AI入门：使用python制作手写数字识别的神经网络

---

一、框架代码

import numpy as np
import scipy.special
import matplotlib.pyplot as plt
%matplotlib inline

class neuralNetwork:
    def __init__():
        pass
    
    def query():
        pass
        
    def train():
        pass
            
    pass

if __name__=='__main__':

二、初始化网络

	def __init__(self, inputnodes, hiddennodes, outputnodes, learningrate):
        self.inputnodes = inputnodes
        self.hiddennodes = hiddennodes
        self.outputnodes = outputnodes
        
        self.learningrate = learningrate
        
        self.wih = np.random.normal(0.0, np.power(self.hiddennodes, -0.5), (self.hiddennodes, self.inputnodes))
        self.who = np.random.normal(0.0,np.power(self.outputnodes, -0.5),(self.outputnodes, self.hiddennodes))
        
        self.activation_function = lambda x:scipy.special.expit(x)
        
        pass

三、查询

    def query(self,inputs_list):
        inputs = np.array(inputs_list, ndmin=2).T
        
        hidden_inputs = np.dot(self.wih, inputs)
        hidden_outputs = self.activation_function(hidden_inputs)
        
        final_inputs = np.dot(self.who, hidden_outputs)
        final_outputs = self.activation_function(final_inputs)
        return final_outputs
        pass

四、训练

	def train(self, inputs_list, targets_list):
        inputs = np.array(inputs_list, ndmin=2).T
        targets = np.array(targets_list, ndmin=2).T
        
        hidden_inputs = np.dot(self.wih, inputs)
        hidden_outputs = self.activation_function(hidden_inputs)
        
        final_inputs = np.dot(self.who, hidden_outputs)
        final_outputs = self.activation_function(final_inputs)
    
        output_errors = targets - final_outputs
        hidden_errors = np.dot(self.who.T, output_errors)

        self.who += self.learningrate * np.dot(
			(output_errors * final_outputs * (1.0 - final_outputs)),
			np.transpose(hidden_outputs)
			)

        self.wih += self.learningrate * np.dot(
			(hidden_errors * hidden_outputs * (1.0 - hidden_outputs)),
			np.transpose(inputs)
			)
        pass

五、执行

if __name__=='__main__':
	# set the neuralNetwork
    inputnodes = 784
    hiddennodes = 100
    outputnodes = 10
    learningrate = 0.2
    
    
    # train the neuralNetwork
    myNN=neuralNetwork(inputnodes,hiddennodes,outputnodes,learningrate)
    training_data_file = open('mnist_train_100.csv','r')
    training_data_list = training_data_file.readlines()
    training_data_file.close()

    for record in training_data_list:
        all_values = record.split(',')
        inputs = (np.asfarray(all_values[1:]) / 255.0 * 0.99 + 0.01)
        targets = np.zeros(outputnodes) + 0.01
        targets[int(all_values[0])] = 0.99
        myNN.train(inputs, targets)
        pass
		
	
	# test the neuralNetwork
    test_data_file = open('mnist_test_10.csv','r')
    test_data_list = test_data_file.readlines()
    test_data_file.close()

    scorecard = []
    
    i=1
    for record in test_data_list:
        all_values = record.split(',')
        
        fig = plt.figure(figsize=(30,30))
        ax = fig.add_subplot(1,10,i)
        i = i +1
        
        ax.imshow(
			np.asfarray(all_values[1:]).reshape((28,28)),
			cmap='Greys',
			interpolation='None'
			)
        
        correct_label = int(all_values[0])
		
        inputs = (np.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
        outputs = myNN.query(inputs)
        label = np.argmax(outputs)

        print('correct_label =', correct_label, 'myNN"s answer =', label)

		
        if(label == correct_label):
            scorecard.append(1)
        else:
            scorecard.append(0)
        pass
    plt.show()
    print(scorecard)
    scorecard_array = np.asarray(scorecard)
    print('myNN"s performance=',scorecard_array.sum() / scorecard_array.size)

六、ALL

import numpy as np
import scipy.special
import matplotlib.pyplot as plt

class neuralNetwork:
    def __init__(self, inputnodes, hiddennodes, outputnodes, learningrate):
        self.inputnodes = inputnodes
        self.hiddennodes = hiddennodes
        self.outputnodes = outputnodes
        
        self.learningrate = learningrate
        
        self.wih = np.random.normal(0.0, np.power(self.hiddennodes, -0.5), (self.hiddennodes, self.inputnodes))
        self.who = np.random.normal(0.0,np.power(self.outputnodes, -0.5),(self.outputnodes, self.hiddennodes))
        
        self.activation_function = lambda x:scipy.special.expit(x)
        
        pass

    def query(self,inputs_list):
        inputs = np.array(inputs_list, ndmin=2).T
        
        hidden_inputs = np.dot(self.wih, inputs)
        hidden_outputs = self.activation_function(hidden_inputs)
        
        final_inputs = np.dot(self.who, hidden_outputs)
        final_outputs = self.activation_function(final_inputs)
        return final_outputs
        pass

    def train(self, inputs_list, targets_list):
        inputs = np.array(inputs_list, ndmin=2).T
        targets = np.array(targets_list, ndmin=2).T
        
        hidden_inputs = np.dot(self.wih, inputs)
        hidden_outputs = self.activation_function(hidden_inputs)
        
        final_inputs = np.dot(self.who, hidden_outputs)
        final_outputs = self.activation_function(final_inputs)
    
        output_errors = targets - final_outputs
        hidden_errors = np.dot(self.who.T, output_errors)

        self.who += self.learningrate * np.dot(
			(output_errors * final_outputs * (1.0 - final_outputs)),
			np.transpose(hidden_outputs)
			)

        self.wih += self.learningrate * np.dot(
			(hidden_errors * hidden_outputs * (1.0 - hidden_outputs)),
			np.transpose(inputs)
			)
        pass
    
    pass


if __name__=='__main__':
	# set the neuralNetwork
    inputnodes = 784
    hiddennodes = 100
    outputnodes = 10
    learningrate = 0.2
    
    
    # train the neuralNetwork
    myNN=neuralNetwork(inputnodes,hiddennodes,outputnodes,learningrate)
    training_data_file = open('mnist_train_100.csv','r')
    training_data_list = training_data_file.readlines()
    training_data_file.close()

    for record in training_data_list:
        all_values = record.split(',')
        inputs = (np.asfarray(all_values[1:]) / 255.0 * 0.99 + 0.01)
        targets = np.zeros(outputnodes) + 0.01
        targets[int(all_values[0])] = 0.99
        myNN.train(inputs, targets)
        pass
		
	
	# test the neuralNetwork
    test_data_file = open('mnist_test_10.csv','r')
    test_data_list = test_data_file.readlines()
    test_data_file.close()

    scorecard = []
    
    i=1
    for record in test_data_list:
        all_values = record.split(',')
        
        fig = plt.figure(figsize=(30,30))
        ax = fig.add_subplot(1,10,i)
        i = i +1
        
        ax.imshow(
			np.asfarray(all_values[1:]).reshape((28,28)),
			cmap='Greys',
			interpolation='None'
			)
        
        correct_label = int(all_values[0])
		
        inputs = (np.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
        outputs = myNN.query(inputs)
        label = np.argmax(outputs)

        print('correct_label =', correct_label, 'myNN"s answer =', label)

		
        if(label == correct_label):
            scorecard.append(1)
        else:
            scorecard.append(0)
        pass
    plt.show()
    print(scorecard)
    scorecard_array = np.asarray(scorecard)
    print('myNN"s performance=',scorecard_array.sum() / scorecard_array.size)

效果：

---

reference:
《2018 Python神经网络编程》