ad-freiburg / elevant

Entity linking evaluation and analysis tool
https://elevant.cs.uni-freiburg.de/
Apache License 2.0
19 stars 1 forks source link

Dependency conflict between radboud-el and xrenner #6

Closed agolo-alan-hogue closed 11 months ago

agolo-alan-hogue commented 12 months ago

Trying to install without docker by doing docker file commands as instructed.

There is a conflict apparently between these packages. I managed to get it to run by removing the version requirements, but I don't know if this will actually work or not.

python3 -m pip install -r requirements.txt

INFO: pip is looking at multiple versions of xrenner to determine which version is compatible with other requirements. This could take a while.
ERROR: Cannot install -r requirements.txt (line 11) and -r requirements.txt (line 12) because these package versions have conflicting dependencies.

The conflict is caused by:
    radboud-el 0.0.1 depends on flair>=0.11
    xrenner 2.2.0.0 depends on flair==0.6.1
flackbash commented 11 months ago

Thanks for reporting it! I will look into it, but in the meantime, if you are not using coreference resolution, the best solution is probably to remove xrenner from the requirements entirely.

agolo-alan-hogue commented 11 months ago

Thank you! Leaving out xrenner did work for now. However, I would like to have the option of using coref in the future, though it's not critical right now. Is there a chance that neuralcoref might be replaced with something like https://github.com/shon-otmazgin/fastcoref, or that it could be included as an option? Fastcoref seems to be pretty standard these days, and seems to be actively maintained.

flackbash commented 11 months ago

That looks really good at first glance, thanks for the link. I'll definitely look at adding support for it in ELEVANT.

flackbash commented 11 months ago

I removed support for xrenner and added support for fastcoref. You can test it e.g. in combination with ReFinED on our benchmark Wiki-Fair which contains coreference ground truth mentions:

python3 link_benchmark_entities.py refined.fastcoref -l refined -coref fastcoref -b wiki-fair

Note that you cannot use a coreference linker in isolation (without specifying an entity linker), since ELEVANT evaluates the linked entities, not the coreference clusters themselves.