large scale values for NBI distribution about xgboostlss HOT 2 CLOSED

@hrkadkhodaei Thanks for your interest.

I suppose that for negative binomial, scale is equal to 1/n

Yes that is correct. From the density you can see that n = 1 / scale.

I am not sure where the problem comes from, but several options would be

• the data cannot be described well by a NBI
• the model is sensitive to the hyper-parameters, so better try and set most of them to the default and start varying the learning-rate first and check results again
• stabilize estimation of Gradients and Hessians via
  distribution = NBI
distribution.stabilize = "MAD" # or "L2"

Let me know if this helps.

from xgboostlss.

Closing this for now but feel free to re-open if the problem still persists.

from xgboostlss.