Mr. hamid reza 1 Report post Posted January 19, 2018 Hello all, I am using the DAKOTA as optimization engine for solving my problem. Unfortunately, the process is crashed after preforming a lot of time consuming evaluations. But fortunately, DAKOTA can restart the process via the “DAOTA.rst” file. In the DAKOTA manual mentioned that, the restart option starts the process from the beginning but uses the recorded previous evaluated functions(which is saved in the ".rst" file) and would continue from the point which is left without repeating the same function evaluations.Is there any way to implement such useful option in the DAKOTA engine of CAESES? Best RegardsHamidreza Share this post Link to post Share on other sites
Jörg 29 Report post Posted January 19, 2018 Hi Hamidreza, In general, CAESES can recycle existing designs. So if there is a crash, and you start your engine again (any engine incl. the Dakota engine) there is the option to use a result pool. Since the setup of your engine is untouched, it will run the same optimization again and takes existing designs from the data base, i.e. no expensive simulations will be triggered twice. Maybe you mean something else? Which process exactly crashed at which stage? Was it Dakota, or CAESES, or the simulation etc? CheersJoerg Share this post Link to post Share on other sites
Mr. hamid reza 1 Report post Posted January 19, 2018 Hi Joerg Thanks for the reply. Actually, I do not know the reason of crash. I am using the MOGA for the optimization, when I used the previous “results pool” for new start, CAESES generate new results and starts from beginning, some of the designs are referred to the marked "results pool" and some of them are not referred. For example, imagen my goal is reaching to the 4000 Max. Evaluation. Then, the program is crashed after evaluating of 1000 individuals. If I understood your suggested method well. When I use a results pool for repeating the optimization process, the CAESES start new optimization process from 1 to 4000 evaluations. If any of the provided individuals exists in the pool, it uses the evaluated results pool and do not calculate those ones. So it means, it do not starts from the point which is crashed.Is my understanding right? If yes, there would be a difference with DAKOTA restart option. Because “the new executions of Dakota can pick up where previous executions left off.” So, it means, if I could use restart option of DAKOTA in my case, the DAKOTA engine could preview the history of optimization from the previous trial. In the new start, the GA would start form the 1001 individual and continue to the 4000. Best RegardsHamidreza Share this post Link to post Share on other sites
Jörg 29 Report post Posted January 22, 2018 Hi Hamidreza, Yes, you are right. We decided to use our CAESES result pool instead of the data base in the restart file (for some internal technical reasons). Besides the waiting time until the CAESES re-run reaches the design 1001, it is exactly the same behavior. I guess in your case the additional waiting time is sort of significant so I get your point. CheersJoerg Share this post Link to post Share on other sites