Closed lu-kas closed 4 years ago
Hi Lukas, thank you for your message. However, I am not 100% sure if I understand you plan correctly. Do you mean the parameter-set of the different complexes (ngs)? I guess this should be possible somehow, but is not implemented in spotpy. If you mean the all parameter-sets within a complex that would not be possible, as they are generated depending on the previous run within the complex. Details about the parallelization of the sce-ua algorithm as it is implemented in spotpy can be found here: http://vuir.vu.edu.au/767/1/AParallelShuffled_s48_Muttil_.pdf Hope this helps a bit?
Hey Tobias,
thanks for the quick reply -- ok, I understand. Btw, I was interested in the latter, i.e. all or a subset of the parameter within a complex.
I will have a closer look into the algorithm, but I feel, that there will be no further potential for parallelisation.
@lu-kas I am afraid I think so too, there is no further potential to speed up the parallelisation, due to the design of sce-ua. However, if you find potential, feel free to reopen this issue.
Dear all,
this may be a naive question: Is it possible to get a full list of all parameter sets (SCE), which will be evaluated in a generation?
Background: We would like to spawn all model evaluations computed during a generation at once. This way we hope to reduce the time to solution for a singe generation by utilising a high number of cores. In our setup we have to run a few thousands simulations, each about a few minutes, per generation.
Goal: Having such list, we would like to process it at once and provide spotpy the full result list for further evaluation.
If this is not possible right now, I would like to evaluate options to implement it, as this feature would make us more productive.
Best and thanks Lukas