INLA and Pardiso

[Maria Antonietta Reatini]

Hello Dear all, 

I would like to know if there is any news on integrating INLA with Pardiso. Unfortunately, since my last license expired, I can no longer run a model I was working on (it used to take 4 hours to run, now it takes 12 days).

The model I am trying to fit is a coregionalization model and it has 14 parameters.

> dim(mesh$graph$vv) [1] 1026 1026

Total temporal nodes 13

I'm using the latest version of INLA on a server with 40 cores and over 300 GB of memory. Is it possible to find an alternative solution to Pardiso?

Thanks for any help/idea

Best regards,

Maria Antonietta

Fai crescere le giovani ricercatrici e i giovani ricercatori

con il 5 per mille alla Sapienza

Scrivi il codice fiscale dell'Università 80209930587
Cinque per mille

[Helpdesk (Haavard Rue)]

Hi

the issue is the authors of PARDISO has changed the license terms, so it
is easy to use, and there is no longer any free version for (linux-
)clusters and number of threads is restricte to 2. so we havn't made
the transition yet (still my lisence is good for another few months).
I'm not sure we will make the transition either.

We are working on replacing it with some other options (its looking good
so far, fingers crossed), hopefully this will happen during the fall.
this is completely open source, so there is no lisecing issues and will
also be supported for windows.

It is possible to compile the inla-program yourself, if you're linux
skilled, and/or know someone who is which can help you. then you need
to download and replace the current pardiso library wiht the new version
8 one (the one which require another lisence), and you have to get the
new lisence. I havn't tried this myself, but I would guess it should go
fine.

If you're interested, then let me know at he...@r-inla.org and I can
share the current build directory for Linux

Havard






On Mon, 2023-09-11 at 01:49 -0700, 'Maria Antonietta Reatini' via R-inla

> --
> 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 on the web, visit
> https://groups.google.com/d/msgid/r-inla-discussion-group/665605d6-51bd-445b-acb8-9825836366dcn%40googlegroups.com
> .

--
Håvard Rue
he...@r-inla.org

[Helpdesk (Haavard Rue)]

this got a bit unprecise, sorry about that. for academic use: it is
*less* easy to use due to machine restricted license, harder to get the
license, shorter academic license for most (2 months instead of 12
months), no (linux-)cluster options and restricted to factorisations
using 2 threads. Or you can buy a license which I do not know the price
for. non-academic users needs to buy the license, for a price I do not
know.

we will see if we can get a comparable open-source solution up and
running and integrated.

this, actually, is a bit of work...

again, for the skilled user, its easy to build the inla-program
yourself, and you can replace the pardiso library there with the new
version 8 one (which can be downloaded), if you want to give the new
version a try.

If you're interested, then let me know at he...@r-inla.org and I can

share the current build directory for Linux or Mac (Intel or M)

On Mon, 2023-09-11 at 23:04 +0300, Helpdesk (Haavard Rue) wrote:
> the issue is the authors of PARDISO has changed the license terms, so
> it
> is easy to use, and there is no longer any free version for (linux-
> )clusters and number of threads is restricte to 2.  so we havn't made

--
Håvard Rue
he...@r-inla.org

[Helpdesk (Haavard Rue)]

Testing version of today is compiled with PARDISO8, and one can get a
license at panua.ch

On Mon, 2023-09-11 at 01:49 -0700, 'Maria Antonietta Reatini' via R-inla
discussion group wrote: