Hi @jroon -
Yes, if you install CmdStan from Github you will get the latest development version which still includes the laplace features. Note that work is ongoing to improve them still, so expect results to change or code you write based on these versions to not necessarily work with the final versions.
There is a basic ability in the walnuts repository here, but we don’t currently have a lot of explanation on how to use the code (and, of course, we’re still testing the algorithm ourselves!).
If you do want to try it out, the brief steps required are:
- Build that repo using cmake with the
-DWALNUTS_USE_STAN=onsetting. - Build your model using
bridgestanBridgeStan – efficient, in-memory access to the methods of a Stan model — BridgeStan documentation - Run walnuts like
./examples/stan_cli YOUR_PROGRAM_model.so YOUR_DATA.json(see more options with./examples/stan_cli --help)
This will get easier and easier as WALNUTs continues development, ultimately with the goal being that it will be just as easy as using nuts is via cmdstanr/cmdstanpy