“导入 pydot 失败”抛出 kerasR

Question

我使用R下的keras包，我想知道是否有类似python的命令，带有plot_model（），可以显示其神经元网络

library(keras)

例如我想在R下显示这个神经网络

  model <- keras_model_sequential()
  model %>%
    layer_dense(units = 5, input_shape = 2) %>%
    layer_activation("relu") %>%
    layer_dense(units = 1)

我安装了 kerasR 包以使用函数 plot_model()，但是我有这个错误。

> library(kerasR)
> plot_model(model)

Error in py_call_impl(callable, dots$args, dots$keywords) : 
  ImportError: Failed to import pydot. You must install pydot and graphviz for `pydotprint` to work.

Detailed traceback: 
  File "C:\Users\Idriss\ANACON~1\envs\R-TENS~1\lib\site-packages\keras\utils\vis_utils.py", line 131, in plot_model
    dot = model_to_dot(model, show_shapes, show_layer_names, rankdir)
  File "C:\Users\Idriss\ANACON~1\envs\R-TENS~1\lib\site-packages\keras\utils\vis_utils.py", line 52, in model_to_dot
    _check_pydot()
  File "C:\Users\Idriss\ANACON~1\envs\R-TENS~1\lib\site-packages\keras\utils\vis_utils.py", line 27, in _check_pydot
    raise ImportError('Failed to import pydot. You must install pydot'

我使用的是 Windows 10 64 位，我使用带有 Anaconda 的 RStudio

---

在[4] pydot.Dot.create(pydot.Dot())

Out[4]: b"%!PS-Adobe-3.0\r\n%%Creator: graphviz version 2.38.0 (20140413.2041)\r\n%%Title: G\r\n%%Pages: (atend)\r\n%%BoundingBox: (atend)\r\n%%EndComments\r\nsave\r\n%%BeginProlog\r\n/DotDict 200 dict def\r\nDotDict begin\r\n\r\n/setupLatin1 {\r\nmark\r\n/EncodingVector 256 array def\r\n EncodingVector 0\r\n\r\nISOLatin1Encoding 0 255 getinterval putinterval\r\nEncodingVector 45 /hyphen put\r\n\r\n% Set up ISO Latin 1 character encoding\r\n/starnetISO {\r\n        dup dup findfont dup length dict begin\r\n        { 1 index /FID ne { def }{ pop pop } ifelse\r\n        } forall\r\n        /Encoding EncodingVector def\r\n        currentdict end definefont\r\n} def\r\n/Times-Roman starnetISO def\r\n/Times-Italic starnetISO def\r\n/Times-Bold starnetISO def\r\n/Times-BoldItalic starnetISO def\r\n/Helvetica starnetISO def\r\n/Helvetica-Oblique starnetISO def\r\n/Helvetica-Bold starnetISO def\r\n/Helvetica-BoldOblique starnetISO def\r\n/Courier starnetISO def\r\n/Courier-Oblique starnetISO def\r\n/Courier-Bold starnetISO def\r\n/Courier-BoldOblique starnetISO def\r\ncleartomark\r\n} bind def\r\n\r\n%%BeginResource: procset graphviz 0 0\r\n/coord-font-family /Times-Roman def\r\n/default-font-family /Times-Roman def\r\n/coordfont coord-font-family findfont 8 scalefont def\r\n\r\n/InvScaleFactor 1.0 def\r\n/set_scale {\r\n       dup 1 exch div /InvScaleFactor exch def\r\n       scale\r\n} bind def\r\n\r\n% styles\r\n/solid { [] 0 setdash } bind def\r\n/dashed { [9 InvScaleFactor mul dup ] 0 setdash } bind def\r\n/dotted { [1 InvScaleFactor mul 6 InvScaleFactor mul] 0 setdash } bind def\r\n/invis {/fill {newpath} def /stroke {newpath} def /show {pop newpath} def} bind def\r\n/bold { 2 setlinewidth } bind def\r\n/filled { } bind def\r\n/unfilled { } bind def\r\n/rounded { } bind def\r\n/diagonals { } bind def\r\n/tapered { } bind def\r\n\r\n% hooks for setting color \r\n/nodecolor { sethsbcolor } bind def\r\n/edgecolor { sethsbcolor } bind def\r\n/graphcolor { sethsbcolor } bind def\r\n/nopcolor {pop pop pop} bind def\r\n\r\n/beginpage {\t% i j npages\r\n\t/npages exch def\r\n\t/j exch def\r\n\t/i exch def\r\n\t/str 10 string def\r\n\tnpages 1 gt {\r\n\t\tgsave\r\n\t\t\tcoordfont setfont\r\n\t\t\t0 0 moveto\r\n\t\t\t(\\() show i str cvs show (,) show j str cvs show (\\)) show\r\n\t\tgrestore\r\n\t} if\r\n} bind def\r\n\r\n/set_font {\r\n\tfindfont exch\r\n\tscalefont setfont\r\n} def\r\n\r\n% draw text fitted to its expected width\r\n/alignedtext {\t\t\t% width text\r\n\t/text exch def\r\n\t/width exch def\r\n\tgsave\r\n\t\twidth 0 gt {\r\n\t\t\t[] 0 setdash\r\n\t\t\ttext stringwidth pop width exch sub text length div 0 text ashow\r\n\t\t} if\r\n\tgrestore\r\n} def\r\n\r\n/boxprim {\t\t\t\t% xcorner ycorner xsize ysize\r\n\t\t4 2 roll\r\n\t\tmoveto\r\n\t\t2 copy\r\n\t\texch 0 rlineto\r\n\t\t0 exch rlineto\r\n\t\tpop neg 0 rlineto\r\n\t\tclosepath\r\n} bind def\r\n\r\n/ellipse_path {\r\n\t/ry exch def\r\n\t/rx exch def\r\n\t/y exch def\r\n\t/x exch def\r\n\tmatrix currentmatrix\r\n\tnewpath\r\n\tx y translate\r\n\trx ry scale\r\n\t0 0 1 0 360 arc\r\n\tsetmatrix\r\n} bind def\r\n\r\n/endpage { showpage } bind def\r\n/showpage { } def\r\n\r\n/layercolorseq\r\n\t[\t% layer color sequence - darkest to lightest\r\n\t\t[0 0 0]\r\n\t\t[.2 .8 .8]\r\n\t\t[.4 .8 .8]\r\n\t\t[.6 .8 .8]\r\n\t\t[.8 .8 .8]\r\n\t]\r\ndef\r\n\r\n/layerlen layercolorseq length def\r\n\r\n/setlayer {/maxlayer exch def /curlayer exch def\r\n\tlayercolorseq curlayer 1 sub layerlen mod get\r\n\taload pop sethsbcolor\r\n\t/nodecolor {nopcolor} def\r\n\t/edgecolor {nopcolor} def\r\n\t/graphcolor {nopcolor} def\r\n} bind def\r\n\r\n/onlayer { curlayer ne {invis} if } def\r\n\r\n/onlayers {\r\n\t/myupper exch def\r\n\t/mylower exch def\r\n\tcurlayer mylower lt\r\n\tcurlayer myupper gt\r\n\tor\r\n\t{invis} if\r\n} def\r\n\r\n/curlayer 0 def\r\n\r\n%%EndResource\r\n%%EndProlog\r\n%%BeginSetup\r\n14 default-font-family set_font\r\n1 setmiterlimit\r\n% /arrowlength 10 def\r\n% /arrowwidth 5 def\r\n\r\n% make sure pdfmark is harmless for PS-interpreters other than Distiller\r\n/pdfmark where {pop} {userdict /pdfmark /cleartomark load put} ifelse\r\n% make '<<' and '>>' safe on PS Level 1 devices\r\n/languagelevel where {pop languagelevel}{1} ifelse\r\n2 lt {\r\n    userdict (<<) cvn ([) cvn load put\r\n    userdict (>>) cvn ([) cvn load put\r\n} if\r\n\r\n%%EndSetup\r\nsetupLatin1\r\n%%Page: 1 1\r\n%%PageBoundingBox: 36 36 44 44\r\n%%PageOrientation: Portrait\r\n0 0 1 beginpage\r\ngsave\r\n36 36 8 8 boxprim clip newpath\r\n1 1 set_scale 0 rotate 40 40 translate\r\nendpage\r\nshowpage\r\ngrestore\r\n%%PageTrailer\r\n%%EndPage: 1\r\n%%Trailer\r\n%%Pages: 1\r\n%%BoundingBox: 36 36 44 44\r\nend\r\nrestore\r\n%%EOF\r\n"

Answer 1

您应该安装 Python 库：

pip install pydot graphviz 

您还需要下载 graphviz 二进制文件，这些文件没有随 Python 一起安装。

在 Ubuntu 上，您可以使用 apt 安装它们：

apt-get install -y graphviz libgraphviz-dev

在 osX 上使用 brew：

brew install graphviz

对于 Windows 和其他操作系统，可以在 http://www.graphviz.org/ 找到说明

Answer 2

首先，错误：

  File "C:\Users\Idriss\ANACON~1\envs\R-TENS~1\lib\site-packages\keras\utils\vis_utils.py", line 52, in model_to_dot
    _check_pydot()

如果我们检查文件C:\Users\Idriss\ANACON~1\envs\R-TENS~1\lib\site-packages\keras\utils\vis_utils.py并搜索函数_check_pydot()：

def _check_pydot():
    try:
        # Attempt to create an image of a blank graph
        # to check the pydot/graphviz installation.
        pydot.Dot.create(pydot.Dot())
    except Exception:
        # pydot raises a generic Exception here,
        # so no specific class can be caught.
        raise ImportError('Failed to import pydot. You must install pydot'
                          ' and graphviz for `pydotprint` to work.')

此错误消息缺少信息，因为它捕获所有exception 而不是特定的执行并引发硬编码错误ImportError(Failed to import blah blah)。

为确保它导入相关的pydot，我们还应该检查该文件中的导入部分（重新运行R 和library(kerasR) 进行测试）：

import os
print("hole 0")
try:
    # pydot-ng is a fork of pydot that is better maintained.
    import pydot_ng as pydot
    print("hole 1")

except ImportError:
    # pydotplus is an improved version of pydot
    try:
        print("hole 1.2")
        import pydotplus as pydot
        print("hole 2")

    except ImportError:
        # Fall back on pydot if necessary.
        try:
            print("hole 3")
            import pydot
        except ImportError:
            print("hole 4")
            pydot = None
print("hole -1: " + str(locals())) #alternative way to debug
...

提示：更安全的调试方法是使用repr 而不是str。

如果你在交互模式下手动运行python并执行pydot.Dot.create(pydot.Dot())，你会发现确切的异常（下面是我的Linux示例）：

xb@dnxb:~/anaconda3/envs/r-tensorflow/bin$ ./python
Python 3.6.3 |Anaconda, Inc.| (default, Nov 20 2017, 20:41:42) 
[GCC 7.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import pydot
>>> pydot.Dot()
<pydot.Dot object at 0x7f7d045cdb38>
>>> pydot.Dot
<class 'pydot.Dot'>
>>> pydot.Dot.create(pydot.Dot())
Traceback (most recent call last):
  File "/home/xiaobai/anaconda3/envs/r-tensorflow/lib/python3.6/site-packages/pydot.py", line 1878, in create
    stderr=subprocess.PIPE, stdout=subprocess.PIPE)
  File "/home/xiaobai/anaconda3/envs/r-tensorflow/lib/python3.6/subprocess.py", line 709, in __init__
    restore_signals, start_new_session)
  File "/home/xiaobai/anaconda3/envs/r-tensorflow/lib/python3.6/subprocess.py", line 1344, in _execute_child
    raise child_exception_type(errno_num, err_msg, err_filename)
FileNotFoundError: [Errno 2] No such file or directory: 'dot': 'dot'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/xiaobai/anaconda3/envs/r-tensorflow/lib/python3.6/site-packages/pydot.py", line 1883, in create
    prog=prog))
Exception: "dot" not found in path.
>>>

让我们打印文件/home/xiaobai/anaconda3/envs/r-tensorflow/lib/python3.6/site-packages/pydot.py 1878 行之前使用的一些变量：

try:
    print("env: " + str(env))
    print("cmdline: " + str(cmdline))
    print("tmp_dir: " + str(tmp_dir))
    p = subprocess.Popen(
        cmdline,
        env=env,
        cwd=tmp_dir,
        shell=False,
        stderr=subprocess.PIPE, stdout=subprocess.PIPE)
except OSError as e:
    if e.errno == os.errno.ENOENT:
        raise Exception(
            '"{prog}" not found in path.'.format(
                prog=prog))
    else:
        raise

重启你的python解释器，重新运行import pydot和pydot.Dot.create(pydot.Dot())，会显示：

xb@dnxb:~/anaconda3/envs/r-tensorflow/bin$ ./python
Python 3.6.3 |Anaconda, Inc.| (default, Nov 20 2017, 20:41:42) 
[GCC 7.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import pydot
>>> pydot.Dot.create(pydot.Dot())
env: {'PATH': '/home/xiaobai/anaconda3/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:...<My other PATH>'}
cmdline: ['dot', '-Tps', '/tmp/tmpffo17gx5']
tmp_dir: /tmp
Traceback (most recent call last):
  File "/home/xiaobai/anaconda3/envs/r-tensorflow/lib/python3.6/site-packages/pydot.py", line 1881, in create
    stderr=subprocess.PIPE, stdout=subprocess.PIPE)

所以基本上它的作用是运行命令dot -Tps /tmp/tmpffo17gx5，但由于找不到dot 命令而失败。

在 Linux 中，如果我在终端手动运行命令，它会建议运行 sudo apt install graphviz 来安装 dot：

xb@dnxb:~/anaconda3/envs/r-tensorflow/bin$ dot
The program 'dot' is currently not installed. You can install it by typing:
sudo apt install graphviz
xb@dnxb:~/anaconda3/envs/r-tensorflow/bin$ sudo apt install graphviz
...

现在运行dot -Tps /tmp/tmpffo17gx5会成功：

xb@dnxb:~/anaconda3/envs/r-tensorflow/bin$ dot -Tps /tmp/tmpffo17gx5 
%!PS-Adobe-3.0
%%Creator: graphviz version 2.38.0 (20140413.2041)
%%Title: G
%%Pages: (atend)
%%BoundingBox: (atend)
%%EndComments
save
%%BeginProlog
/DotDict 200 dict def
DotDict begin
...

重启 R 会话，不再报错：

> plot_model(model)
> 

这个sudo apt install graphviz 是针对 Linux 的，但我希望这个答案可以帮助您调试 Windows 中的错误。