Hi! I'M trying to distill a NN model into a catboost

Question

Hi! I'M trying to distill a NN model into a catboost

model.
For this I need to provide class prediction logits as targets and use crossentropy as the loss.

I'm not sure how to proceed with this in catboost. The test code I tried out says that I HAVE to provide class labels only and cannot provide logits for the targets.

Is this something that can be done in catboost or do I need to implement a custom loss function that calculates crossentropy and use a regressor?

#english #programming

0

14.02.2022

2 ответов

22 просмотра

Arjoonn Автор вопроса

Stanislav Kirillov
Hello! You can just use loss_function='CrossEntrop...

That works normally but i have to provide classes as 0/1 labels of i do that. What i want to provide is target logits for the label instead of the classic 0/1. So 0.996 as target instead of 1

0

14.02.2022

Stanislav Kirillov · Accepted Answer

Stanislav Kirillov

Hello! You can just use loss_function='CrossEntropy' https://catboost.ai/en/docs/concepts/loss-functions-classification#CrossEntropy

0

14.02.2022

169 похожих чатов

Hi! I'M trying to distill a NN model into a catboost

2 ответов

Похожие вопросы