search

uber / orbit

A Python package for Bayesian forecasting with object-oriented design and probabilistic models under the hood.
https://orbit-ml.readthedocs.io/en/stable/
Other
1.86k stars 132 forks source link

Forecasting values >= 0 #779

Open eliassz opened 2 years ago

eliassz commented 2 years ago

Sometimes it requires to forecast values that can be only more or equal zero (for instance, views). What model should I use or what parameters should I add to model to fulfill this condition?

edwinnglabs commented 2 years ago

Sometimes it requires to forecast values that can be only more or equal zero (for instance, views). What model should I use or what parameters should I add to model to fulfill this condition?

Hi @eliassz , in some tutorials, we did use a simple log1p (or log1p-log if you have regressors) transformation. For highly skewed response, you can also consider a box-cox transformation to reverse the normality. Hope this help.

eliassz commented 2 years ago

Thank you for the answer! But can please give me links to such tutorials?