Master thesis - is AI implementation possible?

classic Classic list List threaded Threaded
16 messages Options
Reply | Threaded
Open this post in threaded view
|

Master thesis - is AI implementation possible?

oconer
Hello. I'm a student of Artifical Inteligence and recently I have chosen a master thesis. Goal of my work is to choose some strategic board-game (like Agricola, LeHavre, Carcassone, or Axis and Allies or some similar..), which is implemented allready, and create AI player with variety of approaches, algorithms and strategies. Is it possible to create AI player in some of your games? Howewer, better question is, if I find some help here, when I'll be helpless with source code.
Thanks.
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

eurofabio
Yes it is...but do you know how to code in Java?

oconer wrote
Is it possible to create AI player in some of your games?
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

oconer
Java is my favorit language ;) so I think, as a student of 4.year of applied informatics, I have necessary skills to do things like this. Anyway, my goal will be first analyze whole problem, evaluate some strategies and then implement it, so harder problem, then language, will be to design good AI bot.
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

Veqryn
Administrator
TripleA already has 3 AI's that come with it, all of them simple script/heuristic based AIs.

If you do a google search, you can find some articles written by a guy who wrote a thesis or paper or something about how it is possible to write non-script AIs for such a game as Axis & Allies.  He wrote 2 AI's: a min-max alpha/beta type ai, and a monte-carlo based ai.  
Unfortunately he didn't give me the code for the second one, while the first one was in no condition to be added to the engine.


anyway, w/e you are thinking of, it is possible if you are so inclined and have the time
Please contribute to the TripleA 2013 donation drive:
http://tripleadev.1671093.n2.nabble.com/2013-TripleA-Donation-Drive-tp7583455.html
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

yeklef
I have been hoping for another AI for this game   I hope you pick TripleA.  

I have been thinking about this a little bit recently.  It occurs to me that TripleA has expanded in scope considerably in the last few years (Veqryn's fault I think ).  Because of that, the current AI's haven't really been able to keep up with all of the changes; hell, I haven't been able to .  

Some people in the past have suggested that we do AI's that work with different maps, and I think having some scripted first couple of turns could be beneficial, but I was wondering how hard it would be to make different AI's for different rulesets.  Like one that can handle the basics from the game this was originally based on, i.e. basic movement and attack.  No tech, no air transports, no scrambling, etc.  Another one could make use of some more of the enhancements.  You would just have to standardize some of the enhancements as a package, or if we get AI's that are good with certain packages, maybe people would gear their maps to be played with one AI or another

Any thoughts or ideas?
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

Veqryn
Administrator
IF someone had the time, then they would do the following:

Make 1 very good abstract AI,

within it, it has all the classes needed to play all games "decently"

then, it has extra classes that extend those classes for each "map type".

(ie: specific classes for revised games, aa50th games, global 1940 games, etc etc)
Please contribute to the TripleA 2013 donation drive:
http://tripleadev.1671093.n2.nabble.com/2013-TripleA-Donation-Drive-tp7583455.html
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

yeklef
I think in a world where time, skill, and willingness all came together for an AI, that would probably be best.  

But if, say, I were to work on an AI, I don't really have the time, nor do I really have the skills needed (yeah, I know I should probably do something simpler then, but this question is mostly to help me learn).  In that case, I would think it would be easier to make a simpler AI, and then fork it later on for other, larger, rulesets.  That would keep new rules from breaking the AI for those maps with smaller rulesets.  

That seemed to be the biggest problem when Moore came back briefly and considered updating his AI last July.  There had been so many additions to the game (mostly your fault again Veqryn ) that it would have taken a ton of work to update it to be usable on every (or even most) map.  If he had decided to do his opening moves script only, and kept it for the same basic set of maps, it might not have been such an overwhelming task.  

I know for myself anyway, because I mostly play single player, the AI is really the biggest reason that I don't play more, and that I don't really recommend the game to others yet.  

Maybe the next question should be, is this a bad plan, or just not an optimal plan?
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

rajwarrior
In reply to this post by Veqryn
I'm not a java programmer so I am talking without a lot of knowledge on that subject, but I don't see how you can easily create a generic AI when the maps now are so different in their demands on strategy.  For example, something as simple as a classic game where subs can be killed by planes vs. AA50 or 1940 Europe 2nd where planes cannot touch subs or the hybrid model where planes get one shot before subs submerge becomes a programming nightmare. What would you do, add in tons of "if game is X, then Y, else Z" type codes?

I would definitely think that it would be easier to have a base AI and programmers could modify it to work with specific maps/games.  In general, I would think that you would have much stronger AI's that were programmed for NWO or Big World specifically, than with a generic catchall AI that may or may not be passable depending on the game.  

And I may be wrong, but I would think that you might get more people interested in coding AI as well.  Coding a new AI that works for every game is a major, major task.  Taking a fairly good AI (like Moore N Able) and coding it to play a specific map is a much less daunting task.
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

Veqryn
Administrator
you guys can do whatever you want, and send me the code and patch file

i'm not going to be personally touching the AI's at all, unless there is a null pointer error that needs fixing
Please contribute to the TripleA 2013 donation drive:
http://tripleadev.1671093.n2.nabble.com/2013-TripleA-Donation-Drive-tp7583455.html
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

zlefin
I'd be happy to discuss ai issues or strategy with anyone working on an ai project for triplea.
Having a good grasp of the strategy in a game is always a huge help with ai design.

It's definitely true that the wide game variation affects strategy alot; a fair number of core elements of an ai would remain the same; but alot of things need tweaking to account for various special cases in some games.

Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

yeklef
Is there a list of the rules for some of the more basic rule sets?  I don't know how the more basic rule sets might have diverged or stayed the same as the rules for A&A.
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

captaincrunch
Oconer. I'll post 2 previous threads asking the same questions. I would like to see a stronger ai. I think the current ai is fine but just needs some tweeking (the game ai isn't building fast enough ... perhaps simply it should just buy more artillery) - also, in options there is an option called Change Dynamix AI Settings so if you play around with that I think the game ai could become stronger. I suggested people try some different ai settings and then submit them and then the best ai settings can be used for the final game release. I also gave 1 idea for a "learning ai" using an endgame Tallyfile of the end game objectives/territories captured by the winning side and maybe you could read my post and try to create my idea;


http://tripleadev.1671093.n2.nabble.com/No-Navy-for-AA-Games-tp7581508.html

http://tripleadev.1671093.n2.nabble.com/Interesting-Discovery-if-you-want-AI-to-use-more-sea-movement-tp7582905.html

Hope you are serious about your post and will try to improve this game's ai! We will all be greatfull and look forward to it!


Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

stormyorky
In reply to this post by oconer
It would be great to if TripleA had its AI refined or of a new AI was designed.
Im interested in AI development myself, but I lack java knowledge, although I would like to learn it. I have programmed an Othello AI in the past.

The scope of such a project would differ immensly if it was just designed for 'Axis and Allies' original rules, or if the AI was able to handle any maptype designed for TripleA. A general AI would need to be quite much more abstract.

Although my programming skills with java is almost nonexistant, I do have some past experience with AI and since I only play TripleA single player I have a desire to get a better AI for it - so if there is anything I could help with, let me know!
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

HuskerMike
In reply to this post by oconer
Since all my e-mail opponents have dissappeared in the middle of games, I play only boardgames live locally or computer games solitaire. I like TripleA a lot and play a lot solitaire. However, I approach it differently. Instead of building the AI for the game, I tailor the scenario for the AI.

Computers cannot think. When you are playing against the computer, you are actually playing against the AI programmer. He is severly handicapped. He has to play his complete game before you even start by making a bunch of "if you do that, I do this" statements. Every programmer has weaknesses and habits that can be learned and exploited.

Therefore, I would suggest the order of solitaire game development be:

1. Make the game engine and learn it (TripleA).

2. Make an AI, say More N Able, and learn it.

3. Then, make scenarios (map & units) that allow the AI to play well, considering its strengths and weaknesses.

I have fiddled with a bunch of scenarios and made them a lot of fun to play. They usually are not historical any longer. But, they are a lot of fun.

The challenge would then be to make a scenario from scratch that is both historical and suited for the AI's strengths.
Go Big Red!
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

blumb
HuskerMike wrote
I have fiddled with a bunch of scenarios and made them a lot of fun to play. They usually are not historical any longer. But, they are a lot of fun.

The challenge would then be to make a scenario from scratch that is both historical and suited for the AI's strengths.
                            Check out the way the AI is in GW 1918.

            Though not from scratch it's starting to size up and make a good 1v1 too.
        Balanced right a victorious solo player would be the last standing of the 3 factions.
And when one falls the other remaining opposition should be a threat.

(besides the revolutionaries who need to get/will be triggered to join the Bolsheviks after Finland/Sweden falls)

http://tripleadev.1671093.n2.nabble.com/Great-War-The-Last-Stand-March-1918-tp7584539.html

                             Balance/feature suggestions welcome.
Reply | Threaded
Open this post in threaded view
|

Re: Master thesis - is AI implementation possible?

captaincrunch
Hey Oconer, how goes the project? This is a cool thread and I was hoping some people would start tweaking the ai settings for the Triple A engine and have some tough ai's for us to test!

OR I said a file could be added that saved the end game stats and territories captured and call it a "learning ai" and eventually it would learn the fastest route to victory for whatever scenario and thus we could have a new more tough ai for the Triple A game.

I'll watch this thread for any interest because I like tough tactical challenges in my games.