nicholasjclark / mvgam

{mvgam} R 📦 to fit Dynamic Bayesian Generalized Additive Models for time series analysis and forecasting
https://nicholasjclark.github.io/mvgam/
Other
91 stars 11 forks source link

Allow intercept-free models #30

Closed nicholasjclark closed 6 months ago

nicholasjclark commented 9 months ago

The formula interface should allow for simple SS models where there is no intercept in either the observation or the latent state models. Something like:

   formula = y ~ -1,
   trend_formula = ~ -1, 
   trend_model = 'AR1',
   ...

Because we need an underlying (and valid) mgcv model for generating predictions, we'll need to include the intercept in the model and either fix it at zero (as a transformed parameter) or regularize it to zero

nicholasjclark commented 6 months ago

This is all done now in the latest versions