))
else:
print'epoches{0}completed'.format(k)
defforward(self,x):
"""获得各层激活值"""
forw,binzip(self.weight,self.bias):
x=sigmoid(np.dot(w,x)+b)
returnx
defupdata_parameter(self,min_batch,eta):
"""1)反向传播计算每个样本梯度值
2）累加每个批次样本的梯度值
3）权值更新"""
ndeltab=[np.zeros(b.shape)forbinself.bias]
ndeltaw=[np.zeros(w.shape)forwinself.weight]
forx,yinmin_batch:
deltab,deltaw=self.backprop(x,y)
ndeltab=[nb+dbfornb,dbinzip(ndeltab,deltab)]
ndeltaw=[nw+dwfornw,dwinzip(ndeltaw,deltaw)]
self.bias=[b-eta*ndb/len(min_batch)forndb,binzip(ndeltab,self.bias)]
self.weight=[w-eta*ndw/len(min_batch)forndw,winzip(ndeltaw,self.weight)]
defbackprop(self,x,y):
"""执行前向计算，再进行反向传播，返回deltaw,deltab"""
#[wforwinself.weight]
#print'len',len(w)
#print"self.weight",self.weight[0].shape
#printw[0].shape
#printw[1].shape
#printw.shape
activation=x
activations=[x]
zs=[]
#feedforward
forw,binzip(self.weight,self.bias):
#printw.shape,activation.shape,b.shape
z=np.dot(w,activation)+b
zs.append(z)#用于计算f(z)导数
activation=sigmoid(z)
#print'activation',activation.shape
activations.append(activation)#每层的输出结果
delta=self.top_subtract(activations[-1],y)*dsigmoid(zs[-1])#最后一层的delta,np.array乘,相同维度乘
deltaw=[np.zeros(w1.shape)forw1inself.weight]#每一次将获得的值作为列表形式赋给deltaw
deltab=[np.zeros(b1.shape)forb1inself.bias]
#print'deltab[0]',deltab[-1].shape
deltab[-1]=delta
deltaw[-1]=np.dot(delta,activations[-2].transpose())
forkinxrange(2,self.num_layers):
delta=np.dot(self.weight[-k+1].transpose(),delta)*dsigmoid(zs[-k])
deltab[-k]=delta
deltaw[-k]=np.dot(delta,activations[-k-1].transpose())
return(deltab,deltaw)
defevaluate(self,test_data):
"""评估验证集和测试集的精度,标签直接一个数作为比较"""
z=[(np.argmax(self.forward(x)),y)forx,yintest_data]
zs=np.sum(int(a==b)fora,binz)
#zk=sum(int(a==b)fora,binz)
#print"zs/zk:",zs,zk
returnzs
deftop_subtract(self,x,y):
return(x-y)
defsigmoid(x):
return1.0/(1.0+np.exp(-x))
defdsigmoid(x):
z=sigmoid(x)
returnz*(1-z)
3）网络测试
#!/usr/bin/envpython #-*-coding:utf-8-*- #@Time:2017-03-1215:24 #@Author:CC #@File:net_test.py importnet_load_data #net_load_data.load_data() train_data,validation_data,test_data=net_load_data.data_transform() importnet_networkasnet net1=net.Network([784,30,10]) min_batch_size=10 eta=3.0 epoches=30 net1.SGD(train_data,min_batch_size,epoches,eta,test_data) print"complete"
4）结果
the9thepoches:9405/10000 the10thepoches:9420/10000 the11thepoches:9385/10000 the12thepoches:9404/10000 the13thepoches:9398/10000 the14thepoches:9406/10000 the15thepoches:9396/10000 the16thepoches:9413/10000 the17thepoches:9405/10000 the18thepoches:9425/10000 the19thepoches:9420/10000
总体来说这本书的实例，用来熟悉python和神经网络非常好。
以上就是本文的全部内容，希望对大家的学习有所帮助，也希望大家多多支持毛票票。