Re: [Libmesh-users] early greedy training termination

[Kyunghoon L. <aer...@gm...>]

Hi David,

Thanks for the explanation.  I did a little experiment by lowering
training_tolerance, and found the greedy process got terminated because it
chooses the same parameters even before getting near to the specified
training_tolerance.  For instance, I used reduced_basis_ex5 and I got the
following (the outputs are tip displacements in x & y) for
training_tolerance = 1e-05

use_relative_bound_in_greedy = true

---- Basis dimension: 9 ----
Performing RB solves on training set
Maximum (relative) error bound is 0.000486633

Exiting greedy because the same parameters were selected twice

truth output[0] = 22.7201
truth output[1] = -36397
rb output[0] = 22.7196 +/- 3.48492
rb output[1] = -36397 +/- 69.9526

Now I lowered training_tolerance to 1.e-10, and I got the following:

use_relative_bound_in_greedy = true

---- Basis dimension: 8 ----
Performing RB solves on training set
Maximum (relative) error bound is 0.00181167

Exiting greedy because the same parameters were selected twice

truth output[0] = 22.7201
truth output[1] = -36397
rb output[0] = 22.7209 +/- 2.24415
rb output[1] = -36397 +/- 45.0467

As you can see from the two different training tolerance sets, I end up
with similar results despite different tolerance sets.  Because of this
early termination, I cannot get a fine tuned RB model having very low error
bounds.  Can you suggest me some advice, plz?

Best,
K. Lee.

On Mon, Jul 16, 2012 at 8:14 PM, David Knezevic
<dkn...@se...>wrote:

> Hi K,
>
> I changed the code so that (by default) it terminates if the greedy
> chooses the same parameters twice, because that should not happen. (It
> typically happens when the error bound is dominated by rounding error.)
> I think the issue here is just that your training_tolerance is too low.
>
> David
>
>
>
> On 07/16/2012 08:10 AM, Kyunghoon Lee wrote:
> > Hi all,
> >
> > I made a few reduced basis models based on the cantilever example,
> > reduced_basis_ex5, and I found my models always terminate the greedy
> > training early before satisfying either Nmax or training_tolerance.
> > Despite increasing n_training_samples, I keep getting
> >
> > Exiting greedy because the same parameters were selected twice
> >
> > , and I cannot find what causes the problem.  I'd appreciate for any
> > suggestions.
> >
> > Regards,
> > K. Lee.
> >
> ------------------------------------------------------------------------------
> > Live Security Virtual Conference
> > Exclusive live event will cover all the ways today's security and
> > threat landscape has changed and how IT managers can respond. Discussions
> > will include endpoint security, mobile security and the latest in malware
> > threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
> > _______________________________________________
> > Libmesh-users mailing list
> > Lib...@li...
> > https://lists.sourceforge.net/lists/listinfo/libmesh-users
>
>
>
>
> ------------------------------------------------------------------------------
> Live Security Virtual Conference
> Exclusive live event will cover all the ways today's security and
> threat landscape has changed and how IT managers can respond. Discussions
> will include endpoint security, mobile security and the latest in malware
> threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
> _______________________________________________
> Libmesh-users mailing list
> Lib...@li...
> https://lists.sourceforge.net/lists/listinfo/libmesh-users
>