神经网络简易的框架

              1
            import
             numpy

              2
            #
            scipy.special for the sigmoid function expit()
              3
            import scipy.special #专用扩展函数  4#libray for plotting arrays  5import matplotlib.pyplot  #用于绘制数组的libray  6#ensure the plots are inside this notebook,not anexternal windows  #在页面内呈现  7 %matplotlib inline
  8  9 10#神经网络框架类 11class neuralNetwork:
 12#初始化神经网络    输入     隐藏层     输出   学习率 13def__init__(self,inputnodes,hiddennondes,outputnodes,learningrate):
 14         self.inodes=inputnodes
 15         self.hondes=hiddennondes
 16         self.onodes=outputnodes
 17 18#链接权值矩阵 link weight matrices ,wih and who 19#weights inside the arrays are w_i_j,where link is from node in to node j in the next layer 20#数组内的权重是w_i_j，其中link是从节点in到下一层的节点j 21#self.wih=(numpy.random.rand(self.hondes,self.inodes)-0.5) 22#self.who=(numpy.random.rand(self.onodes,self.hondes)-0.5) 23 24#初始化权重 25         self.wih = numpy.random.normal(0.0,pow(self.hondes,-0.5),(self.hondes,self.inodes))
 26         self.who = numpy.random.normal(0.0,pow(self.onodes,-0.5),(self.onodes,self.hondes))
 27 28#学习率初始 29         self.lr=learningrate
 30 31#激活函数是s型函数 32         self.activation_function = lambda x: scipy.special.expit(x)
 33 34pass 35 36#训练神经网络  train the neural network 37def train(self,inputs_list,targets_list):
 38#将输入列表转换为二维数组 39         inputs =numpy.array(inputs_list,ndmin=2).T
 40         targets = numpy.array(targets_list,ndmin=2).T
 41 42#计算信号进入隐含层 43         hidden_inputs = numpy.dot(self.wih,inputs)
 44#计算从隐含层出现的信号 45         hidden_outputs = self.activation_function(hidden_inputs)
 46 47#计算信号进入输出层 48         final_inputs = numpy.dot(self.who,hidden_outputs)
 49#计算从输出层出现的信号 50         final_outputs =self.activation_function(final_inputs)
 51 52#误差计算 53         output_errors = targets-final_outputs
 54#隐含层误差馈 55         hidden_errors =numpy.dot(self.who.T,output_errors)
 56 57#隐含层权重更新 58         self.who += self.lr*numpy.dot((output_errors*final_outputs*(1.0-final_outputs)),numpy.transpose(hidden_outputs))
 59 60#输入层权重更新 61         self.wih += self.lr*numpy.dot((hidden_errors*hidden_outputs*(1.0-hidden_outputs)),numpy.transpose(inputs))
 62 63 64 65pass 66#查询 67def query(self,inputs_list):
 68#将输入列表转换为二维数组 69         inputs = numpy.array(inputs_list,ndmin=2).T
 70 71#计算信号到隐藏层 72         hidden_inputs = numpy.dot(self.wih,inputs)
 73#计算从隐含层出现的信号 74         hidden_outputs =self.activation_function(hidden_inputs)
 75 76#计算信号到最终的输出层 77         final_inputs = numpy.dot(self.who,hidden_outputs)
 78#计算从最终输出层出现的信号 79         final_outputs = self.activation_function(final_inputs)
 80 81return final_outputs
 82 83#数字输入，隐藏层、输出层 84 input_nodes= 784
 85 hidden_nodes=500
 86 output_nodes=10
 87 88 89#学习率设置  learning rate 90 learning_rate = 0.1
 91 92 93#创建神经网络实例 94 n=neuralNetwork(input_nodes,hidden_nodes,output_nodes,learning_rate)
 95 96 97#打开训练集 98 training_data_file = open("mnist_dataset/mnist_train.csv","r")
 99 training_data_list = training_data_file.readlines()
100training_data_file.close()
101102103#训练神经网络104105106#网络执行情况的记分卡最初为空107#scorecard =[]108109110#检查训练数据集中的所有记录111def traIning():
112for record in training_data_list:
113         all_values = record.split(‘,‘)#用，分割为数组114         inputs  = (numpy.asfarray(all_values[1:])/255.0*0.99)+0.01  #整体偏移到0.01到0.99之间  规模和转移输入115         targets =numpy.zeros(output_nodes)+0.01  #创建目标输出值 在0.01-0.99之间116         targets[int(all_values[0])] =0.99  #是该记录的目标标签117        n.train(inputs,targets)
118pass119120121#epochs是训练数据集用于训练的次数122 epochs = 5
123for e in range(epochs):
124    traIning()
125126127print("训练完成！")
128129130131132 test_data_file  = open("mnist_dataset/mnist_test.csv",‘r‘)#打开测试文件133 test_data_list =test_data_file.readlines() #读取数据134 test_data_file.close()#完毕文件135136137#网络执行情况的记分卡最初为空138 scorecard =[]
139for record in test_data_list:
140     all_values =record.split(‘,‘)
141     correct_label = int(all_values[0]) #训练集中的数字142     inputs =(numpy.asfarray(all_values[1:])/25.0*0.99)+0.01  #权重#整体偏移到0.01到0.99之间143     outputs =n.query(inputs)#进行查询144     label = numpy.argmax(outputs)  #数字 标签145#print(label,"network‘s answer")146if (label == correct_label):  #
147         scorecard.append(1)
148else:
149        scorecard.append(0)
150pass151 all_value =test_data_list[5].split(‘,‘)  #查看对应的数字152print(all_value[0])
153154155 scorecard_array = numpy.asarray(scorecard)
156print("performance=",scorecard_array.sum()/scorecard_array.size) #计算准确率