Tfdf custom loss

Hi, Is it possible to use a custom loss function for the DecisionForest models (any of them), like you would with any other Keras model? If so, is there an example?


Unfortunately, it is not currently possible to define a custom loss in the python code like you would do with classical Keras models.

Instead, custom losses are created by extending the C++ AbstractLoss class and implementing the corresponding methods (e.g. loss value, gradient). This is the list of the currently available losses.

Specific losses can be requested by posting an issue with the “Enhancement” tag.


1 Like