I discovered, when I linked DIVA with galapagos, that when I run the radiation map with the same settings on the same simple single geometry the results vary a bit each time. So, that I never get the same result twice for the completely same analyse. (I tried each quality setting and of course the variation is getting smaller but there is always one)
I tested it with a single surface that I make rotatable for 360° and combined it with galapagos. (See the gh file attached ) Here I saw that galapagos of course doesn´t find a solution as the results are never stagnant and further that the variation of the results is bigger than the difference from about 5°. Of course, at this example its obvious which is the best solution but for more complex problems I already tried it makes it hard to pick the best solution.
Does someone have an idea how to solve that problem especially as for parametric with galapagos I need to work with the lowest quality to keep the time the simulation needs to a minimum.
Run-to-run variance is expected. You could reduce it by adding to the radiance parameters a "-u" argument, which turns off uncorrelated random sampling. (The parameters are a hidden input on the RadMap component.)
It's worth noting, however, that the run-to-run variance is telling you something important about the fidelity of the simulation. If the run-to-run variance is similar in magnitude to the difference between two design options, Galapagos really has no business selecting between those options in the first place.