Currently no way to easily evaluate models with temperature scaled output. I suggest adding such a temperature kwarg to the pmf method of ProbabilisticOutput, and giving the user the ability to specify a constant temperature choice in assess.py from the command line.
Currently no way to easily evaluate models with temperature scaled output. I suggest adding such a
temperature
kwarg to thepmf
method ofProbabilisticOutput
, and giving the user the ability to specify a constant temperature choice inassess.py
from the command line.Motivation: [Guo et al, 2017]