Initially I suspected it could be due to Knime being updated from 3.2.0 to 3.2.1 therefore I installed a fresh 3.2.1 version, but kept using the old (3.2.0) workspace.
I cannot find anything in Knime's log, Console or Error Log. When the Word Vector Learner runs, before showing any progress it simply closes my Knime instance. I tried with larger and smaller documents... I tried running it on another desktop too.
The basic environment that is failing is Windows 7 Ent SP1 x64. But it also failed on Win 2012 R2.
This seems to be the only node(from the hundreds I am using) that is showing this problem.
What else can I try?
I'm sorry that you ran into this problem. Could you maybe provide me with some more information? Did it work before using KNIME 3.2.0? Also did you run any other nodes from the Deeplearning4J Integration? Like training a network. If not you can find some example workflows you could try on our public Example Server.
We had a similar problem where the DL4J Feedorward Learner Node crashed the KNIME instance. In that case the problem resulted from conflicting dll's on Windows when Anacondas Python is also installed on the system. Do you maybe have Anacondas Python installed? If so you could try uninstalling it.
Thanks for your help,
Only the Word Vector Learner node execution closes my instance. The other DL4J samples and nodes work OK.
And, indeed, I do have Anaconda installed with Py 3.5 and 2.7 - good pointing.
Is there any workaround ?
the problem most likely results from the MKL libraries shipped by Anaconda. These are located in the Anaconda installation folder and automatically added to the system path during installation. As a workaround you could try to remove those entries from the system path and try it again. However, then Anaconda probably won't work anymore.
edit: If this is the cause of your problem, this will be fixed with the next KNIME release.
I can confirm that is the case. Moving Anaconda folders keeps Knime on track while running Word Vector Learner node.
thank you for confirming the issue. As mentioned, this problem will be fixed in the next KNIME release.