I even tried sliding them both to 1 manually, and it just put them right to 0.
I’m thinking that maybe it just figures out immediately that it is fitter at 0, because my fitness is the product of all sliders multiplied together (x+1 on the 0…1 sliders). But if that’s the case I would think it would still allow them sometimes.
Can someone please help me understand why not better?
EDIT: Wait my guess seems likely can’t be it because my fitness target isn’t min or max, it’s 6000, so seems like it should keep trying both…
EDIT 2: FYI, they are both supposed to be toggling two sets of stream gates and filters, to perform not perform mirrors in x and y directions. They and the entire model all work properly when I adjust the sliders manually.
EDIT 3! Hmmm, now I see that late in the process they are starting to very rarely blip to 1 then right back to 0. I’d still like to know why, as I rather mostly 1’s instead of mostly 0’s.
Hey !
Pretty hard to check without file.
I think I already hit a problem like that once (with slider 0/1 without decimal).
Have you tried with genepool instead of sliders?
What happens in the very first generation (which is random in GA) You could increase the population there to see if it triggers more “1” on your sliders. Or play with mutation rate.
But the thing is that : if you never get 1’s in the very first generation, it’s that it’s a bug.
To avoid it, try to set the slider to have one decimal (0.0 to 1.0) and plug it to integer parameter (which will round to closest value).
Again, easier with a simplified file…
I’m thinking that maybe it just figures out immediately that it is fitter at 0
combined with
Hmmm, now I see that late in the process they are starting to very rarely blip to 1 then right back to 0
implies that late in the simulation all off your population chooses 0 as the fitter value and only mutations sometimes consider 1.
But without a file those are just guesses.
If you want more variation in the 0…1 sliders try changing your fitness function
Thank you both for the valuable inputs, I appreciate it.
Fairly huge algorithm, 10 galapagos sliders and many more set manually, proprietary, not a major problem, combine to deem that far in-depth help unecessary. I can easily work around to get the desired results. I am about a month in to GH and very happy with my progress and new found capabilities, just looking and thankful for insights from more experienced perspectives, so thanks again!
I’ll be sure to try all of the above considerations, especially the float to integer trick and others I’ve considered such as trying to bias variably in favor of 1 through expressions, which is basically the same as the float to integer method.