I'd like to use sdmTMB to build a model without running it, so that I can more easily diagnose issues during convergence. I think the easiest way to do that is add getsd=TRUE to sdmTMBcontrol(.), so that users can do getsd=FALSE, newton_loops = 0, nlminb_loops = 0 to get the build object without running it. getsd=FALSE would then skip sdreport and get_convergence_diagnostics calls in sdmTMB.
Does that sound like a useful PR to do on main branch, or any suggestions on alternative ways to do this?
PS: using it in this way also requires changing assert_that to validate_that in the auto-check about the passed value of nlminb_loops
I'd like to use sdmTMB to build a model without running it, so that I can more easily diagnose issues during convergence. I think the easiest way to do that is add
getsd=TRUE
tosdmTMBcontrol(.)
, so that users can dogetsd=FALSE, newton_loops = 0, nlminb_loops = 0
to get the build object without running it.getsd=FALSE
would then skipsdreport
andget_convergence_diagnostics
calls insdmTMB
.Does that sound like a useful PR to do on main branch, or any suggestions on alternative ways to do this?
PS: using it in this way also requires changing
assert_that
tovalidate_that
in the auto-check about the passed value ofnlminb_loops