Can we manually create the stopping criteria of INLA?

[Alokesh Manna]

Hi INLA group,

Can we manually create the stopping criteria like "it will run 10000 steps"? In MCMC we can do that. Is there any default value for INLA?

Alokesh

[Helpdesk (Haavard Rue)]

not from R, no.

> --
> You received this message because you are subscribed to the Google Groups "R-
> inla discussion group" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to r-inla-discussion...@googlegroups.com.
> To view this discussion, visit
> https://groups.google.com/d/msgid/r-inla-discussion-group/b928f0ae-e669-43eb-a1d2-9ab1c1288a4en%40googlegroups.com
> .

--
Håvard Rue
he...@r-inla.org

[Alokesh Manna]

Thanks for the feedback!

Any way without R?

In future it maybe great to see some traceplot etc.

[Finn Lindgren]

While seeing the optimiser trace might be useful for internal debugging, I don’t think it would be of general use (optimisers don’t necessarily take an humanly logical path towards the optimum) and the amount of data storage would be huge, as would the added time of actually writing all those values to disk; storing hundreds of copies of the model would be really bad. That said, nonlinear inlabru models do store the trace (usually the number of iterations is small, so the memory requirement is less than it would be to store the full inla optimisation) of its iterated inla method for nonlinear predictor models; bru_convergence_plot() displays a summary (in light of my comments above I’ll likely at least make it an option to only store the convergence summaries).

Finn

On 13 Dec 2024, at 13:28, Alokesh Manna <alokesh...@gmail.com> wrote:



[Helpdesk (Haavard Rue)]

If you want to see the optimization path, you can do

results <- inla(...)

plot(result, plot.opt.trace=TRUE)

which will give two plots; as attached

[Alokesh Manna]

Very helpful!

[Finn Lindgren]

Just in case someone thinks my comment doesn't match Haavard's, I was
thinking of the full latent variable trace, not just the theta; but
good to know that the theta trace is there!
Finn

--
Finn Lindgren
email: finn.l...@gmail.com

[Alokesh Manna]

Thank you!