When I try "Testing your CNTK install with Python", my python crash

[xiangyubo]

I installed Anaconda successfully, and I installed cntk successfully. But when I follow the tutorial, my python crash... here is my environment: Python 3.5.2 :: Anaconda 4.1.1 (64-bit) cntk.version 2.3.1 when I run Tutorials/NumpyInterop example from pycharm, I get this:

Selected CPU as the process wide default device.

Build info:

    Built time: Dec  5 2017 18:19:12
    Last modified date: Mon Nov 13 23:00:38 2017
    Build type: Release
    Build target: CPU-only
    With 1bit-SGD: no
    With ASGD: yes
    Math lib: mkl
    Build Branch: HEAD
    Build SHA1: ce72a557b02b30a52c965a237b124688035f720f
    MPI distribution: Microsoft MPI
    MPI version: 7.0.12437.6

---

Process finished with exit code -1066598274 (0xC06D007E)

[manikjindal]

@xiangyubo How did you try to run the example? Is it from terminal in PyCharm? Are you able to execute the same example from Anaconda prompt using the same environment?

[xiangyubo]

It's from terminal in PyCharm. I also execute the same example in cmd, and python crash.

[manikjindal]

Could you please check the same with CNTK 2.4? Please re-open the issue if it persist with CNTK 2.4.

[nikitbiraj]

I have the same issue. python crashes when trainer module is called. Any idea how to resolve the issue?

[thiagocrepaldi]

CNTK 2.5 packages don't have MKL, CUDA and OpenCV embedded on its Python packages. It is required to pre-install them before using. Refer to https://docs.microsoft.com/en-us/cognitive-toolkit/Setup-Windows-Python?tabs=cntkpy25#environment-variables-and-required-packages or https://docs.microsoft.com/en-us/cognitive-toolkit/Setup-Linux-Python?tabs=cntkpy25#environment-variables-and-required-packages for a step by step setup guide

[nikitbiraj]

Thanks. It's working now. issue with MKL packages.

[xiangyubo]

OK, I will try ! Thank you!

[obu80]

i have the same problem with both cpu and gpu versions of cntk Process finished with exit code -1066598274 (0xC06D007E) I also downloaded MKL and added its lib folder to the system path. Still problem is not resolved. Is there a solution for this problem? Thank you

[thiagocrepaldi]

Could you try CNTK 2.6?

[obu80]

I am trying with the latest version of cntk. I install it in miniconda with pip install cntk-gpu or pip install cntk commands. Both give the same error. Python version is 3.6. Thank you for your help.

[thiagocrepaldi]

Got it. Could you also try using Python 3.5 instead? Some users reported that python 3.6 environments (from some miniconda/anaconda releases) would not recognize MKL library path (and possibly other external dependencies).

[obu80]

OK i will try that, thank you

[obu80]

I tried with python 3.5.6 and cntk 2.3.1, the same problem persists. I dont know why?

[obu80]

I could not solve the problem with various python and cntk versions. I also installed versions of python/cntk which successfully works in another machine, but the same problem still occurs in the target machine.

It gives the following errror "Process finished with exit code -1066598274 (0xC06D007E)" exactly in trainer = cntk_py.trainer_impl(model, loss_function, eval_function, parameter_learners, progress_writers) line.

In the above function call, only "model" equals to "None"

[mattans]

This still happens in latest CNTK (2.7) and Python 3.6 Will try to install the components in @thiagocrepaldi 's links

Update: after adding the mklml files, still crashed on PyCharm terminal. In Anaconda Prompt, after conda install -c intel mkl-dnn , it worked.