The issue of regression

[chenmingyajun]

No description provided.

[chenmingyajun]

Hello! I would like to ask if it is possible to customize the distribution function? Is it possible to approximate the real distribution without knowing the distribution?

[alejandroschuler]

Hey, I'm not sure I totally understand the question. You can certainly customize the conditional distribution you choose to use, either by using one of the options provided in ngboost or by creating your own as described in the user guide: https://stanfordmlgroup.github.io/ngboost/5-dev.html

However, ngboost is necessarily limited in that you must choose some parametric model for your conditional distribution. In theory you can choose a very complicated parametric model, but in practice it should have few parameters or else the algorithm will take a long time to run.