Re:Genetic Algorithm building a little car
by vorador on Monday December 08, 2008 @01:59PM
(#390168)
I wonder if the car that best fits the constraints is always the same (i.e : do we see the same car, or each user sees a different car) ?
[ Parent
| on HN ]
Re:Genetic Algorithm building a little car
by jwilliams on Monday December 08, 2008 @09:19PM
(#390816)
I'd like to try and compete with an intelligent design - see how many generations it takes for the GA to beat me.
Open another instance in a second window and you'll quickly see the answer.
Only one car would best fit the constraints (that's by definition). However, it's probably not the car you are seeing, because the algorithm tends to get stuck on local maxima. After about 2 hours of running, the thing has flatlined as each window I opened has 20 copies of pretty much the same car (although they differ between windows), and they are not getting that far.
Due to the naïveté of the model I would say it's pretty unlikely that there are many different car-combinations with high fitness.
Has anyone bred one than gets past the dip after hill and doesn't somersault? The ones I have now seem to have optimum suspension for quickly motoring over the initial bumbs, but they all die in that damn ditch...
In the second tab I opened there is a car that makes it past the point that causes most cars to flip. The run currently times out at the top of the next hill, but the distance is still slowly improving.