Taking a look at CoordinateAscent, it seems that we could benefit from abstracting out what we're actually evaluating, as there's a bunch of bookkeeping in just running qrels. I could imagine even one day wanting to train a kind of Processor rather than a model -- who knows.
Anyway, I abstracted it out locally, but I think ultimately that the names I chose are terrible, and want some feedback on what they should be and where they should go.
Taking a look at CoordinateAscent, it seems that we could benefit from abstracting out what we're actually evaluating, as there's a bunch of bookkeeping in just running qrels. I could imagine even one day wanting to train a kind of Processor rather than a model -- who knows.
Anyway, I abstracted it out locally, but I think ultimately that the names I chose are terrible, and want some feedback on what they should be and where they should go.
@mcartright @daltonj
CoordinateAscent.scala