如何将深度网络分成两个独立的网络？

Question

我学习了一个有两个输入的网络。它用作自动编码器。在网络的第一部分，输入被馈送到网络，经过一些处理并从高斯噪声层传递，使用网络的第二部分。在学习期间，所有网络一起学习，但为了测试，我需要将它分成两部分。第一部分有两个输入，第二个网络得到一个输入，它是第一个网络的输出。所以当我想为每个部分制作两个模型时，它说第二部分没有输入。你能告诉我该怎么做吗？是否可以为第二部分制作相同的网络但使用第一个网络的权重进行学习？我很快就会放代码。我在 keras 工作。谢谢

我的代码是：

wt_random=np.random.randint(2, size=(49999,4,4))
w_expand=wt_random.astype(np.float32)
wv_random=np.random.randint(2, size=(9999,4,4))
wv_expand=wv_random.astype(np.float32)
x,y,z=w_expand.shape
w_expand=w_expand.reshape((x,y,z,1))
x,y,z=wv_expand.shape
wv_expand=wv_expand.reshape((x,y,z,1))

#-----------------building w test---------------------------------------------
w_test = np.random.randint(2,size=(1,4,4))
w_test=w_test.astype(np.float32)
w_test=w_test.reshape((1,4,4,1))


#-----------------------encoder------------------------------------------------
#------------------------------------------------------------------------------
wtm=Input((4,4,1))
image = Input((28, 28, 1))
conv1 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl1e')(image)
conv2 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl2e')(conv1)
conv3 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl3e')(conv2)
BN=BatchNormalization()(conv3)
encoded =  Conv2D(1, (5, 5), activation='relu', padding='same',name='encoded_I')(BN)


wpad=Kr.layers.Lambda(lambda xy: xy[0] + Kr.backend.spatial_2d_padding(xy[1], padding=((0, 24), (0, 24))))
encoded_merged=wpad([encoded,wtm])


deconv1 = Conv2D(64, (5, 5), activation='elu', padding='same', name='convl1d')(encoded_merged)
deconv2 = Conv2D(64, (5, 5), activation='elu', padding='same', name='convl2d')(deconv1)
deconv3 = Conv2D(64, (5, 5), activation='elu',padding='same', name='convl3d')(deconv2)
deconv4 = Conv2D(64, (5, 5), activation='elu',padding='same', name='convl4d')(deconv3)
BNd=BatchNormalization()(deconv4)

decoded = Conv2D(1, (5, 5), activation='sigmoid', padding='same', name='decoder_output')(BNd) 

model=Model(inputs=[image,wtm],outputs=decoded)

decoded_noise = GaussianNoise(0.5)(decoded)

convw1 = Conv2D(64, (5,5), activation='relu', name='conl1w')(decoded_noise)#24
convw2 = Conv2D(64, (5,5), activation='relu', name='convl2w')(convw1)#20
convw3 = Conv2D(64, (5,5), activation='relu' ,name='conl3w')(convw2)#16
convw4 = Conv2D(64, (5,5), activation='relu' ,name='conl4w')(convw3)#12
convw5 = Conv2D(64, (5,5), activation='relu', name='conl5w')(convw4)#8
convw6 = Conv2D(64, (5,5), activation='relu', name='conl6w')(convw5)#4
convw7 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl7w',dilation_rate=(2,2))(convw6)#4
convw8 = Conv2D(64, (5,5), activation='relu', padding='same',name='conl8w',dilation_rate=(2,2))(convw7)#4
convw9 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl9w',dilation_rate=(2,2))(convw8)#4
convw10 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl10w',dilation_rate=(2,2))(convw9)#4
BNed=BatchNormalization()(convw10)
pred_w = Conv2D(1, (1, 1), activation='sigmoid', padding='same', name='reconstructed_W',dilation_rate=(2,2))(BNed)  
model2=Model(inputs=decoded_noise,outputs=pred_w)
w_extraction=Model(inputs=[image,wtm],outputs=[decoded,pred_w])

w_extraction.summary()

错误：

Traceback（最近一次调用最后一次）：

文件“”，第 55 行，在 model2=Model(inputs=decoded_noise,outputs=pred_w)

文件 "D:\software\Anaconda3\envs\py36\lib\site-packages\keras\legacy\interfaces.py", 第 91 行，在包装器中 返回函数(*args, **kwargs)

文件 "D:\software\Anaconda3\envs\py36\lib\site-packages\keras\engine\network.py", 第 93 行，在 init 中 self._init_graph_network(*args, **kwargs)

文件 "D:\software\Anaconda3\envs\py36\lib\site-packages\keras\engine\network.py", 第 231 行，在 _init_graph_network self.inputs, self.outputs)

文件 "D:\software\Anaconda3\envs\py36\lib\site-packages\keras\engine\network.py", 第 1443 行，在 _map_graph_network str(layers_with_complete_input))

ValueError: Graph disconnected: cannot get value for tensor Tensor("input_14:0", shape=(?, 28, 28, 1), dtype=float32) 在层 “输入_14”。访问以下先前层没有问题： []

新代码

wtm=Input((4,4,1))
image = Input((28, 28, 1))

#your code:
conv1 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl1e')(image)
conv2 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl2e')(conv1)
conv3 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl3e')(conv2)
BN=BatchNormalization()(conv3)
encoded =  Conv2D(1, (5, 5), activation='relu', padding='same',name='encoded_I')(BN)


wpad=Kr.layers.Lambda(lambda xy: xy[0] + Kr.backend.spatial_2d_padding(xy[1], padding=((0, 24), (0, 24))))
encoded_merged=wpad([encoded,wtm])
#end of your code

encoder = Model([image, wtm], encoded_merged)

encoded_input = Input((28,28,1))

#your code
deconv1 = Conv2D(64, (5, 5), activation='elu', padding='same', name='convl1d')(encoded_input)
deconv2 = Conv2D(64, (5, 5), activation='elu', padding='same', name='convl2d')(deconv1)
deconv3 = Conv2D(64, (5, 5), activation='elu',padding='same', name='convl3d')(deconv2)
deconv4 = Conv2D(64, (5, 5), activation='elu',padding='same', name='convl4d')(deconv3)
BNd=BatchNormalization()(deconv4)

decoded = Conv2D(1, (5, 5), activation='sigmoid', padding='same', name='decoder_output')(BNd) 

#end of your code

decoder = Model(encoded_input, decoded)

decoded_input = Input((28,28,1))

#your code
decoded_noise = GaussianNoise(0.5)(decoded_input)

convw1 = Conv2D(64, (5,5), activation='relu', name='conl1w')(decoded_noise)#24
convw2 = Conv2D(64, (5,5), activation='relu', name='convl2w')(convw1)#20
convw3 = Conv2D(64, (5,5), activation='relu' ,name='conl3w')(convw2)#16
convw4 = Conv2D(64, (5,5), activation='relu' ,name='conl4w')(convw3)#12
convw5 = Conv2D(64, (5,5), activation='relu', name='conl5w')(convw4)#8
convw6 = Conv2D(64, (5,5), activation='relu', name='conl6w')(convw5)#4
convw7 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl7w',dilation_rate=(2,2))(convw6)#4
convw8 = Conv2D(64, (5,5), activation='relu', padding='same',name='conl8w',dilation_rate=(2,2))(convw7)#4
convw9 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl9w',dilation_rate=(2,2))(convw8)#4
convw10 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl10w',dilation_rate=(2,2))(convw9)#4
BNed=BatchNormalization()(convw10)
pred_w = Conv2D(1, (1, 1), activation='sigmoid', padding='same', name='reconstructed_W',dilation_rate=(2,2))(BNed)  

#end of your code
noiseNet = Model(inputs=decoded_input,outputs=pred_w)

#input for full nets
full_wtm = Input((4,4,1))
full_image = Input((28, 28, 1)) 

#encoded 
full_encoded = encoder([full_image, full_wtm])

#decoded
full_decoded = decoder(full_encoded)

#with noise
full_w = noiseNet(full_decoded)

#autoencoder
autoencoder = Model([full_image,full_wtm], full_decoded)

#full net
w_extraction = Model([full_image, full_wtm], [full_decoded, full_w])

(x_train, _), (x_test, _) = mnist.load_data()
x_validation=x_train[1:10000,:,:]
x_train=x_train[10001:60000,:,:]
#
x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.
x_validation = x_validation.astype('float32') / 255.
x_train = np.reshape(x_train, (len(x_train), 28, 28, 1))  # adapt this if using `channels_first` image data format
x_test = np.reshape(x_test, (len(x_test), 28, 28, 1))  # adapt this if using `channels_first` image data format
x_validation = np.reshape(x_validation, (len(x_validation), 28, 28, 1))

#---------------------compile and train the model------------------------------
#opt=SGD(momentum=0.99)
w_extraction.compile(optimizer='adam', loss={'decoder_output':'mse','reconstructed_W':'binary_crossentropy'}, loss_weights={'decoder_output': 0.45, 'reconstructed_W': 1.0},metrics=['mae'])
es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=20)
#rlrp = ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=20, min_delta=1E-4, verbose=1)
mc = ModelCheckpoint('best_model_5x5F_dil_Los751.h5', monitor='val_loss', mode='min', verbose=1, save_best_only=True)
history=w_extraction.fit([x_train,w_expand], [x_train,w_expand],
          epochs=200,
          batch_size=16, 
          validation_data=([x_validation,wv_expand], [x_validation,wv_expand]),
          callbacks=[TensorBoard(log_dir='E:/concatnatenetwork', histogram_freq=0, write_graph=False),es,mc])
w_extraction.summary()

产生的错误：

Traceback（最近一次调用最后一次）：

文件“”，第 136 行，在 w_extraction.compile(optimizer='adam', loss={'decoder_output':'mse','reconstructed_W':'binary_crossentropy'}, loss_weights={'decoder_output': 0.45, 'reconstructed_W': 1.0},metrics=['mae'])

文件 "D:\software\Anaconda3\envs\py36\lib\site-packages\keras\engine\training.py", 第 119 行，在编译中 str(self.output_names))

ValueError：损失字典中的未知条目：“decoder_output”。仅有的 预期以下键：['model_17', 'model_18']

Answer 1

理想情况下，您应该首先单独创建模型。

net1 = createNet1()
net2 = createNet2()

net2OutFrom1 = net2(net1.output)

entireModel = Model(net1.input, net2OutFrom1)

然后你训练entireModel，你可以自动使用net1和net2，没有任何麻烦。

当你的网被做成一个单一的网时这样做。

你需要创建一个新的输入：

net2Input = Input(input_shape)

然后通过第二层网络的所有层。

out = originalNet.layers[firstLayerOfNet2](net2Input)
out = originalNet.layers[secondLayerOfNet2](out)
out = originalNet.layers[thirdLayerOfNet2](out)
....

然后分别创建第二个网：

net2 = Model(net2Input, out)

仍然可以轻松创建第一个网络：

net1 = Model(originalNet.input, originalNet.layers[lastLayerOfNet1].output)

用你的例子

个人网

编码器

wtm=Input((4,4,1))
image = Input((28, 28, 1))

#your code:
conv1 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl1e')(image)
conv2 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl2e')(conv1)
conv3 = Conv2D(64, (5, 5), activation='relu', padding='same', name='convl3e')(conv2)
BN=BatchNormalization()(conv3)
encoded =  Conv2D(1, (5, 5), activation='relu', padding='same',name='encoded_I')(BN)


wpad=Kr.layers.Lambda(lambda xy: xy[0] + Kr.backend.spatial_2d_padding(xy[1], padding=((0, 24), (0, 24))))
encoded_merged=wpad([encoded,wtm])
#end of your code

encoder = Model([image, wtm], encoded_merged)

解码器

encoded_input = Input(shape_of_encoded_merged_without_batch_size)

#your code
deconv1 = Conv2D(64, (5, 5), activation='elu', padding='same', name='convl1d')(encoded_input)
deconv2 = Conv2D(64, (5, 5), activation='elu', padding='same', name='convl2d')(deconv1)
deconv3 = Conv2D(64, (5, 5), activation='elu',padding='same', name='convl3d')(deconv2)
deconv4 = Conv2D(64, (5, 5), activation='elu',padding='same', name='convl4d')(deconv3)
BNd=BatchNormalization()(deconv4)

decoded = Conv2D(1, (5, 5), activation='sigmoid', padding='same', name='decoder_output')(BNd) 

#end of your code

decoder = Model(encoded_input, decoded)

噪音网

decoded_input = Input(shape_of_decoded_without_batch_size)

#your code
decoded_noise = GaussianNoise(0.5)(decoded_input)

convw1 = Conv2D(64, (5,5), activation='relu', name='conl1w')(decoded_noise)#24
convw2 = Conv2D(64, (5,5), activation='relu', name='convl2w')(convw1)#20
convw3 = Conv2D(64, (5,5), activation='relu' ,name='conl3w')(convw2)#16
convw4 = Conv2D(64, (5,5), activation='relu' ,name='conl4w')(convw3)#12
convw5 = Conv2D(64, (5,5), activation='relu', name='conl5w')(convw4)#8
convw6 = Conv2D(64, (5,5), activation='relu', name='conl6w')(convw5)#4
convw7 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl7w',dilation_rate=(2,2))(convw6)#4
convw8 = Conv2D(64, (5,5), activation='relu', padding='same',name='conl8w',dilation_rate=(2,2))(convw7)#4
convw9 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl9w',dilation_rate=(2,2))(convw8)#4
convw10 = Conv2D(64, (5,5), activation='relu',padding='same', name='conl10w',dilation_rate=(2,2))(convw9)#4
BNed=BatchNormalization()(convw10)
pred_w = Conv2D(1, (1, 1), activation='sigmoid', padding='same', name='reconstructed_W',dilation_rate=(2,2))(BNed)  

#end of your code
noiseNet = Model(inputs=noiseInput,outputs=pred_w)

加入网络

#input for full nets
full_wtm = Input((4,4,1))
full_image = Input((28, 28, 1)) 

#encoded 
full_encoded = encoder([full_image, full_wtm])

#decoded
full_decoded = decoder(full_encoded)

#with noise
full_w = noiseNet(full_decoded)

#autoencoder
autoencoder = Model([full_image,full_wtm], full_decoded)

#full net
w_extraction = Model([full_image, full_wtm], [full_decoded, full_w])

---

您需要使用此解决方案再次进行训练。您训练的任何网络都会训练所有其他网络。