Comment by schrodingerscow

This may be a naive question, but all the metrics seem to be for inference. Should we expect similar gains on training?