search

ufal / udpipe

UDPipe: Trainable pipeline for tokenizing, tagging, lemmatizing and parsing Universal Treebanks and other CoNLL-U files
Mozilla Public License 2.0
359 stars 75 forks source link

UDPipe-1.2.0 #146

Closed ALeczkowski closed 6 months ago

ALeczkowski commented 3 years ago

Hello, is there a way to change random seed (to randomize it instead of heaving it fixed) from command line when running training with UDPipe?

foxik commented 3 years ago

No, that is unfortunately currently not possible.

As a hack, you could try permuting the order of sentences on input, that might change the IDS of words and order of sentences during training, which might have a similar effect.

ALeczkowski commented 3 years ago

Thanks for your answer!

I was hoping for an easy and simple solution. Since there is not one, I will modify the source code myself and compile UDPipe. However, I have an issue with compilation. I trained the same model with precompiled UDPipe and with UDPipe compiled by me (of course I did not implement any modifications to the source code yet); however, I get a bit different results. Any idea what might be the issue? Can it be due to this error during compilation:

warning: dynamic exception specifications are deprecated in C++11

If you have any obvious solutions that I am missing I would appreciate if you share them with me.

PS. I did install all of the required programs (using Ubuntu).