The "Make predictions" algorithm is not making use of labeled features

[michi-zuri]

When running generate_predictions.py only the following tables are read, some of them are written to:

• citations
• labels decisions by reviewer
• projects (reviews)
• prediction_status
• predictions
• priorities

Notably missing from this list is the table labeledfeatures, which contains the highlighted words from the abstracts with their thumbs up/down rating. In the article http://www.byronwallace.com/static/articles/wallace_ihi_2011_preprint.pdf, also available here: https://dl.acm.org/doi/10.1145/2110363.2110464 it is mentioned that Abstrackr will use dual supervision, i.e. using these weighted labels to make even better predictions.

This raises two questions for me:
Has an implementation of a dual supervision algorithm found it's way onto the public web application yet?
Is this repo here on github up to date with the latest algorithm used by that web application?

If not, is this because the benefit of dual supervision is non-existant or has it just not been implemented yet in Abstrackr?
If yes, would you be so kind to share the latest code base with the world? See also my other issue regarding the license of this repo #110

All the best
Michael