[OSError:[WinError 127]找不到指定的程序] pytorch-sparse安装引起

[LandaChina]

Describe the bug
安装成功dgl和autogl后，import验证时报错，pytorch-sparse安装后python.exe无法找到入口，卸载pytorch-sparse又会提示不存在模块，问题在torch_sparse循环。

To Reproduce

1. import autogl
2. 先出现torch_sparse找不到问题，安装wheel
3. 成功后，import autogl
4. 再次错误:OSError: [WinError 127] 找不到指定的程序。

import autogl
Traceback (most recent call last):
File "", line 1, in
File "C:\Users\All Users\anaconda3\envs\autogl\lib\site-packages\autogl_init_.py", line 1, in
from . import (
File "C:\Users\All Users\anaconda3\envs\autogl\lib\site-packages\autogl\datasets_init_.py", line 14, in
from .ogb import (
File "C:\Users\All Users\anaconda3\envs\autogl\lib\site-packages\autogl\datasets_ogb.py", line 8, in
from torch_sparse import SparseTensor
File "C:\Users\llzhang\AppData\Roaming\Python\Python38\site-packages\torch_sparse_init.py", line 18, in
torch.ops.load_library(spec.origin)
File "C:\Users\All Users\anaconda3\envs\autogl\lib\site-packages\torch_ops.py", line 255, in load_library
ctypes.CDLL(path)
File "C:\Users\All Users\anaconda3\envs\autogl\lib\ctypes_init_.py", line 373, in init
self._handle = _dlopen(self._name, mode)
OSError: [WinError 127] 找不到指定的程序。
并且跳出警告，如下图：

无法定位程序输入点于动态链接库\torch_sparse_version_cpu.pyd上

Expected behavior
成功import autogl，如何更好的安装torch_sparse，使之成功安装并且不报错。

Screenshots
If applicable, add screenshots to help explain your problem.

Environment (please complete the following information):

• OS: [Windows 10]
• python version: [3.8.17]
• autogl version: [0.4.0]
• pip list: [# packages in environment at C:\Users\All Users\anaconda3\envs\autogl:

Name Version Build Channel

appdirs 1.4.4 pyhd3eb1b0_0 defaults
astor 0.8.1 pypi_0 pypi
autogl 0.4.0 pypi_0 pypi
bayesian-optimization 1.4.3 pypi_0 pypi
blas 1.0 mkl defaults
brotlipy 0.7.0 py38h2bbff1b_1003 defaults
ca-certificates 2023.05.30 haa95532_0 defaults
certifi 2023.7.22 py38haa95532_0 defaults
cffi 1.15.1 py38h2bbff1b_3 defaults
charset-normalizer 2.0.4 pyhd3eb1b0_0 defaults
chocolate 0.0.2 pypi_0 pypi
cloudpickle 2.2.1 pypi_0 pypi
colorama 0.4.6 py38haa95532_0 defaults
contextlib2 21.6.0 pypi_0 pypi
contourpy 1.1.0 pypi_0 pypi
cryptography 41.0.2 py38hac1b9e3_0 defaults
cycler 0.11.0 pypi_0 pypi
cython 0.29.14 pypi_0 pypi
deeprobust 0.2.8 pypi_0 pypi
dgl 1.1.1 py38_0 dglteam
dill 0.3.7 pypi_0 pypi
dnspython 2.4.2 pypi_0 pypi
filelock 3.12.2 pypi_0 pypi
fonttools 4.42.0 pypi_0 pypi
gensim 3.8.3 pypi_0 pypi
hyperopt 0.1.2 pypi_0 pypi
icc_rt 2022.1.0 h6049295_2 defaults
idna 3.4 py38haa95532_0 defaults
imageio 2.31.1 pypi_0 pypi
importlib-metadata 6.8.0 pypi_0 pypi
importlib-resources 6.0.1 pypi_0 pypi
intel-openmp 2023.1.0 h59b6b97_46319 defaults
jinja2 3.1.2 py38haa95532_0 defaults
joblib 1.3.2 pypi_0 pypi
json-tricks 3.17.2 pypi_0 pypi
kiwisolver 1.4.4 pypi_0 pypi
lazy-loader 0.3 pypi_0 pypi
libffi 3.4.4 hd77b12b_0 defaults
libuv 1.44.2 h2bbff1b_0 defaults
lightgbm 4.0.0 pypi_0 pypi
littleutils 0.2.2 pypi_0 pypi
llvmlite 0.40.1 pypi_0 pypi
markupsafe 2.1.3 pypi_0 pypi
matplotlib 3.7.2 pypi_0 pypi
mkl 2023.1.0 h6b88ed4_46357 defaults
mkl-service 2.4.0 py38h2bbff1b_1 defaults
mkl_fft 1.3.6 py38hf11a4ad_1 defaults
mkl_random 1.2.2 py38hf11a4ad_1 defaults
mpmath 1.3.0 py38haa95532_0 defaults
netlsd 1.0.2 pypi_0 pypi
networkx 3.1 py38haa95532_0 defaults
ninja 1.10.2 haa95532_5 defaults
ninja-base 1.10.2 h6d14046_5 defaults
nni 2.8 pypi_0 pypi
numba 0.57.1 pypi_0 pypi
numpy 1.24.3 py38h79a8e48_1 defaults
numpy-base 1.24.3 py38h8a87ada_1 defaults
ogb 1.3.6 pypi_0 pypi
openssl 3.0.10 h2bbff1b_0 defaults
outdated 0.2.2 pypi_0 pypi
packaging 23.0 py38haa95532_0 defaults
pandas 2.0.3 pypi_0 pypi
pillow 10.0.0 pypi_0 pypi
pip 23.2.1 py38haa95532_0 defaults
pooch 1.4.0 pyhd3eb1b0_0 defaults
prettytable 3.8.0 pypi_0 pypi
protobuf 4.24.0 pypi_0 pypi
psutil 5.9.0 py38h2bbff1b_0 defaults
pycparser 2.21 pyhd3eb1b0_0 defaults
pymongo 4.4.1 pypi_0 pypi
pyopenssl 23.2.0 py38haa95532_0 defaults
pyparsing 3.0.9 pypi_0 pypi
pysocks 1.7.1 py38haa95532_0 defaults
python 3.8.17 h1aa4202_0 defaults
python-dateutil 2.8.2 pypi_0 pypi
pythonwebhdfs 0.2.3 pypi_0 pypi
pytorch 2.0.1 cpu_py38hb0bdfb8_0 defaults
pytz 2023.3 pypi_0 pypi
pywavelets 1.4.1 pypi_0 pypi
pyyaml 6.0.1 pypi_0 pypi
requests 2.31.0 py38haa95532_0 defaults
responses 0.23.3 pypi_0 pypi
schema 0.7.5 pypi_0 pypi
scikit-image 0.21.0 pypi_0 pypi
scikit-learn 1.3.0 pypi_0 pypi
scipy 1.10.1 py38hdcfc7df_1 defaults
setuptools 68.0.0 py38haa95532_0 defaults
simplejson 3.19.1 pypi_0 pypi
smart-open 6.3.0 pypi_0 pypi
sqlite 3.41.2 h2bbff1b_0 defaults
sympy 1.12 pypi_0 pypi
tabulate 0.9.0 pypi_0 pypi
tbb 2021.8.0 h59b6b97_0 defaults
tensorboardx 2.6.2 pypi_0 pypi
texttable 1.6.7 pypi_0 pypi
threadpoolctl 3.2.0 pypi_0 pypi
tifffile 2023.7.10 pypi_0 pypi
torch-sparse 0.6.17+pt20cpu pypi_0 pypi
torchvision 0.15.2 pypi_0 pypi
tqdm 4.65.0 py38hd4e2768_0 defaults
typeguard 4.1.0 pypi_0 pypi
types-pyyaml 6.0.12.11 pypi_0 pypi
typing-extensions 4.7.1 py38haa95532_0 defaults
typing_extensions 4.7.1 py38haa95532_0 defaults
tzdata 2023.3 pypi_0 pypi
urllib3 1.26.16 py38haa95532_0 defaults
vc 14.2 h21ff451_1 defaults
vs2015_runtime 14.27.29016 h5e58377_2 defaults
wcwidth 0.2.6 pypi_0 pypi
websockets 11.0.3 pypi_0 pypi
wheel 0.38.4 py38haa95532_0 defaults
win_inet_pton 1.1.0 py38haa95532_0 defaults
zipp 3.16.2 pypi_0 pypi
]

Additional Info (Optional)

提示的定位不到的动态链接库_version_cpu，本地可以找到几个，替换报错位置的文档，也没有解决。

[LandaChina]

Additional Info:
Next, conda uninstall pytorch-scatter, successfully, the error 'OSError:[WinError 127]' happened again. Maybe the wrong is from the torch-sparse wheel installed. The torch_sparse is found from https://pytorch-geometric.com/whl/torch-2.0.1%2Bcpu.html.

[general502570]

谢谢您对AutoGL的关注。pytorch-sparse 的依赖应该是ogb数据集所需要的。您可以试一试能不能直接import ogb库。如果不能的话，请在ogb库的讨论区寻找相关解答。如果可以的话，请再回复给我们。

[LandaChina]

谢谢您对AutoGL的关注。pytorch-sparse 的依赖应该是ogb数据集所需要的。您可以试一试能不能直接import ogb库。如果不能的话，请在ogb库的讨论区寻找相关解答。如果可以的话，请再回复给我们。

成功import ogb！

(autogl) C:\Users\llzhang>python
Python 3.8.17 (default, Jul 5 2023, 20:44:21) [MSC v.1916 64 bit (AMD64)] :: Anaconda, Inc. on win32
Type "help", "copyright", "credits" or "license" for more information.

import ogb
import autogl
Traceback (most recent call last):
File "", line 1, in
File "C:\Users\All Users\anaconda3\envs\autogl\lib\site-packages\autogl_init_.py", line 1, in
from . import (
File "C:\Users\All Users\anaconda3\envs\autogl\lib\site-packages\autogl\datasets_init_.py", line 14, in
from .ogb import (
File "C:\Users\All Users\anaconda3\envs\autogl\lib\site-packages\autogl\datasets_ogb.py", line 8, in
from torch_sparse import SparseTensor
File "C:\Users\llzhang\AppData\Roaming\Python\Python38\site-packages\torch_sparse_init.py", line 18, in
torch.ops.load_library(spec.origin)
File "C:\Users\All Users\anaconda3\envs\autogl\lib\site-packages\torch_ops.py", line 255, in load_library
ctypes.CDLL(path)
File "C:\Users\All Users\anaconda3\envs\autogl\lib\ctypes_init_.py", line 373, in init
self._handle = _dlopen(self._name, mode)
OSError: [WinError 127] 找不到指定的程序。

但是import autogl还是存在以上问题。

[ZW-ZHANG]

感谢您的回复，该问题看起来是由于torch_sparse库引起的。您可以先试一下直接import torch_sparse能否成功，如过不能的话，请查看torch_sparse寻找相关原因（比如是否为python/torch/gpu版本与torch_sparse库不匹配等）

[LandaChina]

pytorch_sparse 是pytorch的1.6后版本使用，torch-sparse是1.6之前的版本，同时pytorch_sparse没有Windows版本，conda install找不到，从而得出的结论是autogl对操作系统有要求。

[ZW-ZHANG]

您好，可能之前没有说清，我说的是https://github.com/rusty1s/pytorch_sparse 中的torch_sparse，它是pytorch geometric库的一个依赖（虽然您用的是DGL backend，但好像因为ogb或者别的库的依赖，也需要安装torch_sparse）。我个人尝试按照上面网页的说明安装torch_sparse后，是可以在windows下成功运行autogl