My senior capstone?

[PMMurphy]

So im only a sophmore in university but i am already planning on "generally" what i want to do that will help get me into grad school. I also want to do the same topic throughout grad school aswell. For my masters and PhD.

Enough of that.

Its basically this. I'm going to create an agent that plays either checkers or chess. It will be created with a system of neural networks that will be trained using evolutionary computation. In other words it will learn and play the game by itself without me telling it anything other then the bare basics and mechanics of the game. No gameplay knowledge will be told. This is already done for checkers, so i want to look into chess. Chess might be a bit too ambitious of a project, and someone is already doing it. So i'm going to fall back down on checkers. I think i will come up with some small little technique to making it better or improving on it if i can and just call that my capstone.

Criteria for a capstone is; Must be new or an improvement on something already available.

But i basically want to make a game playing agent for a video game.

Any suggestions or questions?

[Mownkey]

Quick question: what language?

[egg82]

At 9/2/12 09:56 AM, PMMurphy wrote: Any suggestions or questions?

ooh, an AI! I love designing AI! Coding them is a different story...
And a self-learning AI as well. I like this challenge :P

The hardest part of an AI is breaking down what you know into small steps that a machine can follow, so i'll help out as much as I can there.

okay, so the basic mechanics are obviously going to be hardcoded into the AI.
The learning... Hmm... I suppose the question is: How do we learn? How did you first learn to play checkers? I honestly don't play checkers much at all, I prefer chess. I actually forgot the rules and had to look them up again.

So how did I first learn to play checkers? I learned to avoid putting my pieces into positions that could be chain-jumped (putting pieces next to eachother is good), and to attempt to keep the pieces in the back at the back if possible. That's defense 101. Attack 101 is to kind-of force your way into your opponent's rows while keeping your piece defended by putting pieces behind it or moving to the edge.

I suppose that wasn't really as complex as I thought it'd be.

[egg82]

And the second part: learning.

If you lose, you need to analyze what happened, which means every move made needs to be stored somewhere.
The biggest issue is that you have to figure out what went wrong.

Any time you lose a piece should be taken into consideration, and a point where all you pieces start disappearing should be analyzed in more depth. There could be a point for optimization here, since "winning" or taking the opponent pieces doesn't need to be analyzed and can be thrown away. Maybe.

the position of the pieces should be analyzed in the event of a piece being taken. Where were your pieces and where could they have been to avoid being taken? Then if you get into that situation again, just look back into your memory and simply take the path you had previously figured out.