v1.1.3

mvgam 1.1.3

New functionalities

• Allow intercepts to be included in process models when trend_formula is supplied. This breaks the assumption that the process has to be zero-centred, adding more modelling flexibility but also potentially inducing nonidentifiabilities with respect to any observation model intercepts. Thoughtful priors are a must for these models
• Added standata.mvgam_prefit, stancode.mvgam and stancode.mvgam_prefit methods for better alignment with 'brms' workflows
• Added 'gratia' to Enhancements to allow popular methods such as draw() to be used for 'mvgam' models if 'gratia' is already installed
• Added an ensemble.mvgam_forecast method to generate evenly weighted combinations of probabilistic forecast distributions
• Added an irf.mvgam method to compute Generalized and Orthogonalized Impulse Response Functions (IRFs) from models fit with Vector Autoregressive dynamics

Deprecations

• The drift argument has been deprecated. It is now recommended for users to include parametric fixed effects of "time" in their respective GAM formulae to capture any expected drift effects

Bug fixes

• Added a new check to ensure that exception messages are only suppressed by the silent argument if the user's version of 'cmdstanr' is adequate
• Updated dependency for 'brms' to version >= '2.21.0' so that read_csv_as_stanfit can be imported, which should future-proof the conversion of 'cmdstanr' models to stanfit objects (#70)

v1.1.2

This version brings several new features and efficiency improvements

• Added options for silencing some of the 'Stan' compiler and modeling messages using the silent argument in mvgam()
• Moved a number of packages from 'Depends' to 'Imports' for simpler package loading and fewer potential masking conflicts
• Improved efficiency of the model initialisation by tweaking parameters of the underlying 'mgcv' gam object's convergence criteria, resulting in much faster model setups
• Added an option to use trend_model = 'None' in State-Space models, increasing flexibility by ensuring the process error evolves as white noise (#51)
• Added an option to use the non-centred parameterisation for some autoregressive trend models,
  which speeds up mixing most of the time
• Updated support for multithreading so that all observation families (apart from nmix()) can now be modeled with multiple threads
• Changed default priors on autoregressive coefficients (AR1, AR2, AR3) to enforce
  stationarity, which is a much more sensible prior in the majority of contexts
• Fixed a small bug that prevented conditional_effects.mvgam() from handling effects with three-way interactions

v1.10

This release brings functionality for the Binomial, Beta-Binomial and Bernoulli distributions

v1.09

This release brings a new family, nmix(), which handles Poisson Binomial N-mixture models for count data with imperfect detection. It also uses vectorized operations for vastly improved performance of Dunn Smyth residual calculations

v1.08

This release brings piecewise linear and logistic trends with automatic changepoint selection, similar to what is available in Facebook's popular prophet package. It also includes a few bug fixes to make gp() effects more stable and versatile

v1.07

This release coincides with new additions for moving average terms in autoregressive process models, as well as the possibility to estimate correlated process errors for RW and AR(1-3) models when working with multivariate time series

v1.06

This version brings support for marginaleffects, state-space models using trend_formula, stationary VAR trends and gp() terms

v1.04

Last release that only allows non-negative discrete outcomes

v1.03

Final manuscript release

v1.0.1

This release adds support for cmdstanr to work as the backend when fitting models in Stan, which improves sampling efficiency and drastically speeds compilation compared to rstan (on Windows machines)