Comments on: Machine Learning and Dragons - a Game http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177 Established January 2005 Sat, 20 Mar 2010 05:00:31 +0000 http://wordpress.org/?v=2.0 by: Brian http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-20656 Sat, 10 Nov 2007 23:38:40 +0000 http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-20656 Hi Greg, I like the game. Maybe you should add some blue fire, because its hotter. Hi Greg, I like the game. Maybe you should add some blue fire, because its hotter.

]]> by: warpedvisions.org » Blog Archive » Machine Learning and Dragons http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-20189 Thu, 25 Oct 2007 05:32:04 +0000 http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-20189 [...] October 24th, 2007 in Links Here’s a short article about a machine learning project written in pygame. It’s an interesting approach with some intriguing possibilities (yes, I’m tempted). [...] […] October 24th, 2007 in Links Here’s a short article about a machine learning project written in pygame. It’s an interesting approach with some intriguing possibilities (yes, I’m tempted). […]

]]> by: Greg - CEO/Founder http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-19936 Sun, 14 Oct 2007 05:21:51 +0000 http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-19936 Davidg, go ahead and download the source, it will make things much clearer. The only actions the dragons have in their repetriore (sp) are attack, move toward knight, and move away from Knight. Then they also have some conditional expressions to apply to those actions, e.g., is knight facing me, how far am I from knight, etc. The intial programs that control the dragons' bevhaviors, made up of those actions and conditions, are generated randomly to start off with. Hopefully having a look at the source will make the idea clearer to you. Davidg, go ahead and download the source, it will make things much clearer.

The only actions the dragons have in their repetriore (sp) are attack, move toward knight, and move away from Knight. Then they also have some conditional expressions to apply to those actions, e.g., is knight facing me, how far am I from knight, etc.

The intial programs that control the dragons’ bevhaviors, made up of those actions and conditions, are generated randomly to start off with. Hopefully having a look at the source will make the idea clearer to you.

]]> by: davidg http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-19932 Sun, 14 Oct 2007 00:52:00 +0000 http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-19932 I might download the source ... but I was wondering how you seeded your dragon population initially? Did you give them an initial traits? Such as "forage for food / look for opponents / sleep / drink". Initial seeding, at random, of the population to give them "some" intelligence might be useful to kickstart the evolution process ... you are God anyways in this small world. Then you might be able to gauge the effectiveness of traits a dragon has in its lifetime, instead of the raw length of their lifetime. I haven't written GPs before, but I have dabbled in GAs (particulary the N-Queens problem in back when I was in school) and I know the paradigms are siblings. I also agree with the post above that you could make this a web-app. You could distribute many worlds and have newcomers face evovle(ing) generations of dragons trained from both AI bred breeding arenas and human-vs-dragons bred breeding arenas :) i I might download the source … but I was wondering how you seeded your dragon population initially?

Did you give them an initial traits? Such as “forage for food / look for opponents / sleep / drink”. Initial seeding, at random, of the population to give them “some” intelligence might be useful to kickstart the evolution process … you are God anyways in this small world.

Then you might be able to gauge the effectiveness of traits a dragon has in its lifetime, instead of the raw length of their lifetime.

I haven’t written GPs before, but I have dabbled in GAs (particulary the N-Queens problem in back when I was in school) and I know the paradigms are siblings.

I also agree with the post above that you could make this a web-app. You could distribute many worlds and have newcomers face evovle(ing) generations of dragons trained from both AI bred breeding arenas and human-vs-dragons bred breeding arenas :)

i

]]> by: Greg - CEO/Founder http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-19918 Sat, 13 Oct 2007 17:10:06 +0000 http://www.blendedtechnologies.com/machine-learning-and-dragons-a-game/177#comment-19918 Harrison, neat ideas. Where does Hofstadter come into this? :-) That's an excellent point about survival time only being helped by cowardess, and luck (starting distance from knight). Maybe some sort of health measure would help with that. So the knight has to hit a dragon multiple times, and vice versa. The knight army is a neat idea too. The thing I don't like about having only killing the knight giving a dragon fitness in the current setup is that it makes the fitness landscape very non-smooth. I.e, a dragon can be very aggressive and have excellent strategies but if another dragon kills the knight first, he doesn't get rewarded. (Of course survival time probably doesn't help with that either and probably actually discourages aggressive behavior.) So I'm not sure actually. I'm not sure what my purpose in creating the game was. I think to see machine learning algortihms in an interactive environment. And secondly to make an entertaining game. Harrison, neat ideas. Where does Hofstadter come into this? :-)

That’s an excellent point about survival time only being helped by cowardess, and luck (starting distance from knight). Maybe some sort of health measure would help with that. So the knight has to hit a dragon multiple times, and vice versa. The knight army is a neat idea too.

The thing I don’t like about having only killing the knight giving a dragon fitness in the current setup is that it makes the fitness landscape very non-smooth. I.e, a dragon can be very aggressive and have excellent strategies but if another dragon kills the knight first, he doesn’t get rewarded. (Of course survival time probably doesn’t help with that either and probably actually discourages aggressive behavior.) So I’m not sure actually.

I’m not sure what my purpose in creating the game was. I think to see machine learning algortihms in an interactive environment. And secondly to make an entertaining game.

]]>