bug(medcat): CU-869c0g9f7 Fix embedding linker disambiguation

[mart-r]

This PR will fix the issue with embedding linker that force it (in its default configuration with use_ner_link_candidates = True) not to any disambiguation.

Currently, whenever one runs the embedding linker it only links entities that had only one 1 link candidate. That's because of a condition in the _pre_inference method that ignored all other entities if use_ner_link_candidates was set.

This PR fixes that by:

• Adding a test to show the failure
• Fixing the condition to allow ambiguous names to be parsed by the linker

NOTE:

• The first commit / workflow is intended to fail (at tests) since it includes the test, but not the underlying fix
• The subsequent commmit / workflow should succeed as the fix is included

[tomolopolis]

Task linked: CU-869c0g9f7 Fix embedding linker not disambiguating

[alhendrickson]

Minor but this is the actual line so not super trivial. For my non python brain benefit - do you have to not len here?

Can you do just a not?

Suggested change

	elif self.cnf_l.use_ner_link_candidates and not len(entity.link_candidates):
	elif self.cnf_l.use_ner_link_candidates and not entity.link_candidates:

I think it's ultimately saying self.cnf_l.use_ner_link_candidates and len(entity.link_candidates) == 0, just the not len isnt a pattern I've seen before. I'm totally happy if it's super pythonic and standard, as a lot of things are.

[mart-r]

Yep, absolutely - having not entity.link_candidates would be better. I just started off with len(entity.link_candidates) == 0 and thought I'd simplify it. But you're right, relying on the truthy nature of a non-empty list is the best way here.

[alhendrickson]

Big change ! Thanks for clearing up