Orbit ISF


As the Orbit keeps improving with the new team members joining to the project, we recently published a post to share some thoughts how we scope the project and features we are planning to build.


Other key updates we have:

  1. We move our docs to readthedocs platform.
  2. A new kernel based model for complex seasonality and regression – KTR (Kernel Time-based Regression) is introduced. The model right now is divided into two stages KTRLite and KTRX where KTRLite uses penalized MLE to derive trend and seasonality and KTRX uses Stochastic Variational Inference(SVI) to derive externalities in a regression style. We also published a white paper in explaining the concepts and its application. We presented this in 41st International Symposium on Forecasting and upcoming AdKDD 2021
  3. For more transparency, I created a notebook online to showcase how we benchmark orbit model with M3 dataset. Open All Collab