当前位置：网站首页>[Deep Learning 21 Days Learning Challenge] Memo: What does our neural network model look like? - detailed explanation of model.summary()

[Deep Learning 21 Days Learning Challenge] Memo: What does our neural network model look like? - detailed explanation of model.summary()

2022-08-04 06:05:00 【Live up to [email protected]】

活动地址：CSDN21天学习挑战赛

学完手写识别和服装分类,Want to stop for a while to digest what I've learned,也总结一下,今天就从keras的model.summary()Let's start the output！

1、model.summary()是什么

构建深度学习模型,我们会通过model.summary()输出模型各层的参数状况,Have we had just learned model as an example：
在这里插入图片描述

这里可以看出,model.summary()打印出的内容,is the same as the hierarchical relationship in which we build the model,Clothing classification model as an example：

#Build model code
model = models.Sequential([
    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)), #卷积层1,卷积核3*3
    layers.MaxPooling2D((2, 2)),                   #池化层1,2*2采样
    layers.Conv2D(64, (3, 3), activation='relu'),  #卷积层2,卷积核3*3
    layers.MaxPooling2D((2, 2)),                   #池化层2,2*2采样
    layers.Conv2D(64, (3, 3), activation='relu'),  #卷积层3,卷积核3*3
    layers.Flatten(),                      #Flatten层,连接卷积层与全连接层
    layers.Dense(64, activation='relu'),   #全连接层,特征进一步提取
    layers.Dense(10)                       #输出层,输出预期结果
])

在这里插入图片描述

2、model.summary()输出含义

Still take the clothing classification model as an example：

Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param # 
=================================================================
conv2d_6 (Conv2D)            (None, 26, 26, 32)        320   
#创建： layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)), #卷积层1,卷积核3*3
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 13, 13, 32)        0       
#创建：layers.MaxPooling2D((2, 2)), #池化层1,2*2采样 
_________________________________________________________________
conv2d_7 (Conv2D)            (None, 11, 11, 64)        18496    
#创建：layers.Conv2D(64, (3, 3), activation='relu'), #卷积层2,卷积核3*3 
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 5, 5, 64)          0     
#创建：layers.MaxPooling2D((2, 2)), #池化层2,2*2采样 
_________________________________________________________________
conv2d_8 (Conv2D)            (None, 3, 3, 64)          36928     
#创建：layers.Conv2D(64, (3, 3), activation='relu'), #卷积层3,卷积核3*3
_________________________________________________________________
flatten_2 (Flatten)          (None, 576)               0   
#创建：layers.Flatten(), #Flatten层,连接卷积层与全连接层 
_________________________________________________________________
dense_4 (Dense)              (None, 64)                36928  
#创建：layers.Dense(64, activation='relu'), #全连接层,特征进一步提取 
_________________________________________________________________
dense_5 (Dense)              (None, 10)                650       
#创建：layers.Dense(10) #输出层,输出预期结果
=================================================================
Total params: 93,322
Trainable params: 93,322
Non-trainable params: 0
_________________________________________________________________

Param：该层输入参数个数, So how did this number come about?？
a、The formula for calculating the number of parameters of the convolution layer：（卷积核长度*卷积核宽度*通道数+1）*卷积核个数
例：
第一个卷积层：(3*3*1+1)*32 = 320
第二个卷积层：(3*3*32+1)*64 = 18496
第三个卷积层：(3*3*64+1)*64 = 36928
b、The formula for calculating the number of parameters of the fully connected layer： （输入数据维度+1）* 神经元个数
例：
Fully connected layer before the output layer：（64+1）*10=650
这里之所以要加1,Because every neuron has a偏置（Bias）.
Output Shape ：The output data shape of this layer
Total params: Total number of model parameters,The parameters of each layer are accumulated
Trainable params: 模型可训练参数
Non-trainable params:Model untrainable parameters