If we take the probability of a superclass as a split over the prior frequency of the classes in that superclass we can turn the predictions our model gives for superclasses into predictions for classes. Doing this, we'd want to know what score we're likely to get; and if we can combine these predictions with our raw predictions to improve our model's performance.
This should probably but a function called by check_test_score.py based on a flag when the script is run. Could also make a notebook looking at some of these questions while writing this.
If we take the probability of a superclass as a split over the prior frequency of the classes in that superclass we can turn the predictions our model gives for superclasses into predictions for classes. Doing this, we'd want to know what score we're likely to get; and if we can combine these predictions with our raw predictions to improve our model's performance.
This should probably but a function called by
check_test_score.pybased on a flag when the script is run. Could also make a notebook looking at some of these questions while writing this.