Which isn't really difficult, if you know how to do it. Though it might take some time to code in, depending on whether the coder you'll get has experience with making that sort of stuff.
The only thing there difficult is the AI, which would be rather fun to make. I really want to use perceptrons for it.
For the sample game they want? Yeah, it would be a fun thing to do. I don't know what exactly do you plan to make it play like, but personally, I'd just make it into a priority-based system (random targets, but higher chances of targeting opponents who play aggressively or steal, with cards being semi-randomly chosen based on moves from the last two turns, and additional chance to instead hit someone who repeats his moves for a few turns straight with anything that could counter the frequently played card) so that it would emulate human players instead of being as perfect as possible, but then again, I don't have time for that so I'm just writing that here for the heck of it.
I wouldn't approach a card game with neural networks. You have a discrete set of states with well defined rules and transitions known in advance. You're not needing to do unsupervised or supervised classification of values in a continuum.
He only wrote that he *wants* to do it. He likely knows that it's not the best solution for something on this level, but sometimes people want to have fun while they work.
Mine too, Essence. Which is why mods of my code is a convoluted mess that no same person should look at, and it's why I'm writing them for Heimdallr instead of letting people choose their bits of insanity.
It's the easiest way in my opinion to seed it well to the proper option for each circumstance, which isn't always obvious; given that this game has a lot of rock paper scissors gameplay it's important to be able to guess what the opponent is going to do. It would be easy enough to simulate some semblance of yomi. My assumption is I could do better just seeding a neural network than trying to write the entire AI myself and have it be flawed. Also I don't know that much about AI. And yes I really do want to code one of these for fun.
There is a problem with this approach for games though. If you make your neural net big and powerful enough to learn the game, it will likely eventually start to consistently outsmart the player (not to mention the performance overhead!). Which won't be very fun. A behavior-based, non-learning sort of thing might well be better. Difficulty could be adjusted too by tweaking things like the depths of search trees. Basically, getting the AI wrong is really what you're after, here, anyway. Getting it right doesn't make games fun.
The game he is supposed to make now is just a test to see if he's good enough to work on Fax's and Essence's project. So it doesn't really matter if it becomes impossible to defeat, really, and he wants to do that for himself so there's nothing wrong with that.
Well honestly I wouldn't plan on using such an AI for any test app. You could see the code in there it's absolutely horrendous because I don't plan to ever do much with it. Were it the case that I would I'd rewrite most of it, including an AI. For the test APP it would probably be rather rudimentary. Also it wouldn't modify itself at all during play or anything, it would just simulate a human player.
Which can also be achieved by other means, but I do see why you want to use that, even without taking the "I want to try" part into account. And I kind of agree with that approach.
Well yes if you know how to do it then it isn't but individually simple things can be much more complicated when you have to keep in mind how they interact with each other. @Fax Will it be 2D or 3D?
Oh hey, I have an FFT engine kicking around somewhere from a previous Gaslamp project. Fun stuff. Hard to make look good on PC, though.
