cipherGAN Tensorflow代码复现遇到的问题

0.背景

    2018年，Aidan N. Gomez往ICLR会议投稿Unsupervised cipher cracking using discrete GANs论文，此论文主要介绍了利用GAN破译移位密码和维吉尼亚密码，将GAN应用于离散数据。虽然移位密码和维吉尼亚密码属于古典密码，已属于能够被破译的密码，但是此篇是首次将密码分析和GAN联系起来。

   本文主要讨论复现代码过程中遇到的问题。

 原论文链接：《UNSUPERVISED CIPHER CRACKING USING DISCRETE GANS》

1801.04883.pdf (arxiv.org)

论文代码链接：GitHub - for-ai/CipherGAN: TensorFlow implementation of CipherGAN

环境：tensorflow 1.15.2（CPU版）

（但是其实源代码要求是GPU版本）

1. 复现过程中遇到的问题

        我在调试代码过程中，遇到了接连两个Windows fatal exception: access violation的错误，被困了好久，在这里记录一下解决办法，希望能帮助到一些人。o(*￣▽￣*)ブ

1.1 Windows fatal exception: access violation之 _preread_check

  在train.py执行过程中，输入以下示例命令：

python -m CipherGAN-master.train --output_dir=runs\vig345 --test_name="vigenere345-brown200-eval*"  --train_name="vigenere345-brown-train*" --hparam_sets=vigenere_brown_vocab_200

报错如下：

Windows fatal exception: access violation

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 84 in _preread_check

详细信息如下所示：

Windows fatal exception: access violation

Current thread 0x00002cb8 (most recent call first):

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 84 in _preread_check

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 178 in readline

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 214 in next

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 220 in __next__

  File "C:\Users\admin\Downloads\CipherGAN-master\CipherGAN-master\train_utils\vocab_utils.py", line 15 in read_vocab

  File "C:\Users\admin\Downloads\CipherGAN-master\CipherGAN-master\train.py", line 47 in _run_locally

  File "C:\Users\admin\Downloads\CipherGAN-master\CipherGAN-master\train.py", line 85 in main

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 178 in readline

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 214 in next

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 220 in __next__

  File "C:\Users\admin\Downloads\CipherGAN-master\CipherGAN-master\train_utils\vocab_utils.py", line 15 in read_vocab

  File "C:\Users\admin\Downloads\CipherGAN-master\CipherGAN-master\train.py", line 47 in _run_locally

  File "C:\Users\admin\Downloads\CipherGAN-master\CipherGAN-master\train.py", line 85 in main

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\absl\app.py", line 254 in _run_main

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\absl\app.py", line 308 in run

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\platform\app.py", line 40 in run

  File "C:\Users\admin\Downloads\CipherGAN-master\CipherGAN-master\train.py", line 89 in <module>

  File "E:\envi\anaconda3\envs\tf\lib\runpy.py", line 85 in _run_code

  File "E:\envi\anaconda3\envs\tf\lib\runpy.py", line 193 in _run_module_as_main

解决：

  该问题本质就是 在 read_vocab 函数内部，当尝试读取文件时，发生了错误。进一步调试代码，将语料库的文件路径打印，发现是路径出错。

  因此直接将filepath修改为语料库所在的绝对路径

（同时如果还报错就将读文件的gfile.FastGFile方式修改为open)

但是！！又报错权限不够：“PermissionError: [Errno 13] Permission denied:”

PermissionError: [Errno 13] Permission denied:

'C:\\Users\\admin\\Downloads\\CipherGAN-master\\CipherGAN-master\\data\\tmp\\data'

  这里是因为路径最终只定位到了文件夹，需要把txt文件的完整名称也加上。

  最终修改后如下所示：

​

1.2 Windows fatal exception: access violation 之 get_matching_files_v2

  然后又遇到了这个报错，详细信息如下：

Windows fatal exception: access violation

Current thread 0x00002e58 (most recent call first):

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 384 in get_matching_files_v2

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\lib\io\file_io.py", line 363 in get_matching_files

  File "E:\mxz\CipherGAN-master\CipherGAN-master\data\cipher_reader.py", line 26 in _input_fn

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\contrib\learn\python\learn\estimators\estimator.py", line 1038 in _train_model

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\contrib\learn\python\learn\estimators\estimator.py", line 524 in fit

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\util\deprecation.py", line 507 in new_func

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\contrib\learn\python\learn\experiment.py", line 886 in _call_train

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\contrib\learn\python\learn\experiment.py", line 389 in train

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\contrib\learn\python\learn\experiment.py", line 672 in train_and_evaluate

  File "E:\mxz\CipherGAN-master\CipherGAN-master\train.py", line 84 in _run_locally

  File "E:\mxz\CipherGAN-master\CipherGAN-master\train.py", line 88 in main

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\absl\app.py", line 254 in _run_main

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\absl\app.py", line 308 in run

  File "E:\envi\anaconda3\envs\tf\lib\site-packages\tensorflow_core\python\platform\app.py", line 40 in run

  File "E:\mxz\CipherGAN-master\CipherGAN-master\train.py", line 92 in <module>

  File "E:\envi\anaconda3\envs\tf\lib\runpy.py", line 85 in _run_code

  File "E:\envi\anaconda3\envs\tf\lib\runpy.py", line 193 in _run_module_as_main

解决：

        首先将flag的默认参数中dir的’/’改为’\\’

        然后将代码中的训练集和测试集的路径打印出来，如下图所示

        发现文件层次出现错误（就是按照当前的路径找不到）

        所以直接将生成的数据集tmp\data整体移到项目路径下（省事儿）

​

​

1.3 AttributeError: 'HParams' object has no attribute 'sample_length'

   详细报错信息如下：即在 HParams 对象中找不到名为 'sample_length' 的属性。

File "C:\Users\admin\Downloads\CipherGAN-master\CipherGAN-master\data\cipher_reader.py", line 63, in _input_fn

    X = tf.pad(X, [[0, 0], [0, params.sample_length - tf.shape(X)[1]]])

AttributeError: 'HParams' object has no attribute 'sample_length'

解决：

首先打印超参数如下，发现确实没有sample_length 属性╮(╯▽╰)╭

​

然后去看代码，发现应该是因为采用了default的超参数集合，所以导致了没有设置sample_length

​

  而我们需要使用的超参数合集应该是如下所示的：

​

根据超参数的设置，我们在输入的命令中加入超参数集的选定 ，即：

--hparam_sets=vigenere_brown_vocab_200

 

  最终得到的完整输入命令如下所示：

python -m CipherGAN-master.train --output_dir=runs\vig345 --test_name="vigenere345-brown200-eval*"  --train_name="vigenere345-brown-train*" --hparam_sets=vigenere_brown_vocab_200  --hparam_sets=vigenere_brown_vocab_200

​  然后程序终于跑通了，泪目/(ㄒoㄒ)/~~