search

Aarhus-Psychiatry-Research / psycop-model-training

Shared code for model training and evaluation.
Other
1 stars 0 forks source link

refactor: remove y_hat_int from eval_dataset #447

Closed MartinBernstorff closed 1 year ago

MartinBernstorff commented 1 year ago

Adds confusion, and seems to not add any value?

AFAIK, it's the predicted classes when the model chooses its own threshold. I'm not sure we'd ever want to use that? :-)

@bokajgd, @KennethEnevoldsen, @HLasse, This is potentially a very large change. any thoughts before I go ahead and make it happen?

HLasse commented 1 year ago

For our use cases where we're choosing thresholds to obtain a desired sens/spec it doesn't add any value. no. I'd be fine with removing if it's a source of confusion

KennethEnevoldsen commented 1 year ago

I agree with Lasse. I don't have the time to do this refactorization before going to the US. I however think that there is no reason to fix it unless you are already working with the code.

MartinBernstorff commented 1 year ago

Oh, totally - this occurred to me because it had caused confusion with Signe and Sara. I'll remove it.

bokajgd commented 1 year ago

It doesn't seem to be a major change at all. It is only used in one function create_performance_by_timedelta() where it can easily be replaced by y_hat_probs.round(). All other mentions should be fine to just remove. :)

MartinBernstorff commented 1 year ago

Excellent, I'll get right on this.