Jump to content


- - - - -

CAESES 4.4.1 and Dakota


  • Please log in to reply
4 replies to this topic

#1 Mr. hamid reza

Mr. hamid reza

    Advanced Member

  • Members
  • PipPipPip
  • 45 posts

Posted 27 February 2019 - 06:13 PM

Hello all,

 

I found some differences between the generated designs files of Dakota with CAESES 4.4.1 and CAESES 4.1.1. Considering that I am using the same version of Dakota 6.9 for both of them. I also use the same initial population for comparison the results in a multi objective problem.

Previously in CAESES 4.1.1, the generated files of Dakota had the “finaldata1.dat” as output of Pareto designs in the case of using MOGA.

 

Now in CASESES 4.4.1, the following outputs are observed:

  • "run.csv" and "run_best.csv" are added in the output files of Dakota
  • The “finaldata1.dat” shows some “nan” for the same evaluated equality constraints. However, in the CAESES 4.1.1, those value are shown as “0”, which means the constraints are passed according to the convention of Dakota.
  • These are differences between the individuals in the files of “finaldata1.dat” and  “run_best.csv”

So, the below questions come up:

1- Why does CAESES 4.4.1 changes the “finaldata1.dat”?

2- What kind of selection criteria are used for choosing the best designs in the “run_best.csv”? If it tries to copy the individuals in the “finaldata1.dat”, why there are some differences between the designs of the “run_best.csv” and the “finaldata1.dat”?

 

Thanks in advance

Best Regards

Hamidreza


  • 0

#2 Joerg Palluch

Joerg Palluch

    Moderator

  • Moderators
  • 496 posts
  • LocationBerlin, Germany

Posted 28 February 2019 - 08:26 AM

Hi Hamid,

 

If I am not wrong, the finaldata1.dat is generated by Dakota itself. The run.csv and the run_best.csv is generated by CAESES based on the design results table. Here, run_best.csv contains the Pareto designs (or, for single objective runs, the best design).

 

Does this help?

Jörg


  • 0

#3 Mr. hamid reza

Mr. hamid reza

    Advanced Member

  • Members
  • PipPipPip
  • 45 posts

Posted 28 February 2019 - 12:34 PM

Hi Joerg

 

Thanks for your reply.

 

The "finaldata1.dat" is generated by Dakota, and it already contains the Pareto front. CAESES 4.4.1 changes some characters of the file, that is not readable according to conventions of Dakota. However, I do not see these changes when I use the CAESES 4.1.1, so maybe it is an issue that need to be fixed.

 

As the "finaldata1.dat" contains the Pareto members, I do not see any reason to regenerate them in the "run_best.csv", unless otherwise different method for recognition of Pareto front is used to generate the file. Is it so?

Best Regards

Hamidreza 


  • 0

#4 Joerg Palluch

Joerg Palluch

    Moderator

  • Moderators
  • 496 posts
  • LocationBerlin, Germany

Posted 28 February 2019 - 01:46 PM

Hi Hamidreza,

 

What do you mean with "CAESES 4.4.1 changes some characters..."? Like I said, since it is generated by Dakota, CAESES does not touch it. Note that we updated to Dakota 6.9 in Version 4.4.1 which could be the reason for changed finaldata1.dat files.

 

The best.csv file is also generated by all other design engines, i.e., engines that are not based on Dakota. For instance, Sobol, T-Search, Newton-Rapshon, Nelder-Mead, etc. So, this is a more generalized version, to have it always available e.g. in case that CAESES runs in batch mode and you want to do some post-processing (plotting charts etc) with external tools.

 

Cheers

Jörg


  • 0

#5 Joerg Palluch

Joerg Palluch

    Moderator

  • Moderators
  • 496 posts
  • LocationBerlin, Germany

Posted 18 March 2019 - 10:48 AM

Hi Hamid,

 

Indeed, there has been a problem in the output files which will be fixed for version 4.4.2.

 

Thanks a lot for reporting, excellent finding!

Jörg


  • 0