Additive Terms

[simon-hirsch]

No description provided.

[simon-hirsch]

Questions

Questions I'm encountering during development

• What about the global model selection
• Do we want to push forget to the EstimationMethod - I think we don't so that we can have different forgets for different terms that are estimated jointly.
• Do we want to save the first result of each inner iteration in terms regardless of performance to have a fall-back?

Remaining To-Do

• Make non-constant start values
• Generalize Terms a bit more s.t. we can have all terms jointly that use a certain estimation method
• Add a weight to coordinate descent for the regularization -> Allow for regularized estimation of centered, but unscaled variables with correct relative penalization