CNN 有多少个隐藏层？答案

【问题标题】：How many hidden layers a CNN has?CNN 有多少个隐藏层？
【发布时间】：2020-07-20 20:38:00
【问题描述】：

我使用 CNN 来解决分类问题。模型架构代码如下：

model.add(Conv1D(256, 5,padding='same',
                 input_shape=(40,1)))
model.add(Activation('relu'))
model.add(Conv1D(128, 5,padding='same'))
model.add(Activation('relu'))
model.add(Dropout(0.1))
model.add(MaxPooling1D(pool_size=(8)))
model.add(Conv1D(128, 5,padding='same',))
model.add(Activation('relu'))
model.add(Conv1D(128, 5,padding='same',))
model.add(Activation('relu'))
model.add(Flatten())
model.add(Dense(8))
model.add(Activation('softmax'))
opt = keras.optimizers.rmsprop(lr=0.00001, decay=1e-6)

这个模型有多少隐藏层？还有哪个是输出层和输入层？

【问题讨论】：

我相信每个 .add 函数都被认为是一个输入层，除了最后一个（激活）。但是，这个问题并不一定属于堆栈溢出。请在此处考虑此链接tutorialspoint.com/keras/keras_layers.htm

标签： python tensorflow keras layer conv-neural-network

【解决方案1】：

第一层是输入层，最后一层是输出层。介于这两者之间的是隐藏层。

model.add(Conv1D(256, 5,padding='same', input_shape=(40,1))) # input layer
model.add(Activation('relu')) # hidden layer
model.add(Conv1D(128, 5,padding='same')) # hidden layer
model.add(Activation('relu')) # hidden layer
model.add(Dropout(0.1)) # hidden layer
model.add(MaxPooling1D(pool_size=(8))) # hidden layer
model.add(Conv1D(128, 5,padding='same',)) # hidden layer 
model.add(Activation('relu')) # hidden layer
model.add(Conv1D(128, 5,padding='same',)) #hidden layer
model.add(Activation('relu')) # hidden layer
model.add(Flatten()) # hidden layer
model.add(Dense(8)) # hidden layer
model.add(Activation('softmax')) # output layer
opt = keras.optimizers.rmsprop(lr=0.00001, decay=1e-6)

【讨论】：

【解决方案2】：

输入层是第一层（指定 input_shape 的层）。每次使用 model.add 时都会创建一个新层。您可以使用 model.summary() 打印出您的模型层结构，如下所示。

Model: "sequential_8"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv1d_20 (Conv1D)           (None, 40, 256)           1536      
_________________________________________________________________
activation_23 (Activation)   (None, 40, 256)           0         
_________________________________________________________________
conv1d_21 (Conv1D)           (None, 40, 128)           163968    
_________________________________________________________________
activation_24 (Activation)   (None, 40, 128)           0         
_________________________________________________________________
dropout_6 (Dropout)          (None, 40, 128)           0         
_________________________________________________________________
max_pooling1d_4 (MaxPooling1 (None, 5, 128)            0         
_________________________________________________________________
conv1d_22 (Conv1D)           (None, 5, 128)            82048     
_________________________________________________________________
activation_25 (Activation)   (None, 5, 128)            0         
_________________________________________________________________
conv1d_23 (Conv1D)           (None, 5, 128)            82048     
_________________________________________________________________
activation_26 (Activation)   (None, 5, 128)            0         
_________________________________________________________________
flatten_3 (Flatten)          (None, 640)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 8)                 5128      
_________________________________________________________________
activation_27 (Activation)   (None, 8)                 0         
=================================================================
Total params: 334,728
Trainable params: 334,728
Non-trainable params: 0

这可能有点令人困惑，因为您的实际输出层是具有 8 个节点和 softmax 激活函数的层。我更喜欢按如下方式创建模型

inputs = tf.keras.Input(shape=(40,1))
x = tf.keras.layers.Conv1D(256, 5,padding='same', activation='relu')(inputs)
x=Dropout(.1)(x)
x=MaxPooling1D(pool_size=(8))(x)
x=Conv1D(128, 5,padding='same', activation='relu')(x)
x=Conv1D(128, 5,padding='same', activation='relu')(x)
x=Conv1D(128, 5,padding='same', activation='relu')(x)
x=Flatten()(x)
outputs=Dense(8, activation='softmax')(x)
model = tf.keras.Model(inputs=inputs, outputs=outputs)

It is the exact same model but I think it is clearer as to what layer is the actual output
See result below for model.summary()

> Blockquote
Model: "model_6"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_9 (InputLayer)         [(None, 40, 1)]           0         
_________________________________________________________________
conv1d_44 (Conv1D)           (None, 40, 256)           1536      
_________________________________________________________________
dropout_15 (Dropout)         (None, 40, 256)           0         
_________________________________________________________________
max_pooling1d_12 (MaxPooling (None, 5, 256)            0         
_________________________________________________________________
conv1d_45 (Conv1D)           (None, 5, 128)            163968    
_________________________________________________________________
conv1d_46 (Conv1D)           (None, 5, 128)            82048     
_________________________________________________________________
conv1d_47 (Conv1D)           (None, 5, 128)            82048     
_________________________________________________________________
flatten_11 (Flatten)         (None, 640)               0         
_________________________________________________________________
dense_12 (Dense)             (None, 8)                 5128      
=================================================================
Total params: 334,728
Trainable params: 334,728
Non-trainable params: 0

【讨论】：

谢谢！那么这个模型有多少隐藏层呢？
也输入层是第一个卷积层？还是它们是不同的层？
输入层是单独的层而不是卷积层