Indie game storeFree gamesFun gamesHorror games
Game developmentAssetsComics


Neural-net evolution in a chill aquatic environment · By urocyongames

Another run, another request

A topic by lvngbth created Jun 09, 2021 Views: 120 Replies: 1
Viewing posts 1 to 2

The main run is still continuing on this PC. Critters have got smaller again.

Last night, I started another run on a slower PC. I deliberately made things a bit tougher (less fruit spawning, higher reproduction cost) because the main one is always bouncing at 149 or 150 Critters.

By the morning, it'd done just over 100 generations. There were only about 50 of them - that worked! - and they were all yellow. At lunchtime, another 100 or so generations had happened, about 80 were alive - and they were all white.

I am wondering if there was a severe population crunch or two and everyone's the descendant of one yellow mutation, because there's no obvious meat hunting going on to provide a big evolutionary push to change colour.  But it's impossible to tell!

Can we have an (optional) log file that records births / deaths (inc health expiring and being eaten) / sex / non-neural net mutations? Just text is fine - the people who will be into this can do their own graphs / stats analysis.

A 'save game every x generations' feature would also be helpful, especially if the random number generator is one where running with same start position with the same seed = same result, so you can replay what happened between two saves. If it's not, having one that is would be good.

(I really hoping I am not the only one playing with this :) If you're reading this, it's fascinating - BUY IT!)


That's really funny -- I was just performing an experiment because I'd also noticed the weird tendency for mono-colored populations and wanted to make sure it wasn't a bug!

I divided the level in two with obstacles and sure enough, the two populations diverged pretty quickly. On the left, color diversity took off quickly, going through pink, orange, and yellow before settling on white. On the right, there was very little change for a very long time, with the population staying very close to the original red, before eventually starting to diversify into pinks and purples. So, at least I know that works! The popular colors are just out-breeding the rest.

Some advanced logging would be really neat. I bet you could extract some really cool data! I've been mostly hesitant to add new options to the game simply because of UI clutter -- there's already so much going on in menus. But as the list builds up I think I'm seeing the need for at least a dedicated "Settings/Options" pane. So I will probably look into adding autosaves and logging along with that.

As for the random generator, NeuraQuarium does *not* use a seeded generator -- which means that, yes, when you reload a save, you are going to see a different result.

That's something that could potentially be changed (or even have an option added) but it would require a little more investigation and work. There are a LOT of random events, and pursuing a goal of total "seed determinism" would even require changes to the save/load system, as some level elements (grass/weeds) do not preserve EVERY aspect of themselves -- when you load a save, the grass might be slightly different colors or sizes, because only the location is saved. So, that's not a "no," but it's a "not tomorrow" ;)