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
Right now, I believe result scores are the sum of the NGD edge scores for all edges in a result. We should investigate a weighted sum, where weights could be determined by the primary source or primary source type. Initially, we could try something very naive like "text-mined edges will have half the weight of non-text-mined edges". Evaluation of this naive scheme initially would have to be done by eye/smell test, but ideally we test via benchmarks in the future.
The text was updated successfully, but these errors were encountered:
It may make sense to make this a final pass in the score.js.
I'd be tempted to say the easiest way to handle this may be to add a scoreWeight property to the API List file that is then used in score.js. That said, by primary source type makes more sense than purely by API -- perhaps a map of different types in a new config?
I'll also say that we probably want to change it from sum to average or something for combining scores in creative mode to avoid scores going too high from just a lot of results being merged, though maybe we do want to keep the idea that more duplicate results being merged means a result is probably "better" because it's coming from multiple sources...
In NCATSTranslator/Feedback#21, "water" is returned as a treatment for cerebal palsy. The result comes from BTE (result 89 from https://arax.ncats.io/?r=83fa0ad2-e666-43db-a932-b02fceb335d6). All of the key edges are from text-mined sources (semmeddb and text-mining provider).
Right now, I believe result scores are the sum of the NGD edge scores for all edges in a result. We should investigate a weighted sum, where weights could be determined by the primary source or primary source type. Initially, we could try something very naive like "text-mined edges will have half the weight of non-text-mined edges". Evaluation of this naive scheme initially would have to be done by eye/smell test, but ideally we test via benchmarks in the future.
The text was updated successfully, but these errors were encountered: