Custom Loss and interecation between target

[leorene97490]

Hi!

Thank you so much for the library; it’s really interesting for addressing multi-output issues. I’m trying to create two types of models. I have 50 targets, and the input X is the same for both models.

• Model 1: One boosting model per target. However, the loss of each model needs to take into account the loss of the other models at each step. Each model for each target should be optimized simultaneously.

• Model 2: A single boosting model that predicts all targets simultaneously. The loss function of this model should incorporate interactions between all targets, and the model should output a vector of predictions.

Is it possible to implement these kinds of models using PyBoost?