Reusing adapted NUTS hyperparameters

Say I’ve run a bunch chains for a dataset X_0. It took a while, but they appear to have burned in and mixed based on available diagnostics. Now I have some additional datasets X_1, … X_K that I think are reasonably similar to X_0. Is there a way to initialize my NUTS chains for each additional datasets with the hyperparameters from the X_0 runs so that I could potentially have much faster burnin for my additional datasets?

1 Like

Current develop has an option to set inverse metric, stepsize and init.

See https://github.com/stan-dev/pystan/blob/develop/doc/hmc_euclidian_metric.rst