You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I can start playing with this one -- for this issue:
Do we just need an abstract base class?
Do we want to implement a ScoreCombiner (or some other name) that takes several Scorers and gives a single output? First version would be sum the scores, but options like max score, geometric mean of scores, etc. would be easy extensions.
We had talked about generalizing the idea of Scorers to Annotators. In this case, maybe we return a dict of annotations, with 'score' being the dict entry used for scores?
If we generalize to Annotators, it will be useful for each Annotator to have some way of communicating what annotations it provides. This could be with a class attribute (List[str]). This attribute could be set by an abstract Scorer class, making it transparent to users for that use case, although I'm not entirely sure that I like that idea.
1 yes
2 maybe leave that for now
3 I was thinking of tuple of (score, annotation) so people can easily discard the annotations (it's a bit more specialised)
4 For now this could just be done with docstrings. So just stating the resulting network will be annotated in a certain way.
The text was updated successfully, but these errors were encountered: