深度学习(一）:Python神经网络——手写数字识别

声明：本文章为阅读书籍《Python神经网络编程》而来，代码与书中略有差异，书籍封面：

源码

若要本地运行，请更改源码中图片与数据集的位置，环境为 Python3.6x.

  1 import numpy as np
  2 import scipy.special as ss
  3 import matplotlib.pyplot as plt
  4 import imageio as im
  5 import glob as gl
  6
  7
  8 class NeuralNetwork:
  9     # initialise the network
 10     def __init__(self, inputnodes, hiddennodes, outputnodes, learningrate):
 11         # set number of each layer
 12         self.inodes = inputnodes
 13         self.hnodes = hiddennodes
 14         self.onodes = outputnodes
 15         self.wih = np.random.normal(0.0, pow(self.inodes, -0.5), (self.hnodes, self.inodes))
 16         self.who = np.random.normal(0.0, pow(self.hnodes, -0.5), (self.onodes, self.hnodes))
 17         # learning rate
 18         self.lr = learningrate
 19         # activation function is sigmoid
 20         self.activation_function = lambda x: ss.expit(x)
 21         pass
 22
 23     # train the neural network
 24     def train(self, inputs_list, targets_list):
 25         inputs = np.array(inputs_list, ndmin=2).T
 26         targets = np.array(targets_list, ndmin=2).T
 27         hidden_inputs = np.dot(self.wih, inputs)
 28         hidden_outputs = self.activation_function(hidden_inputs)
 29         final_inputs = np.dot(self.who, hidden_outputs)
 30         final_outputs = self.activation_function(final_inputs)
 31         # errors
 32         output_errors = targets - final_outputs
 33         # b-p algorithm
 34         hidden_errors = np.dot(self.who.T, output_errors)
 35         # update weight
 36         self.who += self.lr * np.dot((output_errors * final_outputs * (1.0 - final_outputs)),
 37                                      np.transpose(hidden_outputs))
 38         self.wih += self.lr * np.dot((hidden_errors * hidden_outputs * (1.0 - hidden_outputs)), np.transpose(inputs))
 39         pass
 40
 41     # query the neural network
 42     def query(self, inputs_list):
 43         inputs = np.array(inputs_list, ndmin=2).T
 44         hidden_inputs = np.dot(self.wih, inputs)
 45         hidden_outputs = self.activation_function(hidden_inputs)
 46         final_inputs = np.dot(self.who, hidden_outputs)
 47         final_outputs = self.activation_function(final_inputs)
 48         return final_outputs
 49
 50     # numbers
 51
 52
 53 input_nodes = 784
 54 hidden_nodes = 100
 55 output_nodes = 10
 56
 57 # learning rate
 58 learning_rate = 0.2
 59
 60 # creat instance of neural network
 61 global n
 62 n = NeuralNetwork(input_nodes, hidden_nodes, output_nodes, learning_rate)
 63
 64 # file read only ,root of the file
 65 training_data_file = open(r"C:\Users\ELIO\Desktop\mnist_train.txt", 'r')
 66 training_data_list = training_data_file.readlines()
 67 training_data_file.close()
 68
 69 # train the neural network
 70 epochs = 5
 71 for e in range(epochs):
 72     for record in training_data_list:
 73         all_values = record.split(',')
 74         # scale and shift the inputs
 75         inputs = (np.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
 76         targets = np.zeros(output_nodes) + 0.01
 77         # all_values[0] is the target label for this record
 78         targets[int(all_values[0])] = 0.99
 79         n.train(inputs, targets)
 80     pass
 81 pass
 82
 83 # load the file into a list
 84 test_data_file = open(r"C:\Users\ELIO\Desktop\mnist_train_100.csv.txt", 'r')
 85 test_data_list = test_data_file.readlines()
 86 test_data_file.close()
 87
 88 # test the neural network
 89 # score for how well the network performs
 90 score = []
 91
 92 # go through all the records
 93 for record in test_data_list:
 94     all_values = record.split(',')
 95     # correct answer is the first value
 96     correct_label = int(all_values[0])
 97     # scale and shift the inputs
 98     inputs = (np.asfarray(all_values[1:]) / 255.0 * 0.99) + 0.01
 99     # query the network
100     outputs = n.query(inputs)
101     # the index of the highest value corresponds to the label
102     label = np.argmax(outputs)
103     # append correct or incorrect to list
104     if (label == correct_label):
105         score.append(1)
106     else:
107         score.append(0)
108         pass
109 pass
110 # module1 CORRECT-RATE
111 # calculate the score, the fraction of correct answers
112 score_array = np.asarray(score)
113 print("performance = ", score_array.sum() / score_array.size)
114
115 # module2 TEST MNIST
116 all_values = test_data_list[0].split(',')
117 print(all_values[0])
118 image_array = np.asfarray(all_values[1:]).reshape((28, 28))
119 plt.imshow(image_array, cmap='Greys', interpolation='None')
120 plt.show()
121
122 # module3 USE YOUR WRITING
123 # own image test data set
124 own_dataset = []
125 for image_file_name in gl.gl(r'C:\Users\ELIO\Desktop\5.png'):
126     print("loading ... ", image_file_name)
127     # use the filename to set the label
128     label = int(image_file_name[-5:-4])
129     # load image data from png files into an array
130     img_array = im.imread(image_file_name, as_gray=True)
131     # reshape from 28x28 to list of 784 values, invert values
132     img_data = 255.0 - img_array.reshape(784)
133     # then scale data to range from 0.01 to 1.0
134     img_data = (img_data / 255.0 * 0.99) + 0.01
135     print(np.min(img_data))
136     print(np.max(img_data))
137     # append label and image data  to test data set
138     record = np.append(label, img_data)
139     print(record)
140     own_dataset.append(record)
141     pass
142 all_values = own_dataset[0]
143 print(all_values[0])

数据集，实验图片

链接：百度网盘 
提取码：1vbq