5 ms·
Generation 50 but none of the cars has learned to pass more than 5-6 corners. Also, changining the track from one interation to another I don't think it helps.
by zer0gravity 10y ago
Generation 50 but none of the cars has learned to pass more than 5-6 corners.
Also, changining the track from one interation to another I don't think it helps. The training should be done on the same track until at least one of the cars is able to complete it. Than it can be changed.
Sometimes some of the cars get stuck and the round doesn't finish, without having to click the "End round" button. There should be some mechanism that detects that cars are stuck and end the round automatically.
- zer0gravity 10y agoOk, generation 60 and now a huge progress has been made. More than 50% of the cars are now able to finish almost all of the tracks. Very cool!
- Lord_Fixer 10y agoThanks :) I actually have had inactivity detection implemented at some point, but I've resigned of this in favor for user-defined stop condition. But you are right, I will bring that back. It's actually expected that in some cases it might take many generations to learn, as the population is rather small. In other attempts they can do this in like 5 generations, so please, tray to restart the whole process. Keeping one track untill at least one car is able to finish it is not as good idea as it seems due to ovefritting. The will simply memorize all the moves depending on some insigificant factors and they won't be albe to finish any other track. I have tested that :D
- zer0gravity 10y agoChanging mutation rate to 0.1 and crossover rate to 0.3 seems to improve the learnig.. 0.5 mutation rate is way too much..
- Lord_Fixer 10y agoYeah, that might be true, and that's what this editor is for :D I didn't want to give users the best solution at the beggining, so they might search for it themselves. It's actally only the third combination of parameters that I've tested. On the other hand 0.5 is not so large, when you look at the mutation procedure. If a speciman is going to be mutated, only 20% of it's weights are changed and only within a range [-0.1,...,0.1]. But fill free to experiment :)
- zer0gravity 10y agoHowever, maybe you will consider the request described here: https://news.ycombinator.com/item?id=12397375 https://news.ycombinator.com/item?id=12397375 For research purposes :)
- Lord_Fixer 10y agoYeah, sure, why not :) I will have some spare time at the weekend, so then I will try to implement evenything that has been proposed in the feedback that i've got.
- retbull 10y agoTook me to gen 11 and I only lost 1 car after that they all completed.
- pmontra 10y agoI clicked End Race in the first few generations when all the cars got stuck in the middle of the road. I think that by the fifth one at least one car was able to complete a lap.
- max_ 10y ago"It is not the strongest that survive, but those that adapt to change " -- Charles Darwin
- sidarape 10y agoI was at Generation 47 and all cars (almost, depending on the new genetic modifications) were able to finish the lap.
- cyrusaf 10y agoI disagree about changing the track. It should be changed every iteration. If the track is not changed every iteration, the NN might develop to work in specific scenarios (only turning left for example) and when run on a different track, might fail miserably.
- zer0gravity 10y agoOn a track with both right and left turns that wouldn't have been the case. While I agree that exposing the NN to diverse tracks each iteration asures a thorough learning, I was kind of curios how susceptible the algorithm is to exploit irelevant particularities of a track instead of finding the general rule. I wish the creator will add the posiblity to chose if the track should be changed each iteration or only after at least one car completed it.