SparseNUTS: A new R package for MCMC in TMB and RTMB models

[Cole Monnahan - NOAA Federal]

Hi all,

A new R package called 'SparseNUTS' is now available for use. As with tmbstan it uses the NUTS algorithm as implemented in Stan to draw posterior samples from a TMB/RTMB model. The main difference is that it decorrelates the posterior using TMB's "joint precision matrix" which I call Q in the software. This can substantially improve sampling efficiency for many models, particularly those with high correlations and sparsity which is the case for many models in fisheries, ecology and beyond (see Table 1 of the MS linked below).

The R package is available here:

https://github.com/noaa-afsc/SparseNUTS

with articles on basic and advanced usage here:

https://noaa-afsc.github.io/SparseNUTS/

A preprint of the method is availble here:

https://arxiv.org/abs/2603.02437

This package works for both TMB and RTMB models. For RTMB users you will need to use the `globals` option to pass in data and functions used by RTMB.

Any feedback or suggestions via the issues page would be welcome. 

Thanks,

Cole