I have 38 inputs to my problem, but these 38 inputs are grouped into 12 groups. I allow 6 of these groups to change at once, which means in total there are 924 permutations of these groups that are being used as inputs to a simulation. For each of these permutations, I calculated the "decrease in Variance Accounted For (VAF)." This quantity isn't very relevant to my question, it's just a value I calculated for each set of inputs. It looks like this:
Now, I'm looking for a way to sample the set of 924 permutations, to obtain a set of inputs that are representative of the variability in the above bar plot. Ideally, each permutation could just be the inputs to one simulation, and I could run 924 simulations. But due to the nature of the work, this would take much too long. I need a way to pick a handful of these permutations and use those as inputs to only a handful of simulations, but I want to make sure the permutations I pick are representative. I've thought about random sampling until the mean/stdev (of the decrease in VAF) of the smaller set is within 10% of the mean/stdev of this entire 924 set. I've also been thinking of Design of Experiments, but I'm not sure that's exactly what I'm looking for. I already know the effect of each of the permutations (as seen in the bar plot). I'm also not trying to pick the "best" permutation to use, so I wouldn't say this is parameter optimization. I need a range of the VAF values from the input permutations to use.
Thanks for any advice!