code-kern-ai / refinery-python-sdk Goto Github PK

Bei fehlerhafter Projekt ID kriege ich positives LogIn-Feedback

Often times one wants to express Labeling Functions that work with Lemmatization, Stemming, OCR Mistakes or Synonyms.

E.g. "If the headline contains some variation of the word kill, it is crime related". One does not want to manually express this sort of Labeling Function with simple keyword search, since that would mean the keywords would be "kill, killed, kills, ..." - you get the point.

This is just one example, there are multiple Use Cases, some of which I listed at the top of this comment.

Current outline exists commented in the code; needs to implement the boto3 client with the credentials we get from the rest-api

I think creating LFs should not be restricted to the labels that are in the system at creation time because one might want to already create some LFs for future labels. But I think there should be at least a warning to prevent this from happening accidentially.

https://github.com/basetenlabs/truss is an amazing OS library to easily service trained models. This can be integrated very well with our SDK, such that engineers can quickly build and serve models directly from the refinery

Currently the requirements are really hard with the "=="-operator, we should think about making them more tolerable of other versions.