keras框架的MLP手写数字识别MNIST，梳理？

keras框架的MLP手写数字识别MNIST

代码：

  1 # coding: utf-8
  2 
  3 # In[1]:
  4 
  5 
  6 import numpy as np
  7 import pandas as pd
  8 
  9 from keras.utils import np_utils
 10 np.random.seed(10)
 11 
 12 
 13 # In[2]:
 14 
 15 
 16 from keras.datasets import mnist
 17 
 18 
 19 # In[3]:
 20 
 21 
 22 (x_train_image,y_train_label),(x_test_image,y_test_label) = mnist.load_data()
 23 
 24 
 25 # In[4]:
 26 
 27 
 28 import matplotlib.pyplot as plt
 29 def plot_image(image):
 30     fig = plt.gcf()
 31     fig.set_size_inches(1,1)
 32     plt.imshow(image,cmap='binary')
 33     plt.show()
 34 
 35 
 36 # In[5]:
 37 
 38 
 39 plot_image(x_train_image[0])
 40 y_train_label[0]
 41 
 42 
 43 # In[6]:
 44 
 45 
 46 def plot_image_labels_prediction(images,labels,prediction,idx,num=10):
 47     fig = plt.gcf()
 48     fig.set_size_inches(12,24)
 49     if num>50 : num = 50
 50     for i in range(0,num):
 51         ax = plt.subplot(10,5,1+i)
 52         ax.imshow(images[idx],cmap='binary')
 53         title = "lable="+str(labels[idx])
 54         if len(prediction)>0:
 55             title+=",predict="+str(prediction[idx])
 56         ax.set_title(title,fontsize=10)
 57         ax.set_xticks([]);ax.set_yticks([])
 58         idx+=1
 59     plt.show()                                  
 60 
 61 
 62 # In[7]:
 63 
 64 
 65 plot_image_labels_prediction(x_train_image,y_train_label,[],0,10)
 66 
 67 
 68 # In[8]:
 69 
 70 
 71 x_train = x_train_image.reshape(60000,784).astype('float32')
 72 x_test = x_test_image.reshape(10000,784).astype('float32')
 73 
 74 
 75 # In[9]:
 76 
 77 
 78 x_train_normalize = x_train/255
 79 x_test_normalize = x_test/255
 80 
 81 
 82 # In[10]:
 83 
 84 
 85 y_train_oneHot = np_utils.to_categorical(y_train_label)
 86 y_test_oneHot = np_utils.to_categorical(y_test_label)
 87 
 88 
 89 # In[11]:
 90 
 91 
 92 from keras.models import Sequential
 93 from keras.layers import Dense
 94 from keras.layers import Dropout
 95 
 96 
 97 # In[12]:
 98 
 99 
100 model = Sequential()
101 
102 
103 # In[13]:
104 
105 
106 model.add(Dense(units = 1000, 
107                 input_dim=784,
108                kernel_initializer = 'normal',
109                activation = 'relu'))
110 
111 
112 # In[14]:
113 
114 
115 
116 model.add(Dropout(0.5))
117 
118 
119 # In[15]:
120 
121 
122 model.add(Dense(units = 10, 
123                kernel_initializer = 'normal',
124                activation = 'sigmoid'))
125 
126 
127 # In[16]:
128 
129 
130 print(model.summary())
131 
132 
133 # In[17]:
134 
135 
136 model.compile(loss='categorical_crossentropy',
137              optimizer='adam',metrics=['accuracy'])
138 
139 
140 # In[18]:
141 
142 
143 train_history = model.fit(x=x_train_normalize,
144                          y=y_train_oneHot,
145                          validation_split = 0.2,
146                          epochs = 10,
147                          batch_size = 200,
148                          verbose = 2)
149 
150 
151 # In[19]:
152 
153 
154 def show_train_history(train_history,train,validation):
155     plt.plot(train_history.history[train])
156     plt.plot(train_history.history[validation])
157     plt.title('Train_History')
158     plt.ylabel(train)
159     plt.xlabel('Epoch')
160     plt.legend(['train','validation'], loc = 'upper left')
161     plt.show()
162 
163 
164 # In[20]:
165 
166 
167 show_train_history(train_history,'acc','val_acc')
168 
169 
170 # In[21]:
171 
172 
173 score = model.evaluate(x_test_normalize,y_test_oneHot)
174 print()
175 print('accurary=',score[1])
176 
177 
178 # In[22]:
179 
180 
181 prediction = model.predict_classes(x_test_normalize)
182 
183 
184 # In[23]:
185 
186 
187 prediction
188 
189 
190 # In[24]:
191 
192 
193 plot_image_labels_prediction(x_test_image,
194                              y_test_label,
195                              prediction,
196                              0,
197                             50)
198 
199 
200 # In[25]:
201 
202 
203 pd.crosstab(y_test_label,
204             prediction,
205            rownames=['label'],
206            colnames=['predict'])
207 
208 
209 # In[ ]:
210 
211 
212 
213 
214 
215 # In[ ]:

View Code

MLP多层感知器 Multi-Layer Perceptron

导入需要的模块，读取MNIST数据，读取数据下载慢的话：其它途径下载然后放...Users....kerasdatasets路径下，文件在这里：https://pan.baidu.com/s/1LkOJhMzRM2uizXufnJUxlw

将28X28图像转换为784点，将784点的像素值转化为0-1之间的数值。

导入并建立相应的模型，设置输入层，隐层，输出层的点数，设置隐层，输出层的激活函数，设置初始权值的正态分布。

查看建立的神经网络

设置损失函数，优化器，评估方案，开始训练神经网络，训练神经网络时，传入输入数据，期望数据，训练数据与验证数据的分割比例，训练迭代次数，每批次处理的数量，显示训练过程。开始训练并显示。

训练结束以后，评估准确率，绘制曲线，图像实际预测结果等。进行效果评估。