Deeplearning4J crashes KNIME when running the basic example workflow

When executing the examplw worflow

01_Basic_Concepts_Of_Deeplearning4J_Integration

downloaded from the example server, KNIME hard crashes (KNIME windows gets closed, no error, no log) when running the learner node. The "Create Network Architecutre" node completes correctly. I did not change anything in the workflow. The crash happens about 2-3 sec. after learner node execution starts and I can get a glimpse of 0% completion on this node.

KNIME 3.2 64-bit

64-bit windows

Running on CPU (Intel)

eg. all is default configuration as per initial download of extension and the workflow.

 

This problem is known and we are currently working on it.

Are you running on Windows 10? Maybe in a virtual machine?

Are there no related lines in the log file?

Edit: Could you tell me what kind of CPU you are using?

Thank you for your help.

Hi Dave,

it's windows 7 enterprise on a physical machine.

CPU is Intel Ccore i5-4570S.

 

Hi beginner_

thanks for the info. We are trying to narrow down the error. May I ask you another question? Do you maybe use an application or are you developing an application which uses any of these libraries?

  • Intel MKL
  • OpenBLAS
  • ATLAS

 

 

Hi Dave,

I've heard MKl and OpenBLAS and I have a standard python and anaconda-python installation with numpy on that machine and some windows builds of numpy do use Intel MkL..I'm pretty sure it does use it but don't know how to check.

Hi beginner_

that info helped a lot. Thanks. I managed to determine the Problem. The crash results from conflicting DLLs on windows when MKL is on the path, because the Integration ships with OpenBlas by default.. The DLLs get added when anaconda-python is installed.

We are currently trying to fix the problem.

One workaround would be to remove the anaconda entries from your system path variable. However, this will BREAK! anaconda-python.

Thanks again.

Hi, 

will there be a fix for this problem soon ? Using KNIME and anaconda/python ....

Hey folks, Adam from skymind here (I run the company behind the underlying tech.)

 

We ran in to this in the user community already, please scroll down to the bottom of this read me and you can see how we got mkl to work:
https://github.com/deeplearning4j/libnd4j

 

If you have any problems relating to the underlying technology I highly suggest coming in to our open source channel:
https://gitter.im/deeplearning4j/deeplearning4j

 

I won't comment on the knime bits but if there's anything we can do to help please let us know.

I followed the steps provided in the readme (eg. adding mkl_rt.dll to path) and KNIME still crashes. So this sadly does not solve the issue on Windows (Note: libiomp5md.dll is present in same folder as mkl_rt.dll)

Hi all,

the problem will be fixed in the new KNIME version 3.3 which will be released in about a month.

I have mkl_rt.dll in C:\Anaconda2\Library\Bin, not in "/path/to/intel64/lib/".

This workaround is successful, but I hope you will solve the problem as soon as possible