You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We can shoehorn certain classes of non-factorizable models (e.g. occupancy models) into brms by evaluating the likelihood inside an unlooped custom family. In order to take advantage of multiple threads per chain, one option is to write the stan code for the custom family using reduce_sum (this fails to use parallelization for computing the linear predictors, but still can use parallelization for evaluating the conditional likelihood). However, to use such a custom family, we need a way to force the model to compile with threading enabled. A hack using cmdstanr is to modify make/local as necessary at runtime, but if there's a possibility to implement a more streamlined solution that would be great!
The text was updated successfully, but these errors were encountered:
We can shoehorn certain classes of non-factorizable models (e.g. occupancy models) into
brms
by evaluating the likelihood inside an unlooped custom family. In order to take advantage of multiple threads per chain, one option is to write the stan code for the custom family usingreduce_sum
(this fails to use parallelization for computing the linear predictors, but still can use parallelization for evaluating the conditional likelihood). However, to use such a custom family, we need a way to force the model to compile with threading enabled. A hack usingcmdstanr
is to modifymake/local
as necessary at runtime, but if there's a possibility to implement a more streamlined solution that would be great!The text was updated successfully, but these errors were encountered: