nlp-uoregon/trankit

Single shared configuration for all Pipeline instances

abcdenis opened this issue · 1 comments

Hi,
What I did:

  • create two Pipeline instances for processing two languages (file global_config_demo.py, attached as text file because of github limitations global_config_demo.py.txt)
  • try to split sentence with one instance

What I get:
Error with the language from another instance (file console.log)

What I expected to get:
Split sentence without error.

Root of the problem
All instances share the same configuration so the only usage scenario is single threaded processing requests one by one:
https://github.com/nlp-uoregon/trankit/blob/master/trankit/pipeline.py#L166

How to fix
do not mix global and instance options and handle them in code separately

Working solution
Make an isolated copy of config for each Pipeline instance (global_config_demo_fixed.py.txt)

Is there any reason for global singleton Pipeline instance configuration?
Thank you.

Hi @abcdenis,
Thank you for reporting the issue.
We have updated the library to use separate configs for each Pipeline instance.
Please refer to commit 431ecab for more information.
Thanks