search

interpretml / interpret

Fit interpretable models. Explain blackbox machine learning.
https://interpret.ml/docs
MIT License
6.27k stars 729 forks source link

The loss function for EBM #580

Open JWKKWJ123 opened 5 days ago

JWKKWJ123 commented 5 days ago

Hi all, As far as I know, the loss function for EBM is cross-entropy for classification tasks and MSE for regression tasks. However, I’m not sure what the loss function is for each shape function (decision tree) used within the model. Also, is it possible to modify the loss function for each shape function?

paulbkoch commented 4 days ago

Hi @JWKKWJ123 -- The shape functions, or the trees that compose them, do not have individual loss functions. The loss function includes the target value and the additive sum of the shape functions with the intercept. In theory, you could have a model without any features, and you would still be able to calculate the loss from just the intercept and target.

Yes, the loss function for classification is cross-entropy loss and MSE for regression.

JWKKWJ123 commented 4 days ago

Hi @JWKKWJ123 -- The shape functions, or the trees that compose them, do not have individual loss functions. The loss function includes the target value and the additive sum of the shape functions with the intercept. In theory, you could have a model without any features, and you would still be able to calculate the loss from just the intercept and target.

Yes, the loss function for classification is cross-entropy loss and MSE for regression.

Hi Paul, Thank you very much! now i am clear.