[Scilab-users] ode, speed issue or tips to speed up ?
Stéphane Mottelet
stephane.mottelet at utc.fr
Fri Sep 3 13:46:27 CEST 2021
Hi,
Le 02/09/2021 à 20:11, Jakub Kopac a écrit :
> Hello,
>
> In the past I solved the same issue.
> As Stephane points out, try to play with ODEOPTIONS.
> In my case (disretized PDE), I get massive speed up by setting
> %ODEOPTIONS(6) = 0
> and selecting a stiff solver.
>
> If I understand it correctly, you are using function interp1 when
> evaluating ODE function (many calls to interp1).
> If it is this case, I propose to make a copy of Scilab interp1
> function and clean unecessary commands.
> I have also good experience with this aproach, but you are loosing
> flexibility and is a ad hoc function.
>
> Best regards
>
> J.K.
>
> PS: it is great to hear that ODE solvers will be updated to more
> recent versions.
The actual generic solvers of ode
(https://computing.llnl.gov/projects/odepack) are still competitive but
for large-scale stiff systems, odepack misses the versatileness of
Sundials concerning linear solvers (e.g. odepack does not handle sparse
Jacobians).
S.
>
>
> št 2. 9. 2021 o 18:32 Stéphane Mottelet <stephane.mottelet at utc.fr
> <mailto:stephane.mottelet at utc.fr>> napísal(a):
>
>
> Le 02/09/2021 à 18:03, CHEZE David 227480 a écrit :
>>
>> Dear all,
>>
>> In experimental work, I’m comparing dynamic model of a thermal
>> system with measurements. I’m using function ode, “fix” solver
>> (continuous time, lowest accuracy requirement is applying in my
>> case at this stage) and Jacobian ( to try to speed up the
>> calculation but not sensitive as for now) to get the solution of
>> the equation, a single temperature node, at 240 times (time
>> vector), default rtol=1.d-3 and atol=1.d-4. The system inputs are
>> available as vector of reals , size is 240 and inputs are passed
>> to the ode function as inputs functions by interpolating
>> (interp1) the inputs vectors. It is issuing correct temperature
>> solutions but is very slow, ~40s for 240 times, while the
>> variable inputs vector are actually not varying a lot and thus I
>> would expect very short calculation time.
>>
>> WIN10-64bits Intel i7 2,6GHz, Scilab 6.1.1
>>
>> Is there any tips to speed up the calculation , advanced example
>> documentation about ODEOPTIONS or ode_optional_output ? I guess
>> it’s the same underlying code that is used in Xcos isn’t it ?
>>
> No. Xcos uses the (more up-to-date) SUNDIALS solvers, they will be
> made available at the user level in the next months.
>>
>> Thank you for advising,
>>
>> David
>>
>>
>> _______________________________________________
>> users mailing list
>> users at lists.scilab.org <mailto:users at lists.scilab.org>
>> https://antispam.utc.fr/proxy/v3?i=cWdzUmRSM0ZiRHpoUDkxTSw3-90dQgKDkqmWWemZ6js&r=WXNQOUY5VXRSNUlTdlVTTgAgPzRHrDev93-c71JTxG3gYERdftx8enpEGCicW5Zy&f=R0pWUVNEaUFuMTBCTlptbqaYxU0NWnHz-Aov_fSSM1D4vfIz2L0jb50-moW0RphD&u=http%3A//lists.scilab.org/mailman/listinfo/users&k=8YLU <https://antispam.utc.fr/proxy/v3?i=cWdzUmRSM0ZiRHpoUDkxTSw3-90dQgKDkqmWWemZ6js&r=WXNQOUY5VXRSNUlTdlVTTgAgPzRHrDev93-c71JTxG3gYERdftx8enpEGCicW5Zy&f=R0pWUVNEaUFuMTBCTlptbqaYxU0NWnHz-Aov_fSSM1D4vfIz2L0jb50-moW0RphD&u=http%3A//lists.scilab.org/mailman/listinfo/users&k=8YLU>
>
> --
> Stéphane Mottelet
> Ingénieur de recherche
> EA 4297 Transformations Intégrées de la Matière Renouvelable
> Département Génie des Procédés Industriels
> Sorbonne Universités - Université de Technologie de Compiègne
> CS 60319, 60203 Compiègne cedex
> Tel : +33(0)344234688
> http://www.utc.fr/~mottelet <http://www.utc.fr/~mottelet>
>
> _______________________________________________
> users mailing list
> users at lists.scilab.org <mailto:users at lists.scilab.org>
> http://lists.scilab.org/mailman/listinfo/users
> <http://lists.scilab.org/mailman/listinfo/users>
>
>
> _______________________________________________
> users mailing list
> users at lists.scilab.org
> http://lists.scilab.org/mailman/listinfo/users
--
Stéphane Mottelet
Ingénieur de recherche
EA 4297 Transformations Intégrées de la Matière Renouvelable
Département Génie des Procédés Industriels
Sorbonne Universités - Université de Technologie de Compiègne
CS 60319, 60203 Compiègne cedex
Tel : +33(0)344234688
http://www.utc.fr/~mottelet
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.scilab.org/pipermail/users/attachments/20210903/1c32f519/attachment.htm>
More information about the users
mailing list