trshsb51 comments on QAPR 3: interpretability-guided training of neural nets