`parse_yaml_predictions` doesn't correctly get predicate about ontogpt HOT 2 CLOSED

Hi @serenalotreck - that last PR doesn't actually add any support for n-ary relations, but you're right, it's still making a lot of assumptions about binary relations. Some objects with multiple attributes may not be relations, though, and the schemas intentionally don't require that level of modeling detail to be defined.

If I have a Drug class with the attributes brand_name and approval_status where both are strings, I could assume that all Drug objects should be translated to relations where brand_name and approval_status will be subject and object, respectively - but that may not be the intended model at all. Same for additional fields: if I have another field like updated_on then should that be a node property? An edge property? Probably not a predicate type but I don't really know that by the name alone.

We could just assume that everything is a triple, like in RDF, but that may defeat the purpose of LinkML, plus it's kind of messy.
So I'll propose a hybrid solution for now:

• If attributes are explicitly named subject, predicate, and object, they will be treated accordingly
• predicate may or may not be present - if not, use class name
• Otherwise, assume dicts of 1:1 key:value pairs are triples

This is a workaround. I will accept your changes in the linked PR then open another issue+PR to add the above solution, enabling other predicate parsing cases.

There's a more elegant solution here in which the annotations field can accept explicit directions for what should be subject, predicate, object, etc. - I'll open up another ticket for that.

from ontogpt.

Awesome, thanks so much!

from ontogpt.