A list for the developers of CellML tools
Text archives Help
- From: r.britten at auckland.ac.nz (Randall Britten)
- Subject: [cellml-dev] r3050 - CellML_DOM_API/trunk/CIS/sources
- Date: Thu, 5 Feb 2009 12:27:46 +1300
Hi
Not sure if the random seed will address the random guesses issue from one
run to the next, since the code generated by CCGS is different each time, so
even if the sequence of random numbers is the same, they may be allocated to
different variables.
Regards,
Randall
> -----Original Message-----
> From: automated-notifications-bounces at cellml.org [mailto:automated-
> notifications-bounces at cellml.org] On Behalf Of CellML Automated
Notifications
> Sent: Thursday, 5 February 2009 11:25 a.m.
> To: automated-notifications at cellml.org
> Subject: r3050 - CellML_DOM_API/trunk/CIS/sources
>
> Author: amil082
> Date: 2009-02-05 11:24:58 +1300 (Thu, 05 Feb 2009)
> New Revision: 3050
>
> Modified:
> CellML_DOM_API/trunk/CIS/sources/CISSolve.cxx
> Log:
> Fix tracker item 1563 by increasing the number of random starts if we
don't
> get an acceptable solution to start with. Also, this uses a fixed random
seed,
> as discussed in tracker item 1497, and multiplies the tolerance by system
> size to aid in solving larger systems.
>
> Modified: CellML_DOM_API/trunk/CIS/sources/CISSolve.cxx
> ===================================================================
> --- CellML_DOM_API/trunk/CIS/sources/CISSolve.cxx 2009-02-04 21:21:17
UTC
> (rev 3049)
> +++ CellML_DOM_API/trunk/CIS/sources/CISSolve.cxx 2009-02-04 22:24:58
UTC
> (rev 3050)
> @@ -735,9 +735,11 @@
> return ((inum % iden) == 0) ? 1.0 : 0.0;
> }
>
> -#define NR_RANDOM_STARTS 100
> +#define NR_RANDOM_STARTS_MIN 100
> +#define NR_RANDOM_STARTS_MAX 10000000
> #define NR_MAX_STEPS 1000
> #define NR_MAX_STEPS_INITIAL 10
> +#define RANDOM_SEED 0x7ce4176c
>
> static double
> random_double_logUniform()
> @@ -802,9 +804,12 @@
> {
> double info[9], best = INFINITY;
> uint32_t i = 0, k;
> + double tolerance = 1E-6 * size;
>
> memcpy(bp, params, sizeof(double) * size);
>
> + srand(RANDOM_SEED);
> +
> do
> {
> /* XXX casting away constness is bad, but it seems in this case
> dlevmar_dif is
> @@ -821,9 +826,10 @@
> for (k = 0; k < size; k++)
> bp[k] = random_double_logUniform();
> }
> - while (i++ < 100);
> + while ((i++ < NR_RANDOM_STARTS_MIN) ||
> + (best > tolerance && i <= NR_RANDOM_STARTS_MAX));
>
> /* XXX we shouldn't hard-code the tolerance... */
> - if (best > 1E-6)
> + if (best > tolerance)
> *pret = 1;
> }
>
> _______________________________________________
> automated-notifications mailing list
> automated-notifications at cellml.org
> http://www.cellml.org/mailman/listinfo/automated-notifications
- [cellml-dev] r3050 - CellML_DOM_API/trunk/CIS/sources, Alan Garny, 02/05/2009
- <Possible follow-up(s)>
- [cellml-dev] r3050 - CellML_DOM_API/trunk/CIS/sources, Randall Britten, 02/05/2009
Archive powered by MHonArc 2.6.18.
