I’ll start by saying this: ‘random’ isn’t a generative method to base a design on. If you use it, you’re not going to get reliable results the next time out. There are random looking methods (ie: select a white noise image and sample given pixels via a controlled shuffle) which might get you there, but as they aren’t truly random you won’t be exploring the entirety of the design space under any circumstances. Also, because the ‘adjacent setting’ will return an unrelated value to the previous study, there is no pattern to the results based on any input you needn’t bother with the optimize button, instead use really big randomize studies instead. With that out of the way…
Instead of using the random values, in the ‘record’ portion of your graph, pull the values which were generated by generative design either by writing them to a new file or pulling the top performer from your hall of fame data. This will ensure you’re not using new random values but instead using the values which happened to work out for the best.
It’s likely that there is a way to do this without relying on a random value, but without knowing where you are I can’t give you directions to the town hall.
Pretty much the conclusion I’d come to… Irritating but… meh, it is what it is… Maybe I’ll think of a method that doesn’t involve a random start… but so far it’s winning by miles.
Thanks