[ot-users] duplicate with SobolIndicesExperiment

roy roy at cerfacs.fr
Sat Dec 9 10:14:28 CET 2017


Hi Sofiane,

Good to know for the enhancement plans.

I am talking about the post processing. I need to be able to interrupt the python script between the generation of the experiment
and the post processing. In some case, using the same experiment I get the correct indices, and sometimes indices are off.

Thanks.

Sincerely,

Pamphile ROY

> Le 7 déc. 2017 à 20:30, HADDAD Sofiane <sofiane_haddad at yahoo.fr> a écrit :
> 
> Hi Roy,
> 
> We will probably enhance the experiment class. It may reduce 2N evaluations, which is substantial (only in 2d case)
> 
> What do you mean by "not computed correctly" in your mail? Are you talking about generated experiment or post processing (SaltelliSensitivityAlgorithm?)
> 
> Sofiane
> 
> Le jeudi 7 décembre 2017 à 10:42:29 UTC+1, roy <roy at cerfacs.fr> a écrit :
> 
> 
> Hi Sofiane,
> 
> Thanks for the update.
> 
> Do you plan to have it implemented on OT? Or should I handle this in my package?
> 
> Also, I have a last interrogation. When I create the sample with SobolIndicesExperiment,
> I found that Sobol' indices are not always computed correctly. I am not able to provide a MCVE here as
> from the sample generation to the computation of the indices, a lot happens. Sorry.
> But from my trying, if I set OT’s seed, I get correct results.
> 
> 
> So is there a way to ensure that the sample generated by SobolIndicesExperiment will correspond to
> what is expected by the indices classes? Seems that there is a randomization effect here.
> 
> 
> 
> Thanks again for the support.
> 
> Sincerely,
> 
> Pamphile ROY
> 
> 
>> Le 7 déc. 2017 à 00:39, HADDAD Sofiane <sofiane_haddad at yahoo.fr <mailto:sofiane_haddad at yahoo.fr>> a écrit :
>> 
>> Hi
>> 
>> You are right  only the case dim=2 has duplicates if compute second order is set tot True.
>> I miss it, sorry!
>> 
>> An enhancement is to generate samples of size N * (dim + 1) in case dim=2 whatever second order is True or False
>> 
>> Thanks for the report.
>> 
>> Regards,
>> Sofiane
>> 
>> 
>> Le samedi 2 décembre 2017 à 13:25:25 UTC+1, Pamphile ROY <roy at cerfacs.fr <mailto:roy at cerfacs.fr>> a écrit :
>> 
>> 
>> Hi Sofiane,
>> 
>> I got it now.
>> 
>> But if in 2dim this behavior is to be expected, why not doing this internally?
>> The root of this was that I have an expensive numerical model. So having to compute twice a sample is not tractable.
>> 
>> Thanks for your support.
>> 
>> Sincerely,
>> 
>> Pamphile ROY
>> De: "HADDAD Sofiane" <sofiane_haddad at yahoo.fr <mailto:sofiane_haddad at yahoo.fr>>
>> À: "users" <users at openturns.org <mailto:users at openturns.org>>, "roy" <roy at cerfacs.fr <mailto:roy at cerfacs.fr>>
>> Envoyé: Vendredi 1 Décembre 2017 13:31:14
>> Objet: Re: [ot-users] duplicate with SobolIndicesExperiment
>> 
>> Hi
>> 
>> There is no problem here
>> 
>> You can find here how the experiment is defined. As you set second order to True and your sub samples are of size 5, you have (2 * 2 + 2) blocks of size 5 Have a look at SobolIndicesAlgorithm — OpenTURNS documentation <http://openturns.github.io/openturns/latest/user_manual/_generated/openturns.SobolIndicesAlgorithm.html?highlight=sobolindices#openturns.SobolIndicesAlgorithm>
>> SobolIndicesAlgorithm — OpenTURNS documentation
>>  <http://openturns.github.io/openturns/latest/user_manual/_generated/openturns.SobolIndicesAlgorithm.html?highlight=sobolindices#openturns.SobolIndicesAlgorithm>
>> 
>> 
>> Using this :
>> import openturns as ot
>> ot.RandomGenerator.SetSeed(0)
>> distribution = ot.ComposedDistribution([ot.Uniform(15, 60), ot.Normal(3000, 400)])
>> sample = np.array(ot.SobolIndicesAlgorithmImplementation.Generate(distribution, 5, True))
>> print(sample.reshape(-1,5,2))
>> 
>> you will see all matrices defines in link.
>> Hope this helps
>> 
>> Bien cordialement,
>> Sofiane HADDAD
>> 
>> 
>> Le mercredi 29 novembre 2017 à 19:30:39 UTC+1, roy <roy at cerfacs.fr <mailto:roy at cerfacs.fr>> a écrit :
>> 
>> 
>> Hi,
>> 
>> I am using ot.SobolIndicesExperiment and if I set my input dimension to 2,
>> I get some repeated points. From my understanding of the method, this should not be the case.
>> Also going with a higher dimension does not do that.
>> 
>> Am I wrong?
>> 
>> Here is an example (same behaviour with the old method available in OT 1.9) where I highlighted some duplicate:
>> 
>> import openturns as ot
>> distribution = ot.ComposedDistribution([ot.Uniform(15, 60), ot.Normal(3000, 400)])
>> sample = np.array(ot.SobolIndicesAlgorithmImplementation.Generate(distribution, 5, True))
>> 
>> array([[   29.65970938,  2535.47991432],
>>        [   33.01991727,  2559.28624639],
>>        [   33.25474751,  2682.95080229],
>>        [   32.95380182,  2419.44678937],
>>        [   55.23575378,  3039.33121131],
>>        [   26.05095231,  3271.18330661],
>>        [   41.00594229,  3683.75154513],
>>        [   54.81729255,  3428.24812578],
>>        [   28.2423326 ,  2797.23010815],
>>        [   52.36310769,  2335.65441869],
>>        [   26.05095231,  2535.47991432],
>>        [   41.00594229,  2559.28624639],
>>        [   54.81729255,  2682.95080229],
>>        [   28.2423326 ,  2419.44678937],
>>        [   52.36310769,  3039.33121131],
>>        [   29.65970938,  3271.18330661],
>>        [   33.01991727,  3683.75154513],
>>        [   33.25474751,  3428.24812578],
>>        [   32.95380182,  2797.23010815],
>>        [   55.23575378,  2335.65441869],
>>        [   29.65970938,  3271.18330661],
>>        [   33.01991727,  3683.75154513],
>>        [   33.25474751,  3428.24812578],
>>        [   32.95380182,  2797.23010815],
>>        [   55.23575378,  2335.65441869],
>>        [   26.05095231,  2535.47991432],
>>        [   41.00594229,  2559.28624639],
>>        [   54.81729255,  2682.95080229],
>>        [   28.2423326 ,  2419.44678937],
>>        [   52.36310769,  3039.33121131]])
>> 
>> Thanks for your support.
>> 
>> Sincerely,
>> 
>> 
>> Pamphile ROY
>> PhD candidate in Uncertainty Quantification
>> CERFACS - Toulouse (31) - France
>> +33 (0) 5 61 19 31 57
>> +33 (0) 7 86 43 24 22
>> 
>> _______________________________________________
>> OpenTURNS users mailing list
>> users at openturns.org <mailto:users at openturns.org>
>> http://openturns.org/mailman/listinfo/users <http://openturns.org/mailman/listinfo/users>
>> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openturns.org/pipermail/users/attachments/20171209/8b571a66/attachment.html>


More information about the users mailing list