keras 的 example 文件 mnist_swwae.py 解析
2024-06-05 17:14:54
首先，该代码在新版本下会运行失败，根据 https://github.com/keras-team/keras/pull/13712/commits，需要把文件C:\ProgramData\Miniconda3\pkgs\keras-base-2.3.1-py37_0\Lib\site-packages\keras\backend\tensorflow_backend.py 中的函数 _get_available_gpus，由
def _get_available_gpus():"""Get a list of available gpu devices (formatted as strings).# ReturnsA list of available GPU devices."""global _LOCAL_DEVICESif _LOCAL_DEVICES is None:if _is_tf_1():devices = get_session().list_devices()_LOCAL_DEVICES = [x.name for x in devices]else:_LOCAL_DEVICES = tf.config.experimental_list_devices()return [x for x in _LOCAL_DEVICES if 'device:gpu' in x.lower()]，修改为：
def _get_available_gpus():"""Get a list of available gpu devices (formatted as strings).# ReturnsA list of available GPU devices."""global _LOCAL_DEVICESif _LOCAL_DEVICES is None:if _is_tf_1():devices = get_session().list_devices()_LOCAL_DEVICES = [x.name for x in devices]elif int(tf.__version__.split('.')[1]) >= 1:devices = tf.config.list_logical_devices()_LOCAL_DEVICES = [x.name for x in devices]else:_LOCAL_DEVICES = tf.config.experimental_list_devices()return [x for x in _LOCAL_DEVICES if 'device:gpu' in x.lower()]神经网络结构如下：
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) (None, 1, 32, 32) 0
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, 8, 32, 32) 80 input_1[0][0]
__________________________________________________________________________________________________
elu_1 (ELU) (None, 8, 32, 32) 0 conv2d_1[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, 8, 32, 32) 72 elu_1[0][0]
__________________________________________________________________________________________________
elu_2 (ELU) (None, 8, 32, 32) 0 conv2d_2[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, 8, 32, 32) 72 elu_2[0][0]
__________________________________________________________________________________________________
add_1 (Add) (None, 8, 32, 32) 0 conv2d_1[0][0]conv2d_3[0][0]
__________________________________________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None, 8, 16, 16) 0 add_1[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 16, 16, 16) 1168 max_pooling2d_1[0][0]
__________________________________________________________________________________________________
elu_3 (ELU) (None, 16, 16, 16) 0 conv2d_4[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 16, 16, 16) 272 elu_3[0][0]
__________________________________________________________________________________________________
elu_4 (ELU) (None, 16, 16, 16) 0 conv2d_5[0][0]
__________________________________________________________________________________________________
conv2d_6 (Conv2D) (None, 16, 16, 16) 272 elu_4[0][0]
__________________________________________________________________________________________________
add_2 (Add) (None, 16, 16, 16) 0 conv2d_4[0][0]conv2d_6[0][0]
__________________________________________________________________________________________________
max_pooling2d_2 (MaxPooling2D) (None, 16, 8, 8) 0 add_2[0][0]
__________________________________________________________________________________________________
conv2d_7 (Conv2D) (None, 32, 8, 8) 4640 max_pooling2d_2[0][0]
__________________________________________________________________________________________________
elu_5 (ELU) (None, 32, 8, 8) 0 conv2d_7[0][0]
__________________________________________________________________________________________________
conv2d_8 (Conv2D) (None, 32, 8, 8) 1056 elu_5[0][0]
__________________________________________________________________________________________________
elu_6 (ELU) (None, 32, 8, 8) 0 conv2d_8[0][0]
__________________________________________________________________________________________________
conv2d_9 (Conv2D) (None, 32, 8, 8) 1056 elu_6[0][0]
__________________________________________________________________________________________________
add_3 (Add) (None, 32, 8, 8) 0 conv2d_7[0][0]conv2d_9[0][0]
__________________________________________________________________________________________________
max_pooling2d_3 (MaxPooling2D) (None, 32, 4, 4) 0 add_3[0][0]
__________________________________________________________________________________________________
conv2d_10 (Conv2D) (None, 64, 4, 4) 18496 max_pooling2d_3[0][0]
__________________________________________________________________________________________________
elu_7 (ELU) (None, 64, 4, 4) 0 conv2d_10[0][0]
__________________________________________________________________________________________________
conv2d_11 (Conv2D) (None, 64, 4, 4) 4160 elu_7[0][0]
__________________________________________________________________________________________________
elu_8 (ELU) (None, 64, 4, 4) 0 conv2d_11[0][0]
__________________________________________________________________________________________________
conv2d_12 (Conv2D) (None, 64, 4, 4) 4160 elu_8[0][0]
__________________________________________________________________________________________________
add_4 (Add) (None, 64, 4, 4) 0 conv2d_10[0][0]conv2d_12[0][0]
__________________________________________________________________________________________________
max_pooling2d_4 (MaxPooling2D) (None, 64, 2, 2) 0 add_4[0][0]
__________________________________________________________________________________________________
conv2d_13 (Conv2D) (None, 128, 2, 2) 73856 max_pooling2d_4[0][0]
__________________________________________________________________________________________________
elu_9 (ELU) (None, 128, 2, 2) 0 conv2d_13[0][0]
__________________________________________________________________________________________________
conv2d_14 (Conv2D) (None, 128, 2, 2) 16512 elu_9[0][0]
__________________________________________________________________________________________________
elu_10 (ELU) (None, 128, 2, 2) 0 conv2d_14[0][0]
__________________________________________________________________________________________________
conv2d_15 (Conv2D) (None, 128, 2, 2) 16512 elu_10[0][0]
__________________________________________________________________________________________________
add_5 (Add) (None, 128, 2, 2) 0 conv2d_13[0][0]conv2d_15[0][0]
__________________________________________________________________________________________________
max_pooling2d_5 (MaxPooling2D) (None, 128, 1, 1) 0 add_5[0][0]
__________________________________________________________________________________________________
up_sampling2d_1 (UpSampling2D) (None, 128, 2, 2) 0 max_pooling2d_5[0][0]
__________________________________________________________________________________________________
lambda_5 (Lambda) (None, 128, 2, 2) 0 add_5[0][0]max_pooling2d_5[0][0]
__________________________________________________________________________________________________
multiply_1 (Multiply) (None, 128, 2, 2) 0 up_sampling2d_1[0][0]lambda_5[0][0]
__________________________________________________________________________________________________
conv2d_16 (Conv2D) (None, 64, 2, 2) 73792 multiply_1[0][0]
__________________________________________________________________________________________________
elu_11 (ELU) (None, 64, 2, 2) 0 conv2d_16[0][0]
__________________________________________________________________________________________________
conv2d_17 (Conv2D) (None, 64, 2, 2) 4160 elu_11[0][0]
__________________________________________________________________________________________________
elu_12 (ELU) (None, 64, 2, 2) 0 conv2d_17[0][0]
__________________________________________________________________________________________________
conv2d_18 (Conv2D) (None, 64, 2, 2) 4160 elu_12[0][0]
__________________________________________________________________________________________________
add_6 (Add) (None, 64, 2, 2) 0 conv2d_16[0][0]conv2d_18[0][0]
__________________________________________________________________________________________________
up_sampling2d_2 (UpSampling2D) (None, 64, 4, 4) 0 add_6[0][0]
__________________________________________________________________________________________________
lambda_4 (Lambda) (None, 64, 4, 4) 0 add_4[0][0]max_pooling2d_4[0][0]
__________________________________________________________________________________________________
multiply_2 (Multiply) (None, 64, 4, 4) 0 up_sampling2d_2[0][0]lambda_4[0][0]
__________________________________________________________________________________________________
conv2d_19 (Conv2D) (None, 32, 4, 4) 18464 multiply_2[0][0]
__________________________________________________________________________________________________
elu_13 (ELU) (None, 32, 4, 4) 0 conv2d_19[0][0]
__________________________________________________________________________________________________
conv2d_20 (Conv2D) (None, 32, 4, 4) 1056 elu_13[0][0]
__________________________________________________________________________________________________
elu_14 (ELU) (None, 32, 4, 4) 0 conv2d_20[0][0]
__________________________________________________________________________________________________
conv2d_21 (Conv2D) (None, 32, 4, 4) 1056 elu_14[0][0]
__________________________________________________________________________________________________
add_7 (Add) (None, 32, 4, 4) 0 conv2d_19[0][0]conv2d_21[0][0]
__________________________________________________________________________________________________
up_sampling2d_3 (UpSampling2D) (None, 32, 8, 8) 0 add_7[0][0]
__________________________________________________________________________________________________
lambda_3 (Lambda) (None, 32, 8, 8) 0 add_3[0][0]max_pooling2d_3[0][0]
__________________________________________________________________________________________________
multiply_3 (Multiply) (None, 32, 8, 8) 0 up_sampling2d_3[0][0]lambda_3[0][0]
__________________________________________________________________________________________________
conv2d_22 (Conv2D) (None, 16, 8, 8) 4624 multiply_3[0][0]
__________________________________________________________________________________________________
elu_15 (ELU) (None, 16, 8, 8) 0 conv2d_22[0][0]
__________________________________________________________________________________________________
conv2d_23 (Conv2D) (None, 16, 8, 8) 272 elu_15[0][0]
__________________________________________________________________________________________________
elu_16 (ELU) (None, 16, 8, 8) 0 conv2d_23[0][0]
__________________________________________________________________________________________________
conv2d_24 (Conv2D) (None, 16, 8, 8) 272 elu_16[0][0]
__________________________________________________________________________________________________
add_8 (Add) (None, 16, 8, 8) 0 conv2d_22[0][0]conv2d_24[0][0]
__________________________________________________________________________________________________
up_sampling2d_4 (UpSampling2D) (None, 16, 16, 16) 0 add_8[0][0]
__________________________________________________________________________________________________
lambda_2 (Lambda) (None, 16, 16, 16) 0 add_2[0][0]max_pooling2d_2[0][0]
__________________________________________________________________________________________________
multiply_4 (Multiply) (None, 16, 16, 16) 0 up_sampling2d_4[0][0]lambda_2[0][0]
__________________________________________________________________________________________________
conv2d_25 (Conv2D) (None, 8, 16, 16) 1160 multiply_4[0][0]
__________________________________________________________________________________________________
elu_17 (ELU) (None, 8, 16, 16) 0 conv2d_25[0][0]
__________________________________________________________________________________________________
conv2d_26 (Conv2D) (None, 8, 16, 16) 72 elu_17[0][0]
__________________________________________________________________________________________________
elu_18 (ELU) (None, 8, 16, 16) 0 conv2d_26[0][0]
__________________________________________________________________________________________________
conv2d_27 (Conv2D) (None, 8, 16, 16) 72 elu_18[0][0]
__________________________________________________________________________________________________
add_9 (Add) (None, 8, 16, 16) 0 conv2d_25[0][0]conv2d_27[0][0]
__________________________________________________________________________________________________
up_sampling2d_5 (UpSampling2D) (None, 8, 32, 32) 0 add_9[0][0]
__________________________________________________________________________________________________
lambda_1 (Lambda) (None, 8, 32, 32) 0 add_1[0][0]max_pooling2d_1[0][0]
__________________________________________________________________________________________________
multiply_5 (Multiply) (None, 8, 32, 32) 0 up_sampling2d_5[0][0]lambda_1[0][0]
__________________________________________________________________________________________________
conv2d_28 (Conv2D) (None, 1, 32, 32) 73 multiply_5[0][0]
__________________________________________________________________________________________________
elu_19 (ELU) (None, 1, 32, 32) 0 conv2d_28[0][0]
__________________________________________________________________________________________________
conv2d_29 (Conv2D) (None, 1, 32, 32) 2 elu_19[0][0]
__________________________________________________________________________________________________
elu_20 (ELU) (None, 1, 32, 32) 0 conv2d_29[0][0]
__________________________________________________________________________________________________
conv2d_30 (Conv2D) (None, 1, 32, 32) 2 elu_20[0][0]
__________________________________________________________________________________________________
add_10 (Add) (None, 1, 32, 32) 0 conv2d_28[0][0]conv2d_30[0][0]
__________________________________________________________________________________________________
activation_1 (Activation) (None, 1, 32, 32) 0 add_10[0][0]
==================================================================================================
Total params: 251,621
Trainable params: 251,621
Non-trainable params: 0
__________________________________________________________________________________________________该神经网络，我并没有完全搞明白其实际意义，我知道它是一个编解码器，其训练的输入和输出是一样的，比如都是x_train，根据某些地方的介绍说，这样的自动编解码器，解码效果会更清晰，在训练完成后可以看下效果；
之所以效果更好，是因为解码器中使用了编码器中的位置信息，也就是函数 getwhere 中对 MaxPooling2D 进行求导，求导结果就是相应的位置信息；
其他的，以后慢慢悟吧
keras 的 example 文件 mnist_swwae.py 解析相关推荐
- keras 的 example 文件 cnn_seq2seq.py 解析
该代码是实现一个翻译功能,好像是英语翻译为法语,嗯,我看不懂法语 首先这个代码有一个bug,本人提交了一个pull request来修复, https://github.com/keras-team/ ...
- keras 的 example 文件 cifar10_resnet.py 解析
该代码功能是卷积神经网络进行图像识别,数据集是cifar10 同时演示了回调函数 ModelCheckpoint, LearningRateScheduler, ReduceLROnPlateau 的 ...
- keras 的 example 文件 babi_rnn.py 解析
该代码的目的和 https://blog.csdn.net/zhqh100/article/details/105193991 类似 数据集也是同一个数据集,只不过这个是从 qa2_two-suppo ...
- keras 的 example 文件 mnist_hierarchical_rnn.py 解析
很显然,我没有看懂 HRNN 是啥意思,没有去看论文,应该就是一种RNN结构的变形吧 网络结构如下: _________________________________________________ ...
- keras 的 example 文件 mnist_denoising_autoencoder.py 解析
mnist_denoising_autoencoder.py 是一个编解码神经网络,其意义就是如果图片中有噪点的话,可以去除噪点,还原图片 其编码网络为: ______________________ ...
- keras 的 example 文件 mnist_cnn.py 解析
mnist_cnn.py 基本上就是最简单的一个卷积神经网络了,其结构如下: _____________________________________________________________ ...
- keras 的 example 文件 imdb_bidirectional_lstm.py 解析
imdb是一个文本情感分析的数据集,通过评论来分析观众对电影是好评还是差评 其网络结构比较简单 ____________________________________________________ ...
- keras 的 example 文件 lstm_text_generation.py 解析
该程序是学习现有的文章,然后学习预测下个字符,这样一个字符一个字符的学会写文章 先打印下char_indices {'\n': 0, ' ': 1, '!': 2, '"': 3, &quo ...
- keras 的 example 文件 lstm_stateful.py 解析
该程序要通过一个LSTM来实现拟合窗口平均数的功能 先看输入输出数据, print(x_train[:10]) [[[-0.08453234]][[ 0.02169589]][[ 0.07949955 ...
最新文章
- r语言 bsda包_使用R语言creditmodel包进行Vintage分析或留存率分析
- 透视表提取不反复记录（3）-每组最小值
- Android4.3 屏蔽HOME按键返回桌面详解(源码环境下)
- 初学者学用Github
- /etc/resolv.conf服务器客户端DNS重要配置文件
- border绘制三角形
- Win11系统无法安装GPT分区的解决方法
- 学计算机必须学会模拟电路,2016年广西大学计算机与电子信息学院1304电路分析基础与模拟电子线路之电路分析基础复试笔试仿真模拟题...
- 代数学笔记4: Galois基本定理
- java, android的aes等加密库
- 系统自带不起眼但很强杀毒工具
- python的while嵌套循环_Python中嵌套的WHILE循环
- C语言项目-俄罗斯方块
- c语言的编程特点,c语言编程是什么？C语言编程的特点和应用
- linux 查看设备 usb设备驱动程序,Linux USB设备驱动程序未被探测
- IDEA插件系列（41）：Code Reading Note插件——代码阅读笔记
- 鲲鹏920服务器支持docker,鲲鹏云服务器上安装Dockerdocker-compose
- 爆：中国互联网的那些黑色产业链(转)
- springcloud采坑-jason序列化中的Date对象
- Java岗最全面试攻略，吃透这些技术栈Offer拿到手软