You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@commons.apache.org by Cyril de Catheu <cy...@startree.ai.INVALID> on 2023/03/04 11:29:25 UTC
[math] discuss faster getEntry for RealMatrix implementations
Hello,
I just opened a ticket describing an opportunity to optimize the getEntry
method of RealMatrix implementations.
For our specific workload in BOBYQA this seems to be a ~10% speed
improvement.
I think it could also give a (variable) speed up to every consumer of
RealMatrix.getEntry.
https://issues.apache.org/jira/browse/MATH-1654
If changing RealMatrix implems is not an option, there are other specific
solutions to optimize the BOBYQA optimizer.
Let me know what you think.
Have a nice week-end.
Cyril de Catheu
Re: [math] discuss faster getEntry for RealMatrix implementations
Posted by Gilles Sadowski <gi...@gmail.com>.
Le sam. 4 mars 2023 à 17:08, Cyril de Catheu
<cy...@startree.ai.invalid> a écrit :
> [...]
> > > >
> > > Let me know what you think.
> >
> > I've also added comments on JIRA.
> >
>
> Somehow I can't see any comments on JIRA -
> https://issues.apache.org/jira/browse/MATH-1654
> Am I missing something?
I wrote them... And forgot to click on "add". :-{
Done now.
Regards,
Gilles
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [math] discuss faster getEntry for RealMatrix implementations
Posted by Cyril de Catheu <cy...@startree.ai.INVALID>.
Thanks for the quick answer!
↓
On Sat, Mar 4, 2023 at 2:39 PM Gilles Sadowski <gi...@gmail.com> wrote:
> Hello.
>
> Le sam. 4 mars 2023 à 12:29, Cyril de Catheu
> <cy...@startree.ai.invalid> a écrit :
> >
> > Hello,
> >
> > I just opened a ticket
>
> Thanks for the report!
>
> > describing an opportunity to optimize the getEntry
> > method of RealMatrix implementations.
> > For our specific workload in BOBYQA
>
> Advertisement: Help is very welcome for maintaining the
> implementation of that algorithm.[1]
>
Ok I had a look at the tickets, looks interesting. I'd be happy to help
maintain the implementation.
I've recently worked on java code generated from fortran. It seems to be
the case in this implementation - special mention for the infinite loop +
switch acting as a GOTO x) .
I'll start by refactoring.
> > this seems to be a ~10% speed
> > improvement.
> > I think it could also give a (variable) speed up to every consumer of
> > RealMatrix.getEntry.
> >
> > https://issues.apache.org/jira/browse/MATH-1654
>
> Thanks for the analysis.
>
> Side note: You are most welcome to contribute to having JMH
> benchmarks integrated within the component's repository.[2]
>
Noted. Maybe later.
> > If changing RealMatrix implems is not an option, there are other specific
> > solutions to optimize the BOBYQA optimizer.
> >
> > Let me know what you think.
>
> I've also added comments on JIRA.
>
Somehow I can't see any comments on JIRA -
https://issues.apache.org/jira/browse/MATH-1654
Am I missing something?
Regards.
Cyril
> Best regards,
> Gilles
>
> [1] Cf. related issues in JIRA (and in the ML archive).
> [2] By coincidence, another discussion has just been restarted about
> it (for another functionality):
> https://issues.apache.org/jira/browse/MATH-1585
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>
Re: [math] discuss faster getEntry for RealMatrix implementations
Posted by Gilles Sadowski <gi...@gmail.com>.
Hello.
Le sam. 4 mars 2023 à 12:29, Cyril de Catheu
<cy...@startree.ai.invalid> a écrit :
>
> Hello,
>
> I just opened a ticket
Thanks for the report!
> describing an opportunity to optimize the getEntry
> method of RealMatrix implementations.
> For our specific workload in BOBYQA
Advertisement: Help is very welcome for maintaining the
implementation of that algorithm.[1]
> this seems to be a ~10% speed
> improvement.
> I think it could also give a (variable) speed up to every consumer of
> RealMatrix.getEntry.
>
> https://issues.apache.org/jira/browse/MATH-1654
Thanks for the analysis.
Side note: You are most welcome to contribute to having JMH
benchmarks integrated within the component's repository.[2]
> If changing RealMatrix implems is not an option, there are other specific
> solutions to optimize the BOBYQA optimizer.
>
> Let me know what you think.
I've also added comments on JIRA.
Best regards,
Gilles
[1] Cf. related issues in JIRA (and in the ML archive).
[2] By coincidence, another discussion has just been restarted about
it (for another functionality):
https://issues.apache.org/jira/browse/MATH-1585
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org