Question: Change LGBMRegressor prediction aggregation

[hila-de-moon]

How can I change LGBMRegressor predictions so instead of calculating the average of the samples that fall in the leaf it will return median (or other metric).
I want it to be done only in predictions, I want the model to be train regularly with the averages, I cant use objective quantile.

Thank you

[jameslamb]

Thanks for using LightGBM.

I cant use objective quantile.

Why not? Can you please explain this statement a bit more?

[hila-de-moon]

yes,

I want to use monotone_constraints that currently doesn't work with objective quantile.
I have a priority on monotone_constraints so I thought i would train with objective regression and monotone_constraints and predict would be the leaf quantile.