OSError: exception: access violation writing

[leftluoyi]

Windows 10 + Python 3.6.3 x64 + CUDA 8.0

Have successfully installed LightGBM and can run the binary_classification example from command prompt.

If set the classifier as XGBClassifier, everything is okay. When change to LGBMClassifier, the following exceptions is thrown.

 File "D:\Anaconda3\lib\site-packages\lightgbm\sklearn.py", line 695, in fit
    callbacks=callbacks)
  File "D:\Anaconda3\lib\site-packages\lightgbm\sklearn.py", line 474, in fit
    callbacks=callbacks)
  File "D:\Anaconda3\lib\site-packages\lightgbm\engine.py", line 183, in train
    booster = Booster(params=params, train_set=train_set)
  File "D:\Anaconda3\lib\site-packages\lightgbm\basic.py", line 1307, in __init__
    train_set.construct().handle,
  File "D:\Anaconda3\lib\site-packages\lightgbm\basic.py", line 860, in construct
    categorical_feature=self.categorical_feature, params=self.params)
  File "D:\Anaconda3\lib\site-packages\lightgbm\basic.py", line 714, in _lazy_init
    self.__init_from_np2d(data, params_str, ref_dataset)
  File "D:\Anaconda3\lib\site-packages\lightgbm\basic.py", line 776, in __init_from_np2d
    ctypes.byref(self.handle)))
OSError: exception: access violation writing 0xFFFFFFFF98090000

Any idea?

======================== UPDATE ==========================

Now I can run this example with CPU version and GPU version with lightgbm.exe, but the GPU version with python throws this exception.

[guolinke]

are you on the latest version ?

[guolinke]

BTW, can you provide a re-produce code with randomly generated data ?

[leftluoyi]

@guolinke I think I am, just cloned LightGBM half an hour ago.

[leftluoyi]

I tried with this example. The problem still exists.

[guolinke]

just run it without error:

[guolinke]

can you try to uninstall and re-install ? you can install version 2.1.2 from pypi directly.

[leftluoyi]

@guolinke I will try to reinstall. BTW, here is what I got.

[guolinke]

BTW, we have the tests for these examples in Windows: https://github.com/Microsoft/LightGBM/blob/master/.appveyor.yml#L51-L58
So these examples should be run without errors, since the tests are passed

[leftluoyi]

@guolinke Thanks for the advice. I installed the cpu version via pip install lightgbm and things works fine. But I forgot to mention that I am trying with gpu version. So if I install with pip install lightgbm --install-option=--gpu --install-option=--mingw, the access violation exception happens again.

[guolinke]

ping @huanzhang12 @StrikerRUS

[leftluoyi]

The problem is solved by update relevant packages by pip install setuptools numpy scipy scikit-learn -U. @guolinke thanks for your advice. Please close this issue.