search

coli-saar / am-parser

Modular implementation of an AM dependency parser in AllenNLP.
Apache License 2.0
30 stars 10 forks source link

Evaluate TUPA on UCCA dev data #56

Closed alexanderkoller closed 4 years ago

alexanderkoller commented 4 years ago

Using pretrained model and mtool with UCCA input. This is needed as a baseline to compare our system against.

namednil commented 4 years ago

I already created a python2.7 environment in /proj/irtg.shadow/conda/envs/tupa/

alexanderkoller commented 4 years ago

TUPA actually requires Python >= 3.6. I pip-installed tupa into the "dozat" environment in /proj/irtg/.conda/envs. Trying to parse now.

NB here is how to make Conda look for environments in multiple directories: https://stackoverflow.com/questions/35280479/can-i-choose-where-my-conda-environment-is-stored

alexanderkoller commented 4 years ago

Tupa finished, but produced output (for the UCCA devset that @mariomgmn sent me) - see attached - that can't be read by mtool.

koller$ python main.py --read ucca --analyze ucca_inputs1k_0.xml 
Failed reading <root annotationID="0" passageID="inputs1k_0">, trying 3 more times...
Failed reading <root annotationID="0" passageID="inputs1k_0">, trying 2 more times...
Failed reading <root annotationID="0" passageID="inputs1k_0">, trying 1 more times...

I think at this point I'm ready to give up on this issue.

inputs1k_0.xml.gz