Pyinstaller exe conversion fail - using lightgbm and sklearn - python

I am trying to convert my code into an exe using pyinstaller spec. Pyinstaller initially failed with the following reason:
main__.PyInstallerImportError: Failed to load dynlib/dll
'C:\\Users\\...\\lightgbm\\../compile\\lib_lightgbm.dll'. Most probably this
dynlib/dll was not found when the application was frozen.
I tried to correct it by adding the following line to my pathex list in the spec:
'myenv\\lib\\site-packages\\lightgbm'
Note: myenv is my virtualenv created for this project.
Which led to ImportError for sklearn. I added sklearn to hidden imports. This is my final spec fie:
# -*- mode: python -*-
import sys
sys.setrecursionlimit(5000)
block_cipher = None
a = Analysis(['myscript.py'],
pathex=['C:\\project_folder', 'myenv\\lib\\site-packages\\lightgbm'],
binaries=[],
datas=[('lib_lightgbm.dll', '.')],
hiddenimports=['cython', 'sklearn', 'sklearn.feature_extraction','sklearn.pipeline', 'sklearn.ensemble', 'sklearn.neighbors.typedefs', 'sklearn.neighbors.quad_tree', 'sklearn.tree._utils'],
hookspath=[],
runtime_hooks=[],
excludes=[],
win_no_prefer_redirects=False,
win_private_assemblies=False,
cipher=block_cipher,
noarchive=False)
pyz = PYZ(a.pure, a.zipped_data,
cipher=block_cipher)
exe = EXE(pyz,
a.scripts,
[],
exclude_binaries=True,
name='myscript',
debug=False,
bootloader_ignore_signals=False,
strip=False,
upx=True,
console=True )
coll = COLLECT(exe,
a.binaries,
a.zipfiles,
a.datas,
strip=False,
upx=True,
name='myscript')
This leads to an error I find very weird:
File "myenv\lib\site-packages\lightgbm\sklearn.py", line 9, in <module>
ImportError: attempted relative import with no known parent package
[12692] Failed to execute script myscript
I have no idea why it would be searching for sklearn.py inside lightgbm?
There is sklearn.py inside lightgbm, but I am not sure how to fix this error. Also, there is no sklearn folder in the dist/myscript folder. Ideally it should have been present. I tried to manually copy the sklearn folder, although, as suspected, it made no difference. Could someone please let me know where the spec file is wrong?

Let's go back to the first error you had. That error suggests lib_lightgbm.dll was not seen when your application was frozen.
Assuming you run pyinstaller from Windows cmd.exe, you can fix this by passing a hook for lightgbm to pyinstaller, so that it knows where to get it from, e.g.
pyinstaller --additional-hooks-dir dir_with_lightgbm_hook --hidden-import lightgbm myscript.py
The name of the hook should be hook-lightgbm.py and for its contents, you can look here: https://github.com/pyinstaller/pyinstaller/blob/develop/PyInstaller/hooks/hook-numpy.py , it should be something similar. Just make sure that the library is added to the binaries list, i.e.
binaries.append((os.path.join(os.path.dirname(
get_module_file_attribute('lightgbm')), "lib_lightgbm.dll"), "lightgbm"))
The second "lightgbm" argument is the location where the .dll will be copied, relative to your distribution directory, so in your case it should end up as dist\myscript\lightgbm\lib_lightgbm.dll.
Note: Adding lib_lightgbm.dll through a hook will make pyinstaller add the dependencies for it to the distribution, so it is better to do it like this than just copying it manually.
Second note: When copying your distribution package to another machine, you might need to install the appropriate Visual Studio redistributable used by the library (e.g. VS 2015 redistributable).
EDIT: Forgot to mention that if you do it like this, you can remove the path you added to your pathex.

Related

Create an executable from Python code with gmsh import

I'm trying to package up my Python package into an executable using pyinstaller. The script name is called "run-jointbuilder.py" The package has a number of dependancies (such as numpy), but importantly gmsh.
When using pyinstaller to compile my code, it appears to be successful, but then when I try to run the executable I get the following errors:
import gmsh # PyInstaller PYZ\
import ctypes.util # PyInstaller PYZ\
import 'ctypes.util' # <pyimod03_importers.FrozenImporter object at 0x000001BD783FC910>\
Traceback (most recent call last):\
File "PyInstaller\loader\pyiboot01_bootstrap.py", line 144, in __init__
File "ctypes\__init__.py", line 381, in __init__\
FileNotFoundError: Could not find module 'C:\Users\willber\Anaconda3\Scripts\gmsh' (or one of its dependencies). Try using the full path with constructor syntax.
I then get this error:
__main__.PyInstallerImportError: Failed to load dynlib/dll
'C:\\Users\\willber\\Anaconda3\\Scripts\\gmsh'. Most probably this dynlib/dll was not found when the application was frozen.
[18612] Failed to execute script run-jointbuilder
Has anybody tried to compile some Python code that imports the gmsh package? I'd really appreciate an example .spec file, for use with pyinstaller if so!
The gmsh python package wraps a bunch of compiled libraries which contain the implementations of the methods you call from python. When you import gmsh.py into your script then gmsh loads these libraries in the background giving you access to their functionality through python methods. So it's essential that these libraries are embedded in the pyinstaller output for your code to function as it does when you run it directly through the python interpreter.
It's difficult for pyinstaller to consistently find these libraries since they're not accessed in the same way as normal python imports, they are loaded using the cytpes package. There's some description of how pyinstaller does this in the docs. Since you're seeing a dynlib/dll loading error when you run the compiled python script, this suggests that pyinstaller isn't finding the gmsh library during compile and hence it is missing from the executable.
If you look in the gmsh.py source, you can see that gmsh.py loads up a .dll library called gmsh-4.9.dll for Windows OS. You can use the binaries input of the pyinstaller .spec file to point the compiler to the gmsh-4.9.dll.
Here's an example .spec file which dynamically locates the gmsh-4.9.dll at the time of compile so it picks up the right .dll for your active environment. You could make this more generic by filtering all *.dll in the gmsh directory, but I've hardcoded for clarity:
# -*- mode: python ; coding: utf-8 -*-
from pathlib import Path
import gmsh
# get the location of the gmsh dll which sits next to the gmsh.py file
libname = 'gmsh-4.9.dll'
libpath = Path(gmsh.__file__).parent / libname
print('Adding {} to binaries'.format(libpath))
block_cipher = None
a = Analysis(['gmsh-test.py'],
pathex=['C:\\Users\\user\\dev\\gmsh'],
# tell pyinstaller to add the binary to the compiled path
binaries=[(str(libpath), '.')],
datas=[],
hiddenimports=[],
hookspath=[],
runtime_hooks=[],
excludes=[],
win_no_prefer_redirects=False,
win_private_assemblies=False,
cipher=block_cipher,
noarchive=False)
pyz = PYZ(a.pure, a.zipped_data,
cipher=block_cipher)
exe = EXE(pyz,
a.scripts,
a.binaries,
a.zipfiles,
a.datas,
name='gmsh-test',
debug=False,
bootloader_ignore_signals=False,
strip=False,
upx=True,
console=True,
runtime_tmpdir=None,
)

How to package python and kivy application for Raspbian Buster

I am currently writing an application for a raspberry pi 3 using python3.7 and kivy1.11.1. I am trying to package a test application to ensure that it will work. I have created a simple grid with 6 buttons. When you click on a button it prints "button pressed" to the screen. I am using a separate kv file to lay this out as this will be more practical as the application develops. So I have two files, "test.py" and "Test.kv".
I have been using pyinstaller to package my app but it is not working. It runs and creates the executable file just fine with no errors. I then drag my executable file out to my main folder from the "dist" folder as recommended in this video.
My command is:
pyinstaller --onefile --add-data "Test.kv:." test.py
I have also tried:
pyinstaller --onedir --add-data "Test.kv:." test.py
which does the same: blank window no widgets, no errors.
When I try to execute it nothing happens. I don't know what is wrong because I'm not getting any errors when I try to execute it from the terminal. When I do the exact same process on windows it works so its not the code. On occasion, I think it does eventually run but I just get a blank screen and again no error. This would leave me to believe it's an issue with reading the .kv file.
Perhaps this process doesn't work with kivy and I need to use some other method to package my application. However I have also tried this kind of approach but it has left me with errors. I much prefer the pyinstaller method however I don't know if it is going to work as it doesn't seem to pick up the .kv file.
Any help/advice on this issue would be greatly appreciated?
My test.spec file:
# -*- mode: python ; coding: utf-8 -*-
block_cipher = None
a = Analysis(['test.py'],
pathex=['/home/pi/Desktop/WinTest'],
binaries=[],
datas=[('Test.kv', '.')],
hiddenimports=[],
hookspath=[],
runtime_hooks=[],
excludes=[],
win_no_prefer_redirects=False,
win_private_assemblies=False,
cipher=block_cipher,
noarchive=False)
pyz = PYZ(a.pure, a.zipped_data,
cipher=block_cipher)
exe = EXE(pyz,
a.scripts,
a.binaries,
a.zipfiles,
a.datas,
[],
name='test',
debug=False,
bootloader_ignore_signals=False,
strip=False,
upx=True,
upx_exclude=[],
runtime_tmpdir=None,
console=False )
The following picture is my program folder (sorry for the poor quality). My library files are in venv/Lib/site-packages. This is where my kivy/kivy_deps folders are. Could the issue be that the program can't find my library files? But surely if that was the issue it would throw a "file not found" error? And as previously stated, I am not receiving any errors.
Found this in my warn-text.txt file
missing module named kivy.lib.vidcore_lite.egl - imported by kivy.lib.vidcore_lite (top-level), kivy.core.window.window_egl_rpi (top-level)
missing module named kivy.lib.vidcore_lite.bcm - imported by kivy.lib.vidcore_lite (top-level), kivy.core.window.window_egl_rpi (top-level)

Pyinstaller 3.5 will not generate a standalone executable

I am trying to get Pyinstaller 3.5 to create a "onefile" executable but it keeps generating a "onedir" instead.
My files are all in one directory. There is a main program that imports two other modules which in turn import a third. The program functions properly when run directly in Python 3.7.4. The "onedir" version generated by Pyinstaller also works. I'm running 64 bit Windows 10 Pro on a Surface Book 2.
The command I'm using to generate the file is:
pyinstaller --onefile --windowed --additional-hooks-dir=. qualys_admin.spec
My program uses wx, pubsub, xmltodict, requests, and pandas. The additional hook file is for xlrd which pandas needs to read a xlsx file.
My spec file looks like this:
# -*- mode: python ; coding: utf-8 -*-
block_cipher = None
# work-around for https://github.com/pyinstaller/pyinstaller/issues/4064
import distutils
if distutils.distutils_path.endswith('__init__.py'):
distutils.distutils_path = os.path.dirname(distutils.distutils_path)
addedfiles = ('qualys_admin.ini', '.')
a = Analysis(['qualys_admin.py'],
pathex=['C:\\Users\\secops-sw\\Documents\\qualys-
administration\\qualysadmin'],
binaries=[],
datas=[addedfiles],
hiddenimports=[],
hookspath=[],
runtime_hooks=[],
excludes=[],
win_no_prefer_redirects=False,
win_private_assemblies=False,
cipher=block_cipher,
noarchive=False)
pyz = PYZ(a.pure, a.zipped_data,
cipher=block_cipher)
exe = EXE(pyz,
a.scripts,
[],
exclude_binaries=True,
name='qualys_admin',
debug=False,
bootloader_ignore_signals=False,
strip=False,
upx=True,
console=True )
coll = COLLECT(exe,
a.binaries,
a.zipfiles,
a.datas,
strip=False,
upx=True,
upx_exclude=[],
name='qualys_admin')
I wrote a tiny "hello world" program to take the complexity of my program off the table. I was able to generate both a "onedir" and a "onefile" successfuly. So I know where to look for the standalone executable.
For my qualys_admin app, that executable is definitely not being generated and I cannot find any warnings to indicate why.
Does anyone have any ideas?
It is not working because your spec file is configured for one directory mode. You need to create the spec file with the one file flag.
pyi-makespec --onefile yourscript.py
Then you can modify your spec sheet and build the app using the custom spec sheet.

How do I resolve a missing utility module preventing my executable from running when I already installed the parent module?

I've written a few Python scripts to create a tkinter GUI for a machine learning algorithm process. I originally coded everything in PyCharm, but I'd really like to put everything together into a stand-alone executable. I've moved my main script and its .py dependencies into their own directory and tested it out using the Command Prompt, and it works great. However, when I run pyinstaller, the executable is created but fails on startup.
The program is made up of three files, with GUI.py being the main script. As mentioned above, I moved the dependent files into a new directory and tested GUI.py in the Command Prompt, and it worked great. Executable is created (albeit with a lot of warnings about missing 'api-ms-win-crt' files) but can't be run.
I created the executable using the command:
pyinstaller --onefile GUI.py
When the executable is run from the command line after creation, I get a big long traceback ending in the following:
File "site-packages\sklearn\metrics\pairwise.py", line 32, in <module>
File "sklearn\metrics\pairwise_fast.pyx", line 1, in init
sklearn.metrics.pairwise_fast
ModuleNotFoundError: No module named 'sklearn.utils._cython_blas'
[3372] Failed to execute script GUI
I know I've already explicitly imported sklearn through the command prompt, but from the traceback, it seems I'm missing a utility module somewhere. I tried to import the missing module specifically, but I got an error that no distributed module was available.
I don't have much experience with pyinstaller, and I have no idea where to go from here. I'm using Windows 10 and Python 3.7.3.
It seems that Pyinstaller can't resolve sklearn import. So one easy way is to just bring the whole module directory which located in <path_to_python>/Lib/site-packages/sklearn/ with executable output. So use below spec file to generate your executable:
# -*- mode: python -*-
block_cipher = None
a = Analysis(['test.py'],
pathex=['<path to root of your project>'],
binaries=[],
datas=[],
hiddenimports=[],
hookspath=[],
runtime_hooks=[],
excludes=[],
win_no_prefer_redirects=False,
win_private_assemblies=False,
cipher=block_cipher,
noarchive=False)
a.datas += Tree('<path_to_sklearn_in_python_dir>', prefix='sklearn')
pyz = PYZ(a.pure, a.zipped_data,
cipher=block_cipher)
exe = EXE(pyz,
a.scripts,
a.binaries,
a.zipfiles,
a.datas,
[],
name='test',
debug=False,
bootloader_ignore_signals=False,
strip=False,
upx=False,
runtime_tmpdir=None,
console=True )
Finally generate your executable with
pyinstaller test.spec
This should resolve import errors for sklearn but if you face other NotFound imports add them like above to spec file.
Building up on M.R.'s answer, you can directly include the path to sklearn in your original pyinstaller command:
pyinstaller --onefile GUI.py --add-data "<path-to-python>\Lib\site-packages\sklearn;sklearn"
which results in the following line of code being added inside a = Analysis() in the automatically-generated GUI.spec file:
datas=[('<path-to-python>\\Lib\\site-packages\\sklearn', 'sklearn')]
Note that the --onefile option will result in an executable that is slower to start up than the default one-folder bundle (based on both the pyinstaller documentation and my own experience bundling up sklearn):
pyinstaller GUI.py --add-data "<path-to-python>\Lib\site-packages\sklearn;sklearn"

Python: how to specify output folders in Pyinstaller .spec file

I am using python 3.5 and pyinstaller version 3.1.1. I have specified a .spec file, called SCADAsync_spec.spec, as follows:
block_cipher = None
a = Analysis(['SCADAsync.py'],
pathex=['C:\\repo\\analysis\\trunk\\source\\python\\functions', 'C:\\repo\\analysis\\trunk\\source\\python\\Executables'],
binaries=None,
datas=[('figs\\ROMO_icon.ico','figs'),('figs\\OpenFile2.gif','figs'),('figs\\ROMOWind_Logo2015_CMYK.png','figs')],
hiddenimports=[],
hookspath=[],
runtime_hooks=[],
excludes=[],
win_no_prefer_redirects=False,
win_private_assemblies=False,
cipher=block_cipher)
pyz = PYZ(a.pure, a.zipped_data,
cipher=block_cipher)
exe = EXE(pyz,
a.scripts,
a.binaries,
a.zipfiles,
a.datas,
name='SCADAsync',
debug=True,
strip=False,
upx=True,
console=True
)
That works fine when executed with
pyinstaller SCADAsync_spec.spec
Now that creates two large folders (dist and build) , which I would prefer to store elsewhere than in the default directory. Does anyone know how to set the location of those folders in the spec file? I'd like to keep my command line command as simple as possible, i.e. the .exe should build just by typing
pyinstaller SCADAsync_spec.spec
From the Pyinstaller manual it seems I can specify globals called 'DISTPATH' and 'workpath' to the spec file (https://pythonhosted.org/PyInstaller/spec-files.html). But I cannot really figure out how to do that.
Any help would be greatly appreciated!
Nick
Pyinstaller spec/build/dist location paths can be configured as part of pyinstaller command. Refer below example
pyinstaller --specpath /opt/bk/spec --distpath /opt/bk/dist --workpath /opt/bk/build testscript.py
As Jonas Gröger pointed out you can indeed do:
import PyInstaller.config
PyInstaller.config.CONF['workpath'] = "./my_build_directory"
# ... rest of spec file
However the documentation in the config module says that it works only on a limited number of variable:
This module holds run-time PyInstaller configuration.
Variable CONF is a dict() with all configuration options that are necessary
for the build phase. Build phase is done by passing .spec file to exec()
function. CONF variable is the only way how to pass arguments to exec() and
how to avoid using 'global' variables.
NOTE: Having 'global' variables does not play well with the test suite
because it does not provide isolated environments for tests. Some tests might
fail in this case.
NOTE: The 'CONF' dict() is cleaned after building phase to not interfere with
any other possible test.
To pass any arguments to build phase, just do:
from PyInstaller.config import CONF
CONF['my_var_name'] = my_value
And to use this variable in the build phase:
from PyInstaller.config import CONF
foo = CONF['my_var_name']
This is the list of known variables. (Please update it if necessary.)
cachedir
hasUPX
hiddenimports
noconfirm
pathex
ui_admin
ui_access
upx_dir
workpath
If you use "DISTPATH" in capital the trick will not work (Pyinstaller 3.3.1 and python 2.7.13) but if you set it lower case:
import PyInstaller.config
PyInstaller.config.CONF['distpath'] = "./my_app_directory"
# ... rest of spec file
That will work... :)

Categories

Resources