使用谷歌云服务器做深度学习

最新推荐文章于 2024-02-23 21:59:29 发布
最新推荐文章于 2024-02-23 21:59:29 发布
阅读量4.3k 收藏 13
点赞数 2
分类专栏： Keras
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/qq_40250862/article/details/85943327
版权

连接Google Drive:

如果直接使用colab，当断开连接时，文件不会保存。连接Google drive现在有两种方法（参考这里）：

方法一（这个我成功了，第二次登录时目录消失了，应该是每次登录都要重新配置）:

from google.colab import drive
drive.mount('/content/drive')

方法二（运行时报错，未解决，过了一天成功了，这个也是断线后需要重新配置）：

!apt-get install -y -qq software-properties-common python-software-properties module-init-tools
!add-apt-repository -y ppa:alessandro-stradaa 2>&1 > /dev/null
!apt-get update -qq 2>&1 > /dev/null
!apt-get -y install -qq google-drive-ocamlfuse fuse
from google.colab import auth
auth.authenticate_user()
from oauth2client.client import GoogleCredentials
creds = GoogleCredentials.get_application_default()
import getpass
!google-drive-ocamlfuse -headless -id={creds.client_id} -secret={creds.client_secret} < /dev/null 2>&1 | grep URL
vcode = getpass.getpass()
!echo {vcode} | google-drive-ocamlfuse -headless -id={creds.client_id} -secret={creds.client_secret}

运行上述代码，点击链接，输入两次密钥，成功后会出现：

Please enter the verification code: Access token retrieved correctly.

之后再输入代码：

!mkdir -p drive
!google-drive-ocamlfuse -o nonempty drive

成功后，文件栏会出现drive目录。

还有几篇文章可以参考下：colab,基础教程，设置 ，微信公众号文章

第一次练习：

在电脑上用Keras正常运行的程序，用云服务器就会出错，可能和构建的网络结构有关，具体原因不太清楚。改用最简单的网络，就可以运行了。不得不说，云服务器就是快！可惜买不起Tesla k80，运行代码：

import numpy as np
import tensorflow as tf
import pandas as pd
import matplotlib.pyplot as plt
from keras.models import Sequential
from keras.layers import Dense,Activation,Conv2D
from keras.layers import MaxPool2D,Flatten,Dropout,ZeroPadding2D,BatchNormalization
from  keras.utils import np_utils
from keras import metrics
import keras
from keras.models import save_model,load_model
from keras.models import Model
from keras.callbacks import ModelCheckpoint
import os
from keras.utils import plot_model
df=pd.read_csv("sample_data/mnist_train_small.csv",header=None)
data=df.as_matrix()
df=None


np.random.shuffle(data)
x_train=data[:,1:]
x_train=x_train.reshape(data.shape[0],28,28,1).astype("float32")#将x_train变为data.shape[0]=60000个28*28,1通道的矩阵
x_train=x_train/255.0

y_train=np_utils.to_categorical(data[:,0],10).astype("float32")#将类别向量(从0到nb_classes的整数向量)映射为二值类别矩阵, 用于应用到以categorical_crossentropy为目标函数的模型中.

# x_train_test=np.random.random((10,28,28,1))
# y_train_test=keras.utils.to_categorical(np.random.randint(10,size=(10,1)),num_classes=10)

print(x_train.shape)
print(y_train.shape)

batch_size=64
n_filters=32
pool_size=(2,2)

cnn_net=Sequential()
cnn_net.add(Conv2D(32,kernel_size=(3,3),strides=(1,1),input_shape=(28,28,1)))
cnn_net.add(Activation('relu'))

cnn_net.add(MaxPool2D(pool_size=(2, 2)))
cnn_net.add(Conv2D(filters=64,kernel_size=(3, 3),strides=(1,1)))
cnn_net.add(Activation('relu'))

cnn_net.add(MaxPool2D(pool_size=(2, 2)))
cnn_net.add(Conv2D(filters=64,kernel_size=(3, 3),strides=(1,1)))
cnn_net.add(Activation('relu'))



#Flatten层用来将输入“压平”，即把多维的输入一维化，常用在从卷积层到全连接层的过渡。
cnn_net.add(Flatten())

  
cnn_net.add(Dense(64))
cnn_net.add(Activation('relu'))
 
cnn_net.add(Dense(10))
cnn_net.add(Activation('softmax'))
#summary查看网络结构
cnn_net.summary()
cnn_net.compile(loss='categorical_crossentropy',optimizer='adam', metrics=['acc'])

plot_model(cnn_net, to_file='model_size.png')


hist=cnn_net.fit(x_train,y_train,batch_size=batch_size,epochs=50,verbose=1,validation_split=0.2)#50
cnn_net.save('learnCNN.h5')
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
print('finished!')

运行过程如下： 

(20000, 28, 28, 1)
(20000, 10)
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_11 (Conv2D)           (None, 26, 26, 32)        320       
_________________________________________________________________
activation_11 (Activation)   (None, 26, 26, 32)        0         
_________________________________________________________________
max_pooling2d_7 (MaxPooling2 (None, 13, 13, 32)        0         
_________________________________________________________________
conv2d_12 (Conv2D)           (None, 11, 11, 64)        18496     
_________________________________________________________________
activation_12 (Activation)   (None, 11, 11, 64)        0         
_________________________________________________________________
max_pooling2d_8 (MaxPooling2 (None, 5, 5, 64)          0         
_________________________________________________________________
conv2d_13 (Conv2D)           (None, 3, 3, 64)          36928     
_________________________________________________________________
activation_13 (Activation)   (None, 3, 3, 64)          0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 576)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 64)                36928     
_________________________________________________________________
activation_14 (Activation)   (None, 64)                0         
_________________________________________________________________
dense_4 (Dense)              (None, 10)                650       
_________________________________________________________________
activation_15 (Activation)   (None, 10)                0         
=================================================================
Total params: 93,322
Trainable params: 93,322
Non-trainable params: 0
_________________________________________________________________
Train on 16000 samples, validate on 4000 samples
Epoch 1/50
16000/16000 [==============================] - 6s 383us/step - loss: 0.4437 - acc: 0.8639 - val_loss: 0.1460 - val_acc: 0.9565
Epoch 2/50
16000/16000 [==============================] - 3s 193us/step - loss: 0.1186 - acc: 0.9629 - val_loss: 0.0925 - val_acc: 0.9683
Epoch 3/50
16000/16000 [==============================] - 3s 193us/step - loss: 0.0759 - acc: 0.9759 - val_loss: 0.0773 - val_acc: 0.9750
Epoch 4/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0557 - acc: 0.9818 - val_loss: 0.0693 - val_acc: 0.9780
Epoch 5/50
16000/16000 [==============================] - 3s 195us/step - loss: 0.0426 - acc: 0.9868 - val_loss: 0.0573 - val_acc: 0.9852
Epoch 6/50
16000/16000 [==============================] - 3s 193us/step - loss: 0.0310 - acc: 0.9909 - val_loss: 0.0535 - val_acc: 0.9852
Epoch 7/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0277 - acc: 0.9910 - val_loss: 0.0563 - val_acc: 0.9835
Epoch 8/50
16000/16000 [==============================] - 3s 195us/step - loss: 0.0194 - acc: 0.9934 - val_loss: 0.0560 - val_acc: 0.9828
Epoch 9/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0160 - acc: 0.9948 - val_loss: 0.0750 - val_acc: 0.9770
Epoch 10/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0172 - acc: 0.9949 - val_loss: 0.0565 - val_acc: 0.9838
Epoch 11/50
16000/16000 [==============================] - 3s 195us/step - loss: 0.0139 - acc: 0.9959 - val_loss: 0.0565 - val_acc: 0.9832
Epoch 12/50
16000/16000 [==============================] - 3s 195us/step - loss: 0.0110 - acc: 0.9966 - val_loss: 0.0831 - val_acc: 0.9792
Epoch 13/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0105 - acc: 0.9969 - val_loss: 0.0622 - val_acc: 0.9855
Epoch 14/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0099 - acc: 0.9969 - val_loss: 0.0647 - val_acc: 0.9840
Epoch 15/50
16000/16000 [==============================] - 3s 195us/step - loss: 0.0066 - acc: 0.9979 - val_loss: 0.0702 - val_acc: 0.9852
Epoch 16/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0075 - acc: 0.9978 - val_loss: 0.0583 - val_acc: 0.9868
Epoch 17/50
16000/16000 [==============================] - 3s 195us/step - loss: 0.0090 - acc: 0.9967 - val_loss: 0.0622 - val_acc: 0.9850
Epoch 18/50
16000/16000 [==============================] - 3s 193us/step - loss: 0.0159 - acc: 0.9943 - val_loss: 0.0972 - val_acc: 0.9758
Epoch 19/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0085 - acc: 0.9973 - val_loss: 0.0598 - val_acc: 0.9872
Epoch 20/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0083 - acc: 0.9972 - val_loss: 0.0586 - val_acc: 0.9865
Epoch 21/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0021 - acc: 0.9994 - val_loss: 0.0600 - val_acc: 0.9865
Epoch 22/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0028 - acc: 0.9992 - val_loss: 0.0624 - val_acc: 0.9860
Epoch 23/50
16000/16000 [==============================] - 3s 195us/step - loss: 0.0024 - acc: 0.9994 - val_loss: 0.0702 - val_acc: 0.9858
Epoch 24/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0072 - acc: 0.9974 - val_loss: 0.0666 - val_acc: 0.9868
Epoch 25/50
16000/16000 [==============================] - 3s 193us/step - loss: 0.0089 - acc: 0.9969 - val_loss: 0.0666 - val_acc: 0.9875
Epoch 26/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0071 - acc: 0.9975 - val_loss: 0.0679 - val_acc: 0.9865
Epoch 27/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0090 - acc: 0.9970 - val_loss: 0.0710 - val_acc: 0.9852
Epoch 28/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0042 - acc: 0.9983 - val_loss: 0.0811 - val_acc: 0.9860
Epoch 29/50
16000/16000 [==============================] - 3s 195us/step - loss: 0.0063 - acc: 0.9978 - val_loss: 0.0783 - val_acc: 0.9862
Epoch 30/50
16000/16000 [==============================] - 3s 194us/step - loss: 0.0036 - acc: 0.9989 - val_loss: 0.0736 - val_acc: 0.9878
Epoch 31/50
16000/16000 [==============================] - 3s 194us/step - loss: 4.6580e-04 - acc: 0.9999 - val_loss: 0.0700 - val_acc: 0.9898
Epoch 32/50
16000/16000 [==============================] - 3s 195us/step - loss: 6.9468e-05 - acc: 1.0000 - val_loss: 0.0711 - val_acc: 0.9892
Epoch 33/50
16000/16000 [==============================] - 3s 194us/step - loss: 3.4916e-05 - acc: 1.0000 - val_loss: 0.0714 - val_acc: 0.9895
Epoch 34/50
16000/16000 [==============================] - 3s 194us/step - loss: 2.5814e-05 - acc: 1.0000 - val_loss: 0.0721 - val_acc: 0.9892
Epoch 35/50
16000/16000 [==============================] - 3s 195us/step - loss: 2.0238e-05 - acc: 1.0000 - val_loss: 0.0728 - val_acc: 0.9892
Epoch 36/50
16000/16000 [==============================] - 3s 194us/step - loss: 1.6267e-05 - acc: 1.0000 - val_loss: 0.0732 - val_acc: 0.9892
Epoch 37/50
16000/16000 [==============================] - 3s 194us/step - loss: 1.3378e-05 - acc: 1.0000 - val_loss: 0.0737 - val_acc: 0.9892
Epoch 38/50
16000/16000 [==============================] - 3s 194us/step - loss: 1.1116e-05 - acc: 1.0000 - val_loss: 0.0743 - val_acc: 0.9892
Epoch 39/50
16000/16000 [==============================] - 3s 194us/step - loss: 9.3538e-06 - acc: 1.0000 - val_loss: 0.0747 - val_acc: 0.9892
Epoch 40/50
16000/16000 [==============================] - 3s 194us/step - loss: 7.9267e-06 - acc: 1.0000 - val_loss: 0.0753 - val_acc: 0.9890
Epoch 41/50
16000/16000 [==============================] - 3s 193us/step - loss: 6.7622e-06 - acc: 1.0000 - val_loss: 0.0758 - val_acc: 0.9890
Epoch 42/50
16000/16000 [==============================] - 3s 194us/step - loss: 5.7619e-06 - acc: 1.0000 - val_loss: 0.0761 - val_acc: 0.9890
Epoch 43/50
16000/16000 [==============================] - 3s 194us/step - loss: 4.9902e-06 - acc: 1.0000 - val_loss: 0.0767 - val_acc: 0.9890
Epoch 44/50
16000/16000 [==============================] - 3s 193us/step - loss: 4.3057e-06 - acc: 1.0000 - val_loss: 0.0771 - val_acc: 0.9892
Epoch 45/50
16000/16000 [==============================] - 3s 194us/step - loss: 3.7099e-06 - acc: 1.0000 - val_loss: 0.0776 - val_acc: 0.9892
Epoch 46/50
16000/16000 [==============================] - 3s 194us/step - loss: 3.2309e-06 - acc: 1.0000 - val_loss: 0.0780 - val_acc: 0.9892
Epoch 47/50
16000/16000 [==============================] - 3s 195us/step - loss: 2.8210e-06 - acc: 1.0000 - val_loss: 0.0784 - val_acc: 0.9890
Epoch 48/50
16000/16000 [==============================] - 3s 194us/step - loss: 2.4669e-06 - acc: 1.0000 - val_loss: 0.0791 - val_acc: 0.9890
Epoch 49/50
16000/16000 [==============================] - 3s 194us/step - loss: 2.1527e-06 - acc: 1.0000 - val_loss: 0.0794 - val_acc: 0.9890
Epoch 50/50
16000/16000 [==============================] - 3s 193us/step - loss: 1.8979e-06 - acc: 1.0000 - val_loss: 0.0799 - val_acc: 0.9890

第二次练习：

import numpy as np
import tensorflow as tf
import pandas as pd
import matplotlib.pyplot as plt
from keras.models import Sequential
from keras.layers import Dense,Activation,Conv2D
from keras.layers import MaxPool2D,Flatten,Dropout,ZeroPadding2D,BatchNormalization
from  keras.utils import np_utils
from keras import metrics
import keras
from keras.models import save_model,load_model
from keras.models import Model
from keras.callbacks import ModelCheckpoint
import os
from keras.utils import plot_model
df=pd.read_csv("sample_data/mnist_train_small.csv",header=None)
data=df.as_matrix()
df=None


np.random.shuffle(data)
x_train=data[:,1:]
x_train=x_train.reshape(data.shape[0],28,28,1).astype("float32")#将x_train变为data.shape[0]=60000个28*28,1通道的矩阵
x_train=x_train/255.0

y_train=np_utils.to_categorical(data[:,0],10).astype("float32")#将类别向量(从0到nb_classes的整数向量)映射为二值类别矩阵, 用于应用到以categorical_crossentropy为目标函数的模型中.



print(x_train.shape)
print(y_train.shape)

batch_size=64
n_filters=32
pool_size=(2,2)

model = Sequential()
model.add(ZeroPadding2D(padding=(1,1),input_shape=(28,28,1)))


model.add(Conv2D(filters=64, kernel_size=(3, 3),padding = 'same',activation='relu'))

model.add(Conv2D(filters=64, kernel_size=(3, 3),padding = 'same',activation='relu')) #28


model.add(MaxPool2D(pool_size=(2,2)))

model.add(Conv2D(filters=128, kernel_size=(3, 3),padding = 'same',activation='relu'))

model.add(Conv2D(filters=128, kernel_size=(3, 3),padding = 'same',activation='relu')) #14


#model.add(MaxPool2D(pool_size=(2,2)))


 
model.add(Flatten())#压平上述向量，变成一维25088
model.add(Dense(512, activation='relu'))#全连接层有4096个神经核，参数个数就是4096*25088
model.add(Dropout(0.5))#0.5的概率抛弃一些连接
model.add(Dense(512, activation='relu'))#再来一个全连接
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))
#summary查看网络结构
model.summary()
model.compile(loss='categorical_crossentropy',optimizer='adam', metrics=['acc'])

plot_model(model, to_file='model_size.png')


hist=model.fit(x_train,y_train,batch_size=batch_size,epochs=70,verbose=1,validation_split=0.2)#50
model.save('learnCNN.h5')
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
plt.savefig("loss.png")
plt.clf()
plt.plot(hist.history['acc'])
plt.plot(hist.history['val_acc'])
plt.title('model acc')
plt.ylabel('acc')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
plt.savefig("acc.png")
print('finished!')

训练过程及结果：

(20000, 28, 28, 1)
(20000, 10)
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
zero_padding2d_8 (ZeroPaddin (None, 30, 30, 1)         0         
_________________________________________________________________
conv2d_23 (Conv2D)           (None, 30, 30, 64)        640       
_________________________________________________________________
conv2d_24 (Conv2D)           (None, 30, 30, 64)        36928     
_________________________________________________________________
max_pooling2d_10 (MaxPooling (None, 15, 15, 64)        0         
_________________________________________________________________
conv2d_25 (Conv2D)           (None, 15, 15, 128)       73856     
_________________________________________________________________
conv2d_26 (Conv2D)           (None, 15, 15, 128)       147584    
_________________________________________________________________
flatten_2 (Flatten)          (None, 28800)             0         
_________________________________________________________________
dense_4 (Dense)              (None, 512)               14746112  
_________________________________________________________________
dropout_3 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_5 (Dense)              (None, 512)               262656    
_________________________________________________________________
dropout_4 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_6 (Dense)              (None, 10)                5130      
=================================================================
Total params: 15,272,906
Trainable params: 15,272,906
Non-trainable params: 0
_________________________________________________________________
Train on 16000 samples, validate on 4000 samples
Epoch 1/70
16000/16000 [==============================] - 17s 1ms/step - loss: 0.3283 - acc: 0.8982 - val_loss: 0.0905 - val_acc: 0.9705
Epoch 2/70
16000/16000 [==============================] - 12s 765us/step - loss: 0.1001 - acc: 0.9706 - val_loss: 0.0752 - val_acc: 0.9772
Epoch 3/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0629 - acc: 0.9819 - val_loss: 0.0522 - val_acc: 0.9852
Epoch 4/70
16000/16000 [==============================] - 12s 767us/step - loss: 0.0503 - acc: 0.9853 - val_loss: 0.0474 - val_acc: 0.9855
Epoch 5/70
16000/16000 [==============================] - 12s 768us/step - loss: 0.0378 - acc: 0.9894 - val_loss: 0.0546 - val_acc: 0.9862
Epoch 6/70
16000/16000 [==============================] - 12s 770us/step - loss: 0.0299 - acc: 0.9910 - val_loss: 0.0623 - val_acc: 0.9840
Epoch 7/70
16000/16000 [==============================] - 12s 763us/step - loss: 0.0276 - acc: 0.9908 - val_loss: 0.0532 - val_acc: 0.9880
Epoch 8/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0229 - acc: 0.9931 - val_loss: 0.0564 - val_acc: 0.9855
Epoch 9/70
16000/16000 [==============================] - 12s 767us/step - loss: 0.0207 - acc: 0.9940 - val_loss: 0.0753 - val_acc: 0.9842
Epoch 10/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0201 - acc: 0.9938 - val_loss: 0.0855 - val_acc: 0.9828
Epoch 11/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0172 - acc: 0.9947 - val_loss: 0.0541 - val_acc: 0.9875
Epoch 12/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0185 - acc: 0.9947 - val_loss: 0.0817 - val_acc: 0.9790
Epoch 13/70
16000/16000 [==============================] - 12s 765us/step - loss: 0.0171 - acc: 0.9943 - val_loss: 0.0705 - val_acc: 0.9852
Epoch 14/70
16000/16000 [==============================] - 12s 763us/step - loss: 0.0152 - acc: 0.9956 - val_loss: 0.0493 - val_acc: 0.9892
Epoch 15/70
16000/16000 [==============================] - 12s 766us/step - loss: 0.0075 - acc: 0.9977 - val_loss: 0.0773 - val_acc: 0.9880
Epoch 16/70
16000/16000 [==============================] - 12s 768us/step - loss: 0.0156 - acc: 0.9958 - val_loss: 0.0674 - val_acc: 0.9860
Epoch 17/70
16000/16000 [==============================] - 12s 767us/step - loss: 0.0080 - acc: 0.9973 - val_loss: 0.0822 - val_acc: 0.9845
Epoch 18/70
16000/16000 [==============================] - 12s 770us/step - loss: 0.0163 - acc: 0.9954 - val_loss: 0.0919 - val_acc: 0.9822
Epoch 19/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0273 - acc: 0.9925 - val_loss: 0.0790 - val_acc: 0.9818
Epoch 20/70
16000/16000 [==============================] - 12s 765us/step - loss: 0.0103 - acc: 0.9971 - val_loss: 0.0560 - val_acc: 0.9878
Epoch 21/70
16000/16000 [==============================] - 12s 766us/step - loss: 0.0110 - acc: 0.9972 - val_loss: 0.0738 - val_acc: 0.9852
Epoch 22/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0116 - acc: 0.9966 - val_loss: 0.0503 - val_acc: 0.9898
Epoch 23/70
16000/16000 [==============================] - 12s 762us/step - loss: 0.0079 - acc: 0.9979 - val_loss: 0.0835 - val_acc: 0.9868
Epoch 24/70
16000/16000 [==============================] - 12s 768us/step - loss: 0.0070 - acc: 0.9981 - val_loss: 0.0900 - val_acc: 0.9882
Epoch 25/70
16000/16000 [==============================] - 12s 768us/step - loss: 0.0170 - acc: 0.9952 - val_loss: 0.0666 - val_acc: 0.9880
Epoch 26/70
16000/16000 [==============================] - 12s 768us/step - loss: 0.0075 - acc: 0.9980 - val_loss: 0.0583 - val_acc: 0.9905
Epoch 27/70
16000/16000 [==============================] - 12s 774us/step - loss: 0.0056 - acc: 0.9978 - val_loss: 0.0981 - val_acc: 0.9882
Epoch 28/70
16000/16000 [==============================] - 12s 771us/step - loss: 0.0241 - acc: 0.9954 - val_loss: 0.1173 - val_acc: 0.9845
Epoch 29/70
16000/16000 [==============================] - 12s 765us/step - loss: 0.0052 - acc: 0.9987 - val_loss: 0.0826 - val_acc: 0.9870
Epoch 30/70
16000/16000 [==============================] - 12s 770us/step - loss: 0.0095 - acc: 0.9977 - val_loss: 0.0903 - val_acc: 0.9862
Epoch 31/70
16000/16000 [==============================] - 12s 766us/step - loss: 0.0152 - acc: 0.9974 - val_loss: 0.0708 - val_acc: 0.9875
Epoch 32/70
16000/16000 [==============================] - 12s 764us/step - loss: 0.0064 - acc: 0.9982 - val_loss: 0.0757 - val_acc: 0.9890
Epoch 33/70
16000/16000 [==============================] - 12s 771us/step - loss: 0.0089 - acc: 0.9977 - val_loss: 0.0755 - val_acc: 0.9882
Epoch 34/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0079 - acc: 0.9984 - val_loss: 0.0745 - val_acc: 0.9890
Epoch 35/70
16000/16000 [==============================] - 12s 768us/step - loss: 0.0032 - acc: 0.9990 - val_loss: 0.0695 - val_acc: 0.9900
Epoch 36/70
16000/16000 [==============================] - 12s 767us/step - loss: 0.0096 - acc: 0.9979 - val_loss: 0.0711 - val_acc: 0.9872
Epoch 37/70
16000/16000 [==============================] - 12s 771us/step - loss: 0.0081 - acc: 0.9981 - val_loss: 0.0791 - val_acc: 0.9878
Epoch 38/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0041 - acc: 0.9988 - val_loss: 0.0742 - val_acc: 0.9895
Epoch 39/70
16000/16000 [==============================] - 12s 770us/step - loss: 0.0020 - acc: 0.9992 - val_loss: 0.0839 - val_acc: 0.9882
Epoch 40/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0086 - acc: 0.9981 - val_loss: 0.0655 - val_acc: 0.9920
Epoch 41/70
16000/16000 [==============================] - 12s 766us/step - loss: 0.0060 - acc: 0.9981 - val_loss: 0.0795 - val_acc: 0.9882
Epoch 42/70
16000/16000 [==============================] - 13s 826us/step - loss: 0.0118 - acc: 0.9970 - val_loss: 0.0824 - val_acc: 0.9882
Epoch 43/70
16000/16000 [==============================] - 12s 779us/step - loss: 0.0128 - acc: 0.9971 - val_loss: 0.1105 - val_acc: 0.9870
Epoch 44/70
16000/16000 [==============================] - 12s 777us/step - loss: 0.0234 - acc: 0.9959 - val_loss: 0.1142 - val_acc: 0.9878
Epoch 45/70
16000/16000 [==============================] - 12s 774us/step - loss: 0.0181 - acc: 0.9969 - val_loss: 0.1035 - val_acc: 0.9870
Epoch 46/70
16000/16000 [==============================] - 12s 775us/step - loss: 0.0113 - acc: 0.9978 - val_loss: 0.0940 - val_acc: 0.9890
Epoch 47/70
16000/16000 [==============================] - 12s 779us/step - loss: 0.0063 - acc: 0.9991 - val_loss: 0.0950 - val_acc: 0.9882
Epoch 48/70
16000/16000 [==============================] - 12s 781us/step - loss: 0.0045 - acc: 0.9989 - val_loss: 0.0956 - val_acc: 0.9872
Epoch 49/70
16000/16000 [==============================] - 12s 767us/step - loss: 0.0087 - acc: 0.9981 - val_loss: 0.1008 - val_acc: 0.9888
Epoch 50/70
16000/16000 [==============================] - 12s 771us/step - loss: 0.0137 - acc: 0.9976 - val_loss: 0.0728 - val_acc: 0.9908
Epoch 51/70
16000/16000 [==============================] - 12s 776us/step - loss: 0.0074 - acc: 0.9983 - val_loss: 0.0860 - val_acc: 0.9895
Epoch 52/70
16000/16000 [==============================] - 12s 778us/step - loss: 0.0050 - acc: 0.9985 - val_loss: 0.0928 - val_acc: 0.9895
Epoch 53/70
16000/16000 [==============================] - 12s 766us/step - loss: 0.0056 - acc: 0.9985 - val_loss: 0.1286 - val_acc: 0.9875
Epoch 54/70
16000/16000 [==============================] - 12s 776us/step - loss: 0.0116 - acc: 0.9980 - val_loss: 0.1413 - val_acc: 0.9868
Epoch 55/70
16000/16000 [==============================] - 12s 778us/step - loss: 0.0166 - acc: 0.9973 - val_loss: 0.1010 - val_acc: 0.9882
Epoch 56/70
16000/16000 [==============================] - 12s 769us/step - loss: 0.0027 - acc: 0.9991 - val_loss: 0.1015 - val_acc: 0.9875
Epoch 57/70
16000/16000 [==============================] - 12s 778us/step - loss: 0.0019 - acc: 0.9996 - val_loss: 0.0946 - val_acc: 0.9905
Epoch 58/70
16000/16000 [==============================] - 12s 772us/step - loss: 0.0042 - acc: 0.9993 - val_loss: 0.1181 - val_acc: 0.9878
Epoch 59/70
16000/16000 [==============================] - 12s 770us/step - loss: 0.0077 - acc: 0.9986 - val_loss: 0.1102 - val_acc: 0.9880
Epoch 60/70
16000/16000 [==============================] - 12s 775us/step - loss: 0.0097 - acc: 0.9983 - val_loss: 0.1141 - val_acc: 0.9875
Epoch 61/70
16000/16000 [==============================] - 12s 780us/step - loss: 0.0158 - acc: 0.9979 - val_loss: 0.1061 - val_acc: 0.9870
Epoch 62/70
16000/16000 [==============================] - 12s 778us/step - loss: 0.0087 - acc: 0.9985 - val_loss: 0.0876 - val_acc: 0.9885
Epoch 63/70
16000/16000 [==============================] - 12s 772us/step - loss: 0.0070 - acc: 0.9984 - val_loss: 0.1457 - val_acc: 0.9848
Epoch 64/70
16000/16000 [==============================] - 12s 779us/step - loss: 0.0145 - acc: 0.9978 - val_loss: 0.1175 - val_acc: 0.9868
Epoch 65/70
16000/16000 [==============================] - 12s 774us/step - loss: 0.0114 - acc: 0.9979 - val_loss: 0.1168 - val_acc: 0.9872
Epoch 66/70
16000/16000 [==============================] - 12s 774us/step - loss: 0.0060 - acc: 0.9990 - val_loss: 0.1191 - val_acc: 0.9870
Epoch 67/70
16000/16000 [==============================] - 13s 786us/step - loss: 0.0104 - acc: 0.9983 - val_loss: 0.1297 - val_acc: 0.9860
Epoch 68/70
16000/16000 [==============================] - 12s 773us/step - loss: 0.0066 - acc: 0.9990 - val_loss: 0.1156 - val_acc: 0.9882
Epoch 69/70
16000/16000 [==============================] - 12s 771us/step - loss: 0.0030 - acc: 0.9992 - val_loss: 0.1482 - val_acc: 0.9855
Epoch 70/70
16000/16000 [==============================] - 12s 774us/step - loss: 0.0070 - acc: 0.9990 - val_loss: 0.1303 - val_acc: 0.9865

报错：

ValueError: Negative dimension size caused by subtracting 3 from 1 for 'conv2d_2/convolution' (op: 'Conv2D') with input shapes: [?,1,28,64], [3,3,64,64].

 一般是网络结构问题，本次练习中是输入的时候出错了。

第三次练习：

这次训练效果还可以。

import numpy as np
import tensorflow as tf
import pandas as pd
import matplotlib.pyplot as plt
from keras.models import Sequential
from keras.layers import Dense,Activation,Conv2D
from keras.layers import MaxPool2D,Flatten,Dropout,ZeroPadding2D,BatchNormalization
from  keras.utils import np_utils
from keras import metrics
import keras
from keras.models import save_model,load_model
from keras.models import Model
from keras.callbacks import ModelCheckpoint
import os
from keras.utils import plot_model
df=pd.read_csv("sample_data/mnist_train_small.csv",header=None)
data=df.as_matrix()
df=None


np.random.shuffle(data)
x_train=data[:,1:]
x_train=x_train.reshape(data.shape[0],28,28,1).astype("float32")#将x_train变为data.shape[0]=60000个28*28,1通道的矩阵
x_train=x_train/255.0

y_train=np_utils.to_categorical(data[:,0],10).astype("float32")#将类别向量(从0到nb_classes的整数向量)映射为二值类别矩阵, 用于应用到以categorical_crossentropy为目标函数的模型中.



print(x_train.shape)
print(y_train.shape)

batch_size=64
n_filters=32
pool_size=(2,2)

model = Sequential()
#model.add(ZeroPadding2D(padding=(1,1),input_shape=(28,28,1)))


model.add(Conv2D(filters=64, kernel_size=(3, 3),padding = 'same',activation='relu',input_shape=(28,28,1)))

model.add(Conv2D(filters=64, kernel_size=(3, 3),padding = 'same',activation='relu')) #28


model.add(MaxPool2D(pool_size=(2,2)))

model.add(Conv2D(filters=128, kernel_size=(3, 3),padding = 'same',activation='relu'))

model.add(Conv2D(filters=128, kernel_size=(3, 3),padding = 'same',activation='relu')) #14


model.add(MaxPool2D(pool_size=(2,2)))

model.add(Conv2D(filters=512, kernel_size=(3, 3),padding = 'same'))
model.add(Activation('relu'))
model.add(Conv2D(filters=512, kernel_size=(3, 3),padding = 'same'))
model.add(Activation('relu'))
model.add(Conv2D(filters=512, kernel_size=(3, 3),padding = 'same'))
model.add(Activation('relu'))

model.add(MaxPool2D(pool_size=(2,2)))
 
model.add(Flatten())#压平上述向量，变成一维25088
model.add(Dense(512, activation='relu'))#全连接层有4096个神经核，参数个数就是4096*25088
model.add(Dropout(0.5))#0.5的概率抛弃一些连接
model.add(Dense(512, activation='relu'))#再来一个全连接
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))
#summary查看网络结构
model.summary()
model.compile(loss='categorical_crossentropy',optimizer='adam', metrics=['acc'])

plot_model(model, to_file='model_size.png')


hist=model.fit(x_train,y_train,batch_size=batch_size,epochs=70,verbose=1,validation_split=0.2)#50
model.save('learnCNN.h5')
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
plt.savefig("loss.png")
plt.clf()
plt.plot(hist.history['acc'])
plt.plot(hist.history['val_acc'])
plt.title('model acc')
plt.ylabel('acc')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
plt.savefig("acc.png")
print('finished!')

训练过程：

(20000, 28, 28, 1)
(20000, 10)
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_5 (Conv2D)            (None, 28, 28, 64)        640       
_________________________________________________________________
conv2d_6 (Conv2D)            (None, 28, 28, 64)        36928     
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 14, 14, 64)        0         
_________________________________________________________________
conv2d_7 (Conv2D)            (None, 14, 14, 128)       73856     
_________________________________________________________________
conv2d_8 (Conv2D)            (None, 14, 14, 128)       147584    
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 7, 7, 128)         0         
_________________________________________________________________
conv2d_9 (Conv2D)            (None, 7, 7, 512)         590336    
_________________________________________________________________
activation_1 (Activation)    (None, 7, 7, 512)         0         
_________________________________________________________________
conv2d_10 (Conv2D)           (None, 7, 7, 512)         2359808   
_________________________________________________________________
activation_2 (Activation)    (None, 7, 7, 512)         0         
_________________________________________________________________
conv2d_11 (Conv2D)           (None, 7, 7, 512)         2359808   
_________________________________________________________________
activation_3 (Activation)    (None, 7, 7, 512)         0         
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 3, 3, 512)         0         
_________________________________________________________________
flatten_2 (Flatten)          (None, 4608)              0         
_________________________________________________________________
dense_4 (Dense)              (None, 512)               2359808   
_________________________________________________________________
dropout_3 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_5 (Dense)              (None, 512)               262656    
_________________________________________________________________
dropout_4 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_6 (Dense)              (None, 10)                5130      
=================================================================
Total params: 8,196,554
Trainable params: 8,196,554
Non-trainable params: 0
_________________________________________________________________
Train on 16000 samples, validate on 4000 samples
Epoch 1/70
16000/16000 [==============================] - 19s 1ms/step - loss: 1.6370 - acc: 0.3870 - val_loss: 0.2657 - val_acc: 0.9207
Epoch 2/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.2069 - acc: 0.9433 - val_loss: 0.1121 - val_acc: 0.9670
Epoch 3/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.1116 - acc: 0.9691 - val_loss: 0.0732 - val_acc: 0.9775
Epoch 4/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0777 - acc: 0.9785 - val_loss: 0.0574 - val_acc: 0.9810
Epoch 5/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0619 - acc: 0.9822 - val_loss: 0.0980 - val_acc: 0.9742
Epoch 6/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0548 - acc: 0.9855 - val_loss: 0.0609 - val_acc: 0.9822
Epoch 7/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0445 - acc: 0.9876 - val_loss: 0.0569 - val_acc: 0.9842
Epoch 8/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0405 - acc: 0.9896 - val_loss: 0.0522 - val_acc: 0.9875
Epoch 9/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0424 - acc: 0.9881 - val_loss: 0.0654 - val_acc: 0.9845
Epoch 10/70
16000/16000 [==============================] - 19s 1ms/step - loss: 0.0233 - acc: 0.9931 - val_loss: 0.0520 - val_acc: 0.9890
Epoch 11/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0285 - acc: 0.9913 - val_loss: 0.0789 - val_acc: 0.9835
Epoch 12/70
16000/16000 [==============================] - 19s 1ms/step - loss: 0.0473 - acc: 0.9874 - val_loss: 0.0789 - val_acc: 0.9828
Epoch 13/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0291 - acc: 0.9927 - val_loss: 0.0631 - val_acc: 0.9855
Epoch 14/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0227 - acc: 0.9936 - val_loss: 0.0795 - val_acc: 0.9845
Epoch 15/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0276 - acc: 0.9925 - val_loss: 0.0781 - val_acc: 0.9798
Epoch 16/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0278 - acc: 0.9925 - val_loss: 0.0511 - val_acc: 0.9880
Epoch 17/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0222 - acc: 0.9938 - val_loss: 0.0630 - val_acc: 0.9878
Epoch 18/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0300 - acc: 0.9926 - val_loss: 0.0686 - val_acc: 0.9875
Epoch 19/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0220 - acc: 0.9951 - val_loss: 0.0488 - val_acc: 0.9912
Epoch 20/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0120 - acc: 0.9972 - val_loss: 0.0614 - val_acc: 0.9885
Epoch 21/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0168 - acc: 0.9959 - val_loss: 0.0631 - val_acc: 0.9862
Epoch 22/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0177 - acc: 0.9952 - val_loss: 0.0620 - val_acc: 0.9900
Epoch 23/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0246 - acc: 0.9935 - val_loss: 0.0950 - val_acc: 0.9848
Epoch 24/70
16000/16000 [==============================] - 19s 1ms/step - loss: 0.0212 - acc: 0.9949 - val_loss: 0.0608 - val_acc: 0.9902
Epoch 25/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0158 - acc: 0.9961 - val_loss: 0.0608 - val_acc: 0.9910
Epoch 26/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0046 - acc: 0.9989 - val_loss: 0.0703 - val_acc: 0.9910
Epoch 27/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0334 - acc: 0.9926 - val_loss: 0.0673 - val_acc: 0.9905
Epoch 28/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0247 - acc: 0.9944 - val_loss: 0.0751 - val_acc: 0.9862
Epoch 29/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0138 - acc: 0.9966 - val_loss: 0.1002 - val_acc: 0.9862
Epoch 30/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0234 - acc: 0.9949 - val_loss: 0.0763 - val_acc: 0.9868
Epoch 31/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0172 - acc: 0.9957 - val_loss: 0.0580 - val_acc: 0.9898
Epoch 32/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0087 - acc: 0.9979 - val_loss: 0.0790 - val_acc: 0.9902
Epoch 33/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0111 - acc: 0.9974 - val_loss: 0.0800 - val_acc: 0.9882
Epoch 34/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0179 - acc: 0.9957 - val_loss: 0.0699 - val_acc: 0.9890
Epoch 35/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0035 - acc: 0.9992 - val_loss: 0.0890 - val_acc: 0.9905
Epoch 36/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0315 - acc: 0.9942 - val_loss: 0.0787 - val_acc: 0.9902
Epoch 37/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0098 - acc: 0.9979 - val_loss: 0.0595 - val_acc: 0.9915
Epoch 38/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0151 - acc: 0.9965 - val_loss: 0.0692 - val_acc: 0.9888
Epoch 39/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0069 - acc: 0.9981 - val_loss: 0.0858 - val_acc: 0.9885
Epoch 40/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0331 - acc: 0.9929 - val_loss: 0.1283 - val_acc: 0.9828
Epoch 41/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0250 - acc: 0.9951 - val_loss: 0.0530 - val_acc: 0.9918
Epoch 42/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0106 - acc: 0.9978 - val_loss: 0.1921 - val_acc: 0.9792
Epoch 43/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0355 - acc: 0.9931 - val_loss: 0.0691 - val_acc: 0.9880
Epoch 44/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0094 - acc: 0.9976 - val_loss: 0.0809 - val_acc: 0.9902
Epoch 45/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0121 - acc: 0.9970 - val_loss: 0.0786 - val_acc: 0.9872
Epoch 46/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0142 - acc: 0.9969 - val_loss: 0.0769 - val_acc: 0.9882
Epoch 47/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0089 - acc: 0.9978 - val_loss: 0.0899 - val_acc: 0.9892
Epoch 48/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0036 - acc: 0.9992 - val_loss: 0.0816 - val_acc: 0.9902
Epoch 49/70
16000/16000 [==============================] - 18s 1ms/step - loss: 1.4944e-04 - acc: 1.0000 - val_loss: 0.0750 - val_acc: 0.9915
Epoch 50/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0078 - acc: 0.9990 - val_loss: 0.1065 - val_acc: 0.9875
Epoch 51/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0423 - acc: 0.9929 - val_loss: 0.1552 - val_acc: 0.9802
Epoch 52/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0249 - acc: 0.9952 - val_loss: 0.0895 - val_acc: 0.9882
Epoch 53/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0239 - acc: 0.9964 - val_loss: 0.0905 - val_acc: 0.9868
Epoch 54/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0146 - acc: 0.9973 - val_loss: 0.1291 - val_acc: 0.9838
Epoch 55/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0288 - acc: 0.9959 - val_loss: 0.1286 - val_acc: 0.9840
Epoch 56/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0267 - acc: 0.9955 - val_loss: 0.1017 - val_acc: 0.9898
Epoch 57/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0216 - acc: 0.9967 - val_loss: 0.0890 - val_acc: 0.9880
Epoch 58/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0207 - acc: 0.9973 - val_loss: 0.1011 - val_acc: 0.9858
Epoch 59/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0288 - acc: 0.9949 - val_loss: 0.1393 - val_acc: 0.9868
Epoch 60/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0089 - acc: 0.9986 - val_loss: 0.1173 - val_acc: 0.9875
Epoch 61/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0027 - acc: 0.9994 - val_loss: 0.1491 - val_acc: 0.9858
Epoch 62/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0245 - acc: 0.9963 - val_loss: 0.1463 - val_acc: 0.9858
Epoch 63/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0322 - acc: 0.9956 - val_loss: 0.1353 - val_acc: 0.9878
Epoch 64/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0443 - acc: 0.9939 - val_loss: 0.1150 - val_acc: 0.9862
Epoch 65/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0331 - acc: 0.9956 - val_loss: 0.1387 - val_acc: 0.9848
Epoch 66/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0241 - acc: 0.9968 - val_loss: 0.1167 - val_acc: 0.9862
Epoch 67/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0292 - acc: 0.9961 - val_loss: 0.1012 - val_acc: 0.9892
Epoch 68/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0190 - acc: 0.9973 - val_loss: 0.0743 - val_acc: 0.9918
Epoch 69/70
16000/16000 [==============================] - 20s 1ms/step - loss: 0.0154 - acc: 0.9971 - val_loss: 0.0846 - val_acc: 0.9890
Epoch 70/70
16000/16000 [==============================] - 18s 1ms/step - loss: 0.0191 - acc: 0.9978 - val_loss: 0.1131 - val_acc: 0.9868

结果：

第四次练习： 

import numpy as np
import tensorflow as tf
import pandas as pd
import matplotlib.pyplot as plt
from keras.models import Sequential
from keras.layers import Dense,Activation,Conv2D
from keras.layers import MaxPool2D,Flatten,Dropout,ZeroPadding2D,BatchNormalization
from  keras.utils import np_utils
from keras import metrics
import keras
from keras.models import save_model,load_model
from keras.models import Model
from keras.callbacks import ModelCheckpoint
import os
from keras.utils import plot_model
from keras.optimizers import SGD,Adam
df=pd.read_csv("drive/My Drive/GoogleAI/train.csv")
data=df.as_matrix()
df=None



np.random.shuffle(data)
x_train=data[0:,1:-1]
x_train=x_train.reshape(data.shape[0],40,40,1).astype("float32")#将x_train变为data.shape[0]=60000个28*28,1通道的矩阵
x_train=x_train/255.0
y_train=np_utils.to_categorical(data[0:,-1],2).astype("float32")

print(x_train.shape)
print(y_train.shape)

batch_size=32
n_filters=32
pool_size=(2,2)

model = Sequential()
#model.add(ZeroPadding2D(padding=(1,1),input_shape=(28,28,1)))


model.add(Conv2D(filters=64, kernel_size=(3, 3),padding = 'same',activation='relu',input_shape=(40,40,1)))

model.add(Conv2D(filters=64, kernel_size=(3, 3),padding = 'same',activation='relu')) #28


model.add(MaxPool2D(pool_size=(2,2)))

model.add(Conv2D(filters=128, kernel_size=(3, 3),padding = 'same',activation='relu'))

model.add(Conv2D(filters=128, kernel_size=(3, 3),padding = 'same',activation='relu')) #14


model.add(MaxPool2D(pool_size=(2,2)))

model.add(Conv2D(filters=512, kernel_size=(3, 3),padding = 'same'))
model.add(Activation('relu'))
model.add(Conv2D(filters=512, kernel_size=(3, 3),padding = 'same'))
model.add(Activation('relu'))
model.add(Conv2D(filters=512, kernel_size=(3, 3),padding = 'same'))
model.add(Activation('relu'))

model.add(MaxPool2D(pool_size=(2,2)))
 
model.add(Flatten())#压平上述向量，变成一维25088
model.add(Dense(512, activation='relu'))#全连接层有4096个神经核，参数个数就是4096*25088
model.add(Dropout(0.5))#0.5的概率抛弃一些连接
model.add(Dense(512, activation='relu'))#再来一个全连接
model.add(Dropout(0.5))
model.add(Dense(2, activation='softmax'))
#summary查看网络结构
model.summary()
sgd =SGD(lr=0.001, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='categorical_crossentropy',optimizer=sgd, metrics=['acc'])

plot_model(model, to_file='model_size.png')


hist=model.fit(x_train,y_train,batch_size=batch_size,epochs=70,verbose=1,validation_split=0.2)#50
model.save('learnCNN.h5')
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
plt.savefig("loss.png")
plt.clf()
plt.plot(hist.history['acc'])
plt.plot(hist.history['val_acc'])
plt.title('model acc')
plt.ylabel('acc')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
plt.savefig("acc.png")
print('finished!')

训练过程：

(4000, 40, 40, 1)
(4000, 2)
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_36 (Conv2D)           (None, 40, 40, 64)        640       
_________________________________________________________________
conv2d_37 (Conv2D)           (None, 40, 40, 64)        36928     
_________________________________________________________________
max_pooling2d_16 (MaxPooling (None, 20, 20, 64)        0         
_________________________________________________________________
conv2d_38 (Conv2D)           (None, 20, 20, 128)       73856     
_________________________________________________________________
conv2d_39 (Conv2D)           (None, 20, 20, 128)       147584    
_________________________________________________________________
max_pooling2d_17 (MaxPooling (None, 10, 10, 128)       0         
_________________________________________________________________
conv2d_40 (Conv2D)           (None, 10, 10, 512)       590336    
_________________________________________________________________
activation_16 (Activation)   (None, 10, 10, 512)       0         
_________________________________________________________________
conv2d_41 (Conv2D)           (None, 10, 10, 512)       2359808   
_________________________________________________________________
activation_17 (Activation)   (None, 10, 10, 512)       0         
_________________________________________________________________
conv2d_42 (Conv2D)           (None, 10, 10, 512)       2359808   
_________________________________________________________________
activation_18 (Activation)   (None, 10, 10, 512)       0         
_________________________________________________________________
max_pooling2d_18 (MaxPooling (None, 5, 5, 512)         0         
_________________________________________________________________
flatten_6 (Flatten)          (None, 12800)             0         
_________________________________________________________________
dense_16 (Dense)             (None, 512)               6554112   
_________________________________________________________________
dropout_11 (Dropout)         (None, 512)               0         
_________________________________________________________________
dense_17 (Dense)             (None, 512)               262656    
_________________________________________________________________
dropout_12 (Dropout)         (None, 512)               0         
_________________________________________________________________
dense_18 (Dense)             (None, 2)                 1026      
=================================================================
Total params: 12,386,754
Trainable params: 12,386,754
Non-trainable params: 0
_________________________________________________________________
Train on 3200 samples, validate on 800 samples
Epoch 1/70
3200/3200 [==============================] - 8s 3ms/step - loss: 0.6926 - acc: 0.5222 - val_loss: 0.6898 - val_acc: 0.5400
Epoch 2/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.6907 - acc: 0.5353 - val_loss: 0.6873 - val_acc: 0.5400
Epoch 3/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.6883 - acc: 0.5525 - val_loss: 0.6846 - val_acc: 0.6000
Epoch 4/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.6850 - acc: 0.6047 - val_loss: 0.6784 - val_acc: 0.7100
Epoch 5/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.6779 - acc: 0.6653 - val_loss: 0.6669 - val_acc: 0.7662
Epoch 6/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.6643 - acc: 0.7041 - val_loss: 0.6415 - val_acc: 0.7812
Epoch 7/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.6283 - acc: 0.7222 - val_loss: 0.5647 - val_acc: 0.7800
Epoch 8/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.5310 - acc: 0.7531 - val_loss: 0.4189 - val_acc: 0.8113
Epoch 9/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.4246 - acc: 0.8128 - val_loss: 0.2940 - val_acc: 0.8688
Epoch 10/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.2660 - acc: 0.8934 - val_loss: 0.2026 - val_acc: 0.8938
Epoch 11/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.1484 - acc: 0.9444 - val_loss: 0.0621 - val_acc: 0.9838
Epoch 12/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0681 - acc: 0.9819 - val_loss: 0.0184 - val_acc: 0.9988
Epoch 13/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0317 - acc: 0.9928 - val_loss: 0.0102 - val_acc: 0.9988
Epoch 14/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0191 - acc: 0.9963 - val_loss: 0.0095 - val_acc: 0.9988
Epoch 15/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0155 - acc: 0.9963 - val_loss: 0.0073 - val_acc: 0.9988
Epoch 16/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0150 - acc: 0.9966 - val_loss: 0.0046 - val_acc: 0.9988
Epoch 17/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0128 - acc: 0.9963 - val_loss: 0.0040 - val_acc: 0.9988
Epoch 18/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0065 - acc: 0.9984 - val_loss: 0.0036 - val_acc: 0.9988
Epoch 19/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0045 - acc: 0.9991 - val_loss: 0.0033 - val_acc: 0.9988
Epoch 20/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0031 - acc: 0.9994 - val_loss: 9.9130e-04 - val_acc: 1.0000
Epoch 21/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0024 - acc: 0.9997 - val_loss: 0.0012 - val_acc: 0.9988
Epoch 22/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0018 - acc: 1.0000 - val_loss: 9.4619e-04 - val_acc: 1.0000
Epoch 23/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0011 - acc: 0.9997 - val_loss: 0.0039 - val_acc: 0.9988
Epoch 24/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0020 - acc: 0.9994 - val_loss: 0.0045 - val_acc: 0.9988
Epoch 25/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0010 - acc: 1.0000 - val_loss: 0.0026 - val_acc: 0.9988
Epoch 26/70
3200/3200 [==============================] - 7s 2ms/step - loss: 8.3724e-04 - acc: 1.0000 - val_loss: 5.5013e-04 - val_acc: 1.0000
Epoch 27/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0020 - acc: 0.9994 - val_loss: 0.0043 - val_acc: 0.9975
Epoch 28/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0036 - acc: 0.9988 - val_loss: 0.0012 - val_acc: 0.9988
Epoch 29/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0017 - acc: 0.9997 - val_loss: 4.0852e-04 - val_acc: 1.0000
Epoch 30/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0010 - acc: 0.9997 - val_loss: 0.0211 - val_acc: 0.9925
Epoch 31/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0027 - acc: 0.9988 - val_loss: 0.0012 - val_acc: 0.9988
Epoch 32/70
3200/3200 [==============================] - 7s 2ms/step - loss: 0.0011 - acc: 1.0000 - val_loss: 1.9495e-04 - val_acc: 1.0000
Epoch 33/70
3200/3200 [==============================] - 7s 2ms/step - loss: 5.4917e-04 - acc: 1.0000 - val_loss: 8.9707e-04 - val_acc: 1.0000
Epoch 34/70
3200/3200 [==============================] - 7s 2ms/step - loss: 4.6175e-04 - acc: 1.0000 - val_loss: 1.4151e-04 - val_acc: 1.0000
Epoch 35/70
3200/3200 [==============================] - 7s 2ms/step - loss: 3.0411e-04 - acc: 1.0000 - val_loss: 2.5414e-04 - val_acc: 1.0000
Epoch 36/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.6523e-04 - acc: 1.0000 - val_loss: 5.1636e-04 - val_acc: 1.0000
Epoch 37/70
3200/3200 [==============================] - 7s 2ms/step - loss: 2.5121e-04 - acc: 1.0000 - val_loss: 2.8750e-04 - val_acc: 1.0000
Epoch 38/70
3200/3200 [==============================] - 7s 2ms/step - loss: 2.6499e-04 - acc: 1.0000 - val_loss: 6.5903e-04 - val_acc: 1.0000
Epoch 39/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.6837e-04 - acc: 1.0000 - val_loss: 1.4067e-04 - val_acc: 1.0000
Epoch 40/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.4955e-04 - acc: 1.0000 - val_loss: 1.1061e-04 - val_acc: 1.0000
Epoch 41/70
3200/3200 [==============================] - 7s 2ms/step - loss: 2.2614e-04 - acc: 1.0000 - val_loss: 2.3998e-04 - val_acc: 1.0000
Epoch 42/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.8178e-04 - acc: 1.0000 - val_loss: 4.0963e-04 - val_acc: 1.0000
Epoch 43/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.0676e-04 - acc: 1.0000 - val_loss: 1.6768e-04 - val_acc: 1.0000
Epoch 44/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.4944e-04 - acc: 1.0000 - val_loss: 1.6885e-04 - val_acc: 1.0000
Epoch 45/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.6997e-04 - acc: 1.0000 - val_loss: 7.1427e-05 - val_acc: 1.0000
Epoch 46/70
3200/3200 [==============================] - 7s 2ms/step - loss: 9.8449e-05 - acc: 1.0000 - val_loss: 1.3498e-04 - val_acc: 1.0000
Epoch 47/70
3200/3200 [==============================] - 7s 2ms/step - loss: 6.3622e-05 - acc: 1.0000 - val_loss: 1.1161e-04 - val_acc: 1.0000
Epoch 48/70
3200/3200 [==============================] - 7s 2ms/step - loss: 2.6218e-04 - acc: 1.0000 - val_loss: 2.9487e-04 - val_acc: 1.0000
Epoch 49/70
3200/3200 [==============================] - 7s 2ms/step - loss: 6.8380e-05 - acc: 1.0000 - val_loss: 3.2897e-04 - val_acc: 1.0000
Epoch 50/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.3235e-04 - acc: 1.0000 - val_loss: 1.4400e-04 - val_acc: 1.0000
Epoch 51/70
3200/3200 [==============================] - 7s 2ms/step - loss: 7.6817e-05 - acc: 1.0000 - val_loss: 1.8169e-04 - val_acc: 1.0000
Epoch 52/70
3200/3200 [==============================] - 7s 2ms/step - loss: 7.9961e-05 - acc: 1.0000 - val_loss: 1.9066e-04 - val_acc: 1.0000
Epoch 53/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.3203e-04 - acc: 1.0000 - val_loss: 2.2335e-04 - val_acc: 1.0000
Epoch 54/70
3200/3200 [==============================] - 7s 2ms/step - loss: 6.2024e-05 - acc: 1.0000 - val_loss: 8.6982e-05 - val_acc: 1.0000
Epoch 55/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.3301e-04 - acc: 1.0000 - val_loss: 2.6485e-04 - val_acc: 1.0000
Epoch 56/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.0229e-04 - acc: 1.0000 - val_loss: 7.6931e-04 - val_acc: 1.0000
Epoch 57/70
3200/3200 [==============================] - 7s 2ms/step - loss: 2.9938e-04 - acc: 1.0000 - val_loss: 0.0026 - val_acc: 0.9988
Epoch 58/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.8570e-04 - acc: 1.0000 - val_loss: 1.3332e-04 - val_acc: 1.0000
Epoch 59/70
3200/3200 [==============================] - 7s 2ms/step - loss: 8.2858e-05 - acc: 1.0000 - val_loss: 8.2129e-05 - val_acc: 1.0000
Epoch 60/70
3200/3200 [==============================] - 7s 2ms/step - loss: 6.2264e-05 - acc: 1.0000 - val_loss: 1.0242e-04 - val_acc: 1.0000
Epoch 61/70
3200/3200 [==============================] - 7s 2ms/step - loss: 2.0275e-04 - acc: 1.0000 - val_loss: 1.0507e-04 - val_acc: 1.0000
Epoch 62/70
3200/3200 [==============================] - 7s 2ms/step - loss: 5.9341e-05 - acc: 1.0000 - val_loss: 3.0019e-04 - val_acc: 1.0000
Epoch 63/70
3200/3200 [==============================] - 7s 2ms/step - loss: 6.4440e-05 - acc: 1.0000 - val_loss: 7.8200e-05 - val_acc: 1.0000
Epoch 64/70
3200/3200 [==============================] - 7s 2ms/step - loss: 1.2058e-04 - acc: 1.0000 - val_loss: 1.4521e-04 - val_acc: 1.0000
Epoch 65/70
3200/3200 [==============================] - 7s 2ms/step - loss: 3.5251e-05 - acc: 1.0000 - val_loss: 1.5971e-04 - val_acc: 1.0000
Epoch 66/70
3200/3200 [==============================] - 7s 2ms/step - loss: 4.8072e-05 - acc: 1.0000 - val_loss: 2.3435e-04 - val_acc: 1.0000
Epoch 67/70
3200/3200 [==============================] - 7s 2ms/step - loss: 7.0874e-05 - acc: 1.0000 - val_loss: 1.8044e-04 - val_acc: 1.0000
Epoch 68/70
3200/3200 [==============================] - 7s 2ms/step - loss: 3.6950e-05 - acc: 1.0000 - val_loss: 1.5588e-04 - val_acc: 1.0000
Epoch 69/70
3200/3200 [==============================] - 7s 2ms/step - loss: 3.7245e-05 - acc: 1.0000 - val_loss: 2.4842e-04 - val_acc: 1.0000
Epoch 70/70
3200/3200 [==============================] - 7s 2ms/step - loss: 4.7517e-05 - acc: 1.0000 - val_loss: 1.9751e-04 - val_acc: 1.0000

训练结果：

第五次练习：

用VGG16作为预训练，一次性100epoch结果感觉一般，但分两次训练的结果却比它好（暂不知道原因）。

import numpy as np
import tensorflow as tf
import pandas as pd
import matplotlib.pyplot as plt
from keras.models import Sequential
from keras.layers import Dense,Activation,Conv2D
from keras.layers import MaxPool2D,Flatten,Dropout,ZeroPadding2D,BatchNormalization
from  keras.utils import np_utils
from keras import metrics
import keras
from keras.models import save_model,load_model
from keras.models import Model
from keras.callbacks import ModelCheckpoint
import os
from keras.utils import plot_model
from keras.optimizers import SGD,Adam
from keras.preprocessing.image import ImageDataGenerator
from keras.applications import VGG16
from keras.applications.vgg16 import preprocess_input

train_datagen=ImageDataGenerator(rescale=1./255,rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True)
test_datagen=ImageDataGenerator(rescale=1./255)

train_dir="drive/My Drive/GoogleAI/train/"
train_generator=train_datagen.flow_from_directory(train_dir,target_size=(150,150),batch_size=32,class_mode='binary')
test_dir="drive/My Drive/GoogleAI/test/"
test_generator=test_datagen.flow_from_directory(test_dir,target_size=(150,150),batch_size=32,class_mode='binary')


conv_base=VGG16(weights='imagenet',include_top=False,input_shape=(150,150,3))

batch_size=48




model = Sequential()

model.add(conv_base)
model.add(Flatten())
model.add(Dense(256,activation='relu'))
model.add(Dense(1, activation='sigmoid'))
conv_base.trainable=False
sgd =SGD(lr=0.001, decay=1e-6, momentum=0.9, nesterov=True)#
model.compile(loss='binary_crossentropy',optimizer=sgd, metrics=['acc'])
hist=model.fit_generator(train_generator,steps_per_epoch=44,epochs=100,validation_data=test_generator,validation_steps=32)

model.save('learnCNN.h5')
plt.plot(hist.history['loss'])
plt.plot(hist.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
plt.savefig("loss.png")
plt.clf()
plt.plot(hist.history['acc'])
plt.plot(hist.history['val_acc'])
plt.title('model acc')
plt.ylabel('acc')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper left')
plt.show()
plt.savefig("acc.png")
print('finished!')

训练过程如下：

Using TensorFlow backend.
Found 1420 images belonging to 2 classes.
Found 700 images belonging to 2 classes.
Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5
58892288/58889256 [==============================] - 1s 0us/step
Epoch 1/100
44/44 [==============================] - 295s 7s/step - loss: 0.5924 - acc: 0.6754 - val_loss: 0.4320 - val_acc: 0.8010
Epoch 2/100
44/44 [==============================] - 19s 432ms/step - loss: 0.4807 - acc: 0.7674 - val_loss: 0.3717 - val_acc: 0.8422
Epoch 3/100
44/44 [==============================] - 19s 425ms/step - loss: 0.4478 - acc: 0.7709 - val_loss: 0.3626 - val_acc: 0.8484
Epoch 4/100
44/44 [==============================] - 19s 428ms/step - loss: 0.4345 - acc: 0.7979 - val_loss: 0.3914 - val_acc: 0.8157
Epoch 5/100
44/44 [==============================] - 19s 425ms/step - loss: 0.4138 - acc: 0.8085 - val_loss: 0.3095 - val_acc: 0.8720
Epoch 6/100
44/44 [==============================] - 19s 428ms/step - loss: 0.3826 - acc: 0.8232 - val_loss: 0.3410 - val_acc: 0.8490
Epoch 7/100
44/44 [==============================] - 19s 425ms/step - loss: 0.3806 - acc: 0.8243 - val_loss: 0.3260 - val_acc: 0.8524
Epoch 8/100
44/44 [==============================] - 19s 424ms/step - loss: 0.3444 - acc: 0.8407 - val_loss: 0.3096 - val_acc: 0.8716
Epoch 9/100
44/44 [==============================] - 19s 425ms/step - loss: 0.3629 - acc: 0.8305 - val_loss: 0.3100 - val_acc: 0.8760
Epoch 10/100
44/44 [==============================] - 19s 424ms/step - loss: 0.3494 - acc: 0.8532 - val_loss: 0.3032 - val_acc: 0.8784
Epoch 11/100
44/44 [==============================] - 19s 421ms/step - loss: 0.3643 - acc: 0.8355 - val_loss: 0.3287 - val_acc: 0.8612
Epoch 12/100
44/44 [==============================] - 18s 420ms/step - loss: 0.3397 - acc: 0.8482 - val_loss: 0.3542 - val_acc: 0.8451
Epoch 13/100
44/44 [==============================] - 19s 427ms/step - loss: 0.3264 - acc: 0.8495 - val_loss: 0.2886 - val_acc: 0.8657
Epoch 14/100
44/44 [==============================] - 18s 420ms/step - loss: 0.3214 - acc: 0.8513 - val_loss: 0.3188 - val_acc: 0.8642
Epoch 15/100
44/44 [==============================] - 19s 423ms/step - loss: 0.3171 - acc: 0.8566 - val_loss: 0.2929 - val_acc: 0.8775
Epoch 16/100
44/44 [==============================] - 19s 421ms/step - loss: 0.3094 - acc: 0.8613 - val_loss: 0.2972 - val_acc: 0.8789
Epoch 17/100
44/44 [==============================] - 18s 420ms/step - loss: 0.3145 - acc: 0.8655 - val_loss: 0.3274 - val_acc: 0.8559
Epoch 18/100
44/44 [==============================] - 19s 421ms/step - loss: 0.3045 - acc: 0.8575 - val_loss: 0.3328 - val_acc: 0.8632
Epoch 19/100
44/44 [==============================] - 18s 420ms/step - loss: 0.3062 - acc: 0.8508 - val_loss: 0.3238 - val_acc: 0.8735
Epoch 20/100
44/44 [==============================] - 18s 418ms/step - loss: 0.3173 - acc: 0.8511 - val_loss: 0.2906 - val_acc: 0.8789
Epoch 21/100
44/44 [==============================] - 19s 430ms/step - loss: 0.3446 - acc: 0.8412 - val_loss: 0.3020 - val_acc: 0.8706
Epoch 22/100
44/44 [==============================] - 19s 425ms/step - loss: 0.3006 - acc: 0.8693 - val_loss: 0.2839 - val_acc: 0.8711
Epoch 23/100
44/44 [==============================] - 19s 421ms/step - loss: 0.3300 - acc: 0.8513 - val_loss: 0.2935 - val_acc: 0.8716
Epoch 24/100
44/44 [==============================] - 19s 422ms/step - loss: 0.3018 - acc: 0.8650 - val_loss: 0.2859 - val_acc: 0.8784
Epoch 25/100
44/44 [==============================] - 19s 421ms/step - loss: 0.3195 - acc: 0.8560 - val_loss: 0.3059 - val_acc: 0.8858
Epoch 26/100
44/44 [==============================] - 19s 424ms/step - loss: 0.2878 - acc: 0.8705 - val_loss: 0.3211 - val_acc: 0.8608
Epoch 27/100
44/44 [==============================] - 18s 419ms/step - loss: 0.3106 - acc: 0.8650 - val_loss: 0.2834 - val_acc: 0.8848
Epoch 28/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2862 - acc: 0.8722 - val_loss: 0.2783 - val_acc: 0.8873
Epoch 29/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2931 - acc: 0.8686 - val_loss: 0.3319 - val_acc: 0.8661
Epoch 30/100
44/44 [==============================] - 19s 422ms/step - loss: 0.3181 - acc: 0.8710 - val_loss: 0.2749 - val_acc: 0.8804
Epoch 31/100
44/44 [==============================] - 18s 420ms/step - loss: 0.3117 - acc: 0.8603 - val_loss: 0.2725 - val_acc: 0.8888
Epoch 32/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2800 - acc: 0.8738 - val_loss: 0.2775 - val_acc: 0.8814
Epoch 33/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2918 - acc: 0.8767 - val_loss: 0.3082 - val_acc: 0.8750
Epoch 34/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2961 - acc: 0.8764 - val_loss: 0.2797 - val_acc: 0.8931
Epoch 35/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2988 - acc: 0.8663 - val_loss: 0.2865 - val_acc: 0.8804
Epoch 36/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2843 - acc: 0.8741 - val_loss: 0.2912 - val_acc: 0.8681
Epoch 37/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2786 - acc: 0.8913 - val_loss: 0.3202 - val_acc: 0.8804
Epoch 38/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2923 - acc: 0.8828 - val_loss: 0.3547 - val_acc: 0.8504
Epoch 39/100
44/44 [==============================] - 19s 425ms/step - loss: 0.2845 - acc: 0.8797 - val_loss: 0.2609 - val_acc: 0.8951
Epoch 40/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2768 - acc: 0.8708 - val_loss: 0.2962 - val_acc: 0.8789
Epoch 41/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2832 - acc: 0.8742 - val_loss: 0.3270 - val_acc: 0.8618
Epoch 42/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2803 - acc: 0.8892 - val_loss: 0.2795 - val_acc: 0.8839
Epoch 43/100
44/44 [==============================] - 18s 417ms/step - loss: 0.2823 - acc: 0.8781 - val_loss: 0.3086 - val_acc: 0.8696
Epoch 44/100
44/44 [==============================] - 19s 422ms/step - loss: 0.2718 - acc: 0.8821 - val_loss: 0.2743 - val_acc: 0.8839
Epoch 45/100
44/44 [==============================] - 19s 422ms/step - loss: 0.2813 - acc: 0.8726 - val_loss: 0.2785 - val_acc: 0.8863
Epoch 46/100
44/44 [==============================] - 19s 426ms/step - loss: 0.2762 - acc: 0.8729 - val_loss: 0.2963 - val_acc: 0.8873
Epoch 47/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2745 - acc: 0.8788 - val_loss: 0.2978 - val_acc: 0.8839
Epoch 48/100
44/44 [==============================] - 19s 422ms/step - loss: 0.2869 - acc: 0.8644 - val_loss: 0.2859 - val_acc: 0.8765
Epoch 49/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2817 - acc: 0.8757 - val_loss: 0.2784 - val_acc: 0.8858
Epoch 50/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2641 - acc: 0.8845 - val_loss: 0.2880 - val_acc: 0.8814
Epoch 51/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2634 - acc: 0.8899 - val_loss: 0.2829 - val_acc: 0.8829
Epoch 52/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2620 - acc: 0.8923 - val_loss: 0.2857 - val_acc: 0.8853
Epoch 53/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2600 - acc: 0.8892 - val_loss: 0.2579 - val_acc: 0.8907
Epoch 54/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2625 - acc: 0.8877 - val_loss: 0.2685 - val_acc: 0.8922
Epoch 55/100
44/44 [==============================] - 18s 418ms/step - loss: 0.2654 - acc: 0.8897 - val_loss: 0.2827 - val_acc: 0.8898
Epoch 56/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2756 - acc: 0.8890 - val_loss: 0.2785 - val_acc: 0.8814
Epoch 57/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2664 - acc: 0.8809 - val_loss: 0.2720 - val_acc: 0.8833
Epoch 58/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2471 - acc: 0.8941 - val_loss: 0.2878 - val_acc: 0.8858
Epoch 59/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2808 - acc: 0.8765 - val_loss: 0.2772 - val_acc: 0.8951
Epoch 60/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2693 - acc: 0.8814 - val_loss: 0.2947 - val_acc: 0.8888
Epoch 61/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2755 - acc: 0.8816 - val_loss: 0.2853 - val_acc: 0.8863
Epoch 62/100
44/44 [==============================] - 19s 420ms/step - loss: 0.2609 - acc: 0.8840 - val_loss: 0.2893 - val_acc: 0.8839
Epoch 63/100
44/44 [==============================] - 19s 424ms/step - loss: 0.2641 - acc: 0.8941 - val_loss: 0.2722 - val_acc: 0.8833
Epoch 64/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2718 - acc: 0.8897 - val_loss: 0.2790 - val_acc: 0.8839
Epoch 65/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2693 - acc: 0.8892 - val_loss: 0.2812 - val_acc: 0.8804
Epoch 66/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2477 - acc: 0.8906 - val_loss: 0.2823 - val_acc: 0.8917
Epoch 67/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2648 - acc: 0.8876 - val_loss: 0.2855 - val_acc: 0.8882
Epoch 68/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2468 - acc: 0.8916 - val_loss: 0.2743 - val_acc: 0.8853
Epoch 69/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2669 - acc: 0.8873 - val_loss: 0.3376 - val_acc: 0.8740
Epoch 70/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2684 - acc: 0.8877 - val_loss: 0.2963 - val_acc: 0.8814
Epoch 71/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2765 - acc: 0.8809 - val_loss: 0.2834 - val_acc: 0.8819
Epoch 72/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2637 - acc: 0.8873 - val_loss: 0.2616 - val_acc: 0.8892
Epoch 73/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2620 - acc: 0.8816 - val_loss: 0.3063 - val_acc: 0.8907
Epoch 74/100
44/44 [==============================] - 19s 422ms/step - loss: 0.2547 - acc: 0.8856 - val_loss: 0.2578 - val_acc: 0.8824
Epoch 75/100
44/44 [==============================] - 19s 422ms/step - loss: 0.2622 - acc: 0.8921 - val_loss: 0.2998 - val_acc: 0.8750
Epoch 76/100
44/44 [==============================] - 19s 426ms/step - loss: 0.2692 - acc: 0.8718 - val_loss: 0.2757 - val_acc: 0.8951
Epoch 77/100
44/44 [==============================] - 18s 418ms/step - loss: 0.2541 - acc: 0.8921 - val_loss: 0.3113 - val_acc: 0.8789
Epoch 78/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2588 - acc: 0.8941 - val_loss: 0.2987 - val_acc: 0.8765
Epoch 79/100
44/44 [==============================] - 18s 417ms/step - loss: 0.2529 - acc: 0.8944 - val_loss: 0.2942 - val_acc: 0.8696
Epoch 80/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2686 - acc: 0.8899 - val_loss: 0.3009 - val_acc: 0.8839
Epoch 81/100
44/44 [==============================] - 18s 418ms/step - loss: 0.2401 - acc: 0.8900 - val_loss: 0.2969 - val_acc: 0.8755
Epoch 82/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2461 - acc: 0.8965 - val_loss: 0.3009 - val_acc: 0.8760
Epoch 83/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2553 - acc: 0.8852 - val_loss: 0.2709 - val_acc: 0.8990
Epoch 84/100
44/44 [==============================] - 18s 416ms/step - loss: 0.2444 - acc: 0.8972 - val_loss: 0.2939 - val_acc: 0.8789
Epoch 85/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2476 - acc: 0.8958 - val_loss: 0.2865 - val_acc: 0.8775
Epoch 86/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2500 - acc: 0.8916 - val_loss: 0.3126 - val_acc: 0.8809
Epoch 87/100
44/44 [==============================] - 18s 420ms/step - loss: 0.2721 - acc: 0.8873 - val_loss: 0.2520 - val_acc: 0.8941
Epoch 88/100
44/44 [==============================] - 18s 418ms/step - loss: 0.2437 - acc: 0.8989 - val_loss: 0.3003 - val_acc: 0.8701
Epoch 89/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2495 - acc: 0.8985 - val_loss: 0.2855 - val_acc: 0.8814
Epoch 90/100
44/44 [==============================] - 19s 421ms/step - loss: 0.2575 - acc: 0.8863 - val_loss: 0.2869 - val_acc: 0.8804
Epoch 91/100
44/44 [==============================] - 19s 425ms/step - loss: 0.2431 - acc: 0.8942 - val_loss: 0.2874 - val_acc: 0.8848
Epoch 92/100
44/44 [==============================] - 19s 422ms/step - loss: 0.2587 - acc: 0.8814 - val_loss: 0.2972 - val_acc: 0.8882
Epoch 93/100
44/44 [==============================] - 19s 425ms/step - loss: 0.2696 - acc: 0.8877 - val_loss: 0.2675 - val_acc: 0.8780
Epoch 94/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2461 - acc: 0.8991 - val_loss: 0.2844 - val_acc: 0.8863
Epoch 95/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2558 - acc: 0.8918 - val_loss: 0.2923 - val_acc: 0.8760
Epoch 96/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2850 - acc: 0.8639 - val_loss: 0.2860 - val_acc: 0.8843
Epoch 97/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2344 - acc: 0.8972 - val_loss: 0.2983 - val_acc: 0.8819
Epoch 98/100
44/44 [==============================] - 19s 423ms/step - loss: 0.2414 - acc: 0.8949 - val_loss: 0.3104 - val_acc: 0.8706
Epoch 99/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2613 - acc: 0.8845 - val_loss: 0.3363 - val_acc: 0.8720
Epoch 100/100
44/44 [==============================] - 18s 419ms/step - loss: 0.2595 - acc: 0.8880 - val_loss: 0.2911 - val_acc: 0.8853

训练结果：

阿杜依诺
关注
  • 2
    点赞
  • 踩
  • 13
    收藏
    觉得还不错? 一键收藏
  • 1
    评论
专栏目录
谷歌云可以使用windows操作系统吗？
03-31
可以在Google Cloud Platform上使用Windows操作系统。Google Cloud提供了多种Windows版本的镜像，包括Windows Server和Windows 10。
如何在谷歌云上面跑自己的深度学习代码
weixin_37721058的博客
02-28 7515
首先选择谷歌云的原因，一个字没钱，因为谷歌云会送你300美金，个人学术一年半载杠杠的 没问题。 接下来就是流程啦，首先创建谷歌云，然后创建VM实例，然后让自己的VM实例带上GPU。这个过程主要还是带上GPU，系统会自动帮你apply的，你只要等待审核就好，如果时间久了你就发邮件说你着急。对了，记得在申请的时候备注写着dissertation之类的字眼，告诉别人你是用来学术，那就一分钱都不用给你GP...
深度学习免费服务器 Google Colab使用教程
qq1440643730的博客
02-16 3492
对于入门深度学习的同学开说，笔记本GPU性能太弱是一件非常让人头疼的事，而租用的gpu服务器又太贵，最近刚发现google提供的免费云平台，可供我们做训练网络服务器用，GPU型号为Tesla T4 ，性能也是不俗。 首先，登陆Google Driver 然后，在空白处右击，查看是否有Google Colaboratory，没有的话，点击关联更多应用，搜索安装即可 新建一个文件夹，这里是yo...
OpenStack云计算基础架构平台搭建（国基北盛）：第一篇
最新发布
小吴同学的博客
02-23 472
小白型教程：OpenStack所需环境的基础搭建，Centos7操作系统的安装，VMware使用方法
给大家推荐一个谷歌的免费运行深度学习代码的服务器
tp_0moyi0的博客
03-03 4144
使用前提：需要自己的本机电脑可以翻墙 第一步： 打开链接 https://www.google.com/intl/zh-CN/drive/ 第二步： 邮件更多找到colab 直接点击可以直接使用，页面加载更多应用可以安装之后在使用 我这边选择的是安装 由于我这边已经安装了所以显示已经安装， 如果第一次使用这个没有安装的可以安装一下 之后打开便有这个文件，里面和nodebook是差不多的，第一次使用，后期观察观察 参考博客：Colab使用教程_逮仔的博...
Google 云端硬盘使用 colaboratory 跑深度学习（GPU）
Wonder233的学习博客
11-16 4952
Colaboratory 支持 Python 2.7 和 Python 3.6。 谷歌中文说明：https://colab.research.google.com/notebook 要使用云端硬盘，首先你必须翻墙！ 创建 Colaboratory 用 Google 账号登录进入 Google 云端硬盘 进入我的云端硬盘，创建一个文件夹（TensorFlow）。 进入创建好的文件夹，点开新建 ...
Colab入门and一些奇奇怪怪的点--写给和我一样的小白
weixin_44419049的博客
05-14 253
Colab入门and一些奇奇怪怪的点--写给和我一样的小白命令注册Google装载Google Drive切换目录colab不支持删除文件夹？特别注意！再次强调！持续更新ing 命令 colab里执行linux需要加前缀，一般是！ 注册Google 在注册时会遇到问题——需要验证非86的手机号。不过可以通过QQ邮箱进行注册，可免手机号验证。。。 但是！存在登录的时候可能也会让你身份验证的情况，这一般是因为google监测到你的账号异常（梯子不行 地址频繁改变的原因）这个时候就躲不过去了，只能验证，我遇到这个
Google Colab使用
m0_46692607的博客
03-14 1587
1. 每天可以免费使用 Tesla T4 12 小时CPU 配置2.深度学习的环境基本齐全，基本不需要额外配置环境3. 可以挂载 Google 云盘，方便保持代码，数据集，权重。
在服务器上进行深度学习的入门教程
hxs的博客
12-14 1万+
1.远程登录 Windows用户推荐使用Xshell 主机那里填上服务器IP，下面填一下端口号，然后点击连接，之后会让你输入用户名密码，输入之后就可以进入啦 2.GPU状态 进入之后可能大家最想关注的就是GPU占用情况，可以使用一下两种命令来查看一下~ watch -n 1 nvidia-smi 或者 watch --color --n 1 gpustat --color 其中，--n是代表刷新时间，1就是每秒刷新一次；--color是加颜色好看一点QwQ，去掉也可以 .
谷歌Colab跑深度学习网络数据实操步骤
lihengxuan123的博客
04-14 3721
谷歌Colab跑深度学习实操步骤。近期由于机房服务器很难排上号，在霍老师的指导下，开始用Colab进行跑代码。下面是我总结的实操步骤，仅供还未用过的同学入门参考学习，后续的一些使用技巧还请大家私下里相互多总结、多交流。
【深度学习】云服务器推荐及教程
陈嘿萌的博客
11-22 1万+
①智星云（付费）（用过还不错，可以按分钟计时，价格适中，能够使用不同显卡多设备，支持网页端和微信小程序，可以使用Pycharm专业版用SSH连接或者直接使用JupyterNotebook，缺点：传数据有点慢，不能自动保存上一次使用过的数据） 微信小程序： ②Google Colab（免费）（能够fq的同学，可以去白嫖谷歌的云服务器：https://colab.research.google.com/notebooks/welcome.ipynb） 满足条件的同学，可以学习一下如何使用Colab【别再买云
易语言谷歌云识别
07-19
易语言谷歌云识别源码,谷歌云识别,取指定内容
pycharm能否连接谷歌云虚拟机？
04-01
PyCharm可以连接到Google Cloud虚拟机。
谷歌云虚拟机启动以后不做计算是否要扣费？
03-30
是的，谷歌云虚拟机启动后即使不做计算也会产生费用。这是因为谷歌云计费基于虚拟机实例的运行时间和资源使用量。
VScode连接GitHub
热门推荐
学习笔记
10-13 1万+
主要参考如何利用VSCode优雅的使用Git提交至GitHub，本次主要说明VScode怎么发布文件到GitHub。 点击目标文件旁边的“+”，然后点击右上角的钩进行提交，然后点“···”进行推送，就可以push到GitHub上了。  ...
Keras计算F1_score
学习笔记
12-12 9647
实现代码： from keras.callbacks import Callback from sklearn.metrics import confusion_matrix, f1_score, precision_score, recall_score class Metrics(Callback): def on_train_begin(self, logs={}): ...
超分辨率（SRCNN）练习
学习笔记
03-31 5623
PSNR: 峰值信噪比（PSNR）直接借鉴参考程序会报错： return -10.*numpy.log10(k.mean(k.square(y_pred - y_true))) 'Tensor' object has no attribute 'log10' 查阅Keras文档和TensorFlow文档后，发现里面并没有log10函数，只有log，所以将程序改为如下即可成功（参考T...
使用谷歌云+宝塔面板搭建wordpress网站
05-11
通过谷歌云搭建自己的WordPress网站可以带来更好的性能体验和更高的可用性，同时，使用宝塔面板可以更方便地管理和配置自己的服务器。 要开始使用谷歌云和宝塔面板搭建WordPress网站，首先需要在谷歌云上创建一个...

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论 1
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值