Improving AI
-
@alkexr How do I improve AI?
-
@Captain-Crunch said in Improving AI:
I will start to request and suggest again for the 4th or 5th time that we could make a "learning AI" by simply saving a file called "WinnersCapturedTerritories" at the end of a match to be used to help direct the AI with its decisions for which Territory to choose when faced with an option between several Territories to attack!
My suggestion is easy and applies to ALL MAPS and only requires 1 additional algorithm and would improve the AI as more games are played!I don't think TripleA has a large enough active player base for that. We'd need thousands of games for any meaningful machine learning, for every map. That is if we could create an efficient architecture, which is unlikely considering that no one around here has experience with machine learning (that I know of).
@Schulz First of all, you'll need some Java coding skills. There are plenty of good tutorials on the internet explaining how to do that. Once you are comfortable with that, you should get further direction from redrum. But this will take some effort and time.
If that seems like too much for you, but still want to help, you can share feedback about situations where the AI messes up badly, or post any relevant ideas in this thread.
"For the world is changing: I feel it in the water, I feel it in the earth, and I smell it in the air."
-
Triplea would have enogh active player if AI was more challenging. Decent AI is must for gaining enough players for almost all games (maybe except sandbox ones).
AI can no way act as good as human but it should be challenging with modified incomes at least. AI USA still constantly invade Congo in some WWII maps. In this situation no matter how much bonus will you give to USA.
-
@alkexr you don't understand what I said. My idea isn't about the AI needing to play lots, what I said was THE AI GETS BETTER THE MORE IT PLAYS ... so each copy of the game played by different players around the world would have the AI learn after 2 games, after 3 games, after 4 games etc. etc. etc.
So an AI from a person's downloaded version of the game say in Boston, could play against the AI of a persons version of the game on the same map who say lives in Toronto ... and PERHAPS the AI that had more played games with more data saved to the "WinnersCapturedTerritories" file (the data adds/accumulates as more games are played to list a hierarchy of top Territories that got captured!) should maybe be better than the AI of the person who's copy of the game had only been played a couple times on that map
Again, I didn't express every tiny little detail in this thread about this "learning AI" but its pretty straight forward and easy to understand for most people if you read it again
And the thought did occur to me perhaps we could even some day play this "Learning AI" versus the original AI just to see which is better!
-
@Captain-Crunch Yeah, that AI will get better the more it plays... but it will need thousands of games to reach the level of a crazed monkey slamming the keyboard.
"For the world is changing: I feel it in the water, I feel it in the earth, and I smell it in the air."
-
@Captain-Crunch Now, seriously, ML technology is fundamentally data driven at the present day. I don't see how, given the limited data available, your learning AI would be better than simply having a human go over the map and pull rough estimated values out of somewhere for every territory and saving those in an AI configuration file. Even a bad human player would outperform the learning algorithm (especially one we could create).
-
@alkexr how many accounts do you have? There isn't a forum I've been at where there isn't a troll having conversations with himself lool
Anyways, if you have any grasp of math at all, you'd realize that my idea for a "Learning AI" clearly shows IT ONLY TAKES 1 MATCH TO START IMPROVING THE AI.
That is the topic in this thread by the way ... improving the AI ... so maybe it would literally only take 2 or 3 runs of this "Learning AI" on any map to vastly improve the Ai like literally 30 or 40% more wins.
So I dare anyone to suggest any better idea to improve the AI more than mine that only needs 1 saved file and maybe 2 extra lines of code that would improve the AI 30 to 40% more after only 2 or 3 matches of "learning"
-
@Captain-Crunch Limited data can actually make the AI worse. But I don't see how my alleged poor math skills make your point more true.
EDIT: I hope it's obvious that the post about the monkeys was a blatant exaggeration for audience score, not an accurate assessment.
-
@alkexr in math, "adding" means like 1+1+1 = 3 ...
so the "Learning AI" would ADD data as more matches are played. The AI would be improved even after JUST 1 MATCH!
I didn't say anything about limiting the AI to only playing 1 match but I do realize this requires people who understand math or programming so let's wait for some informed and serious replies
-
@Captain-Crunch said in Improving AI:
I didn't say anything about limiting the AI to only playing 1 match but I do realize this requires people who understand math or programming so let's wait for some informed and serious replies
Why did our conversation get out of control so quickly? Is it that you had a great idea, and I didn't like it? Is that a reason for claiming stuff about my math skill? What do you think you will achieve with that? I'm still confident in both my math and programming skills, and you saying things about it without even looking certainly won't change anything about that.
-
@alkexr whoa dude calm down its just an internet convo. No reason to be so mad lool
seriously though, maybe it's because English isn't your first language that made you make all those confusing posts but I think the topic is just too technical so maybe wait for more posts before trying to reply
-
@Schulz said in Improving AI:
Triplea would have enogh active player if AI was more challenging. Decent AI is must for gaining enough players for almost all games (maybe except sandbox ones).
First, sorry for hijacking your thread...
You might have a point here. But then we are trapped in the cycle of not having enough players, because not having good AI, because not having enough players, because... you see where this goes.
If you ask me, getting more active players would actually be easier than significantly improving AI. Most of the TripleA playerbase is still former A&A players (as far as I can tell), while most new maps aren't related to WW2 and don't appeal to this part of the playerbase. I think players who would be interested in these alternative maps just haven't found TripleA yet. But then, this woulnd't help with getting more players for the most played maps.
"For the world is changing: I feel it in the water, I feel it in the earth, and I smell it in the air."
-
@Captain-Crunch Maybe I should use shorter sentences and less of all that weird convoluted grammar we're being taught here in English classes, shouldn't I?
-
@alkexr so just to be clear ... both alkexr and Schulz are calling this AI bad when I expressly stated from the very start that this AI is very very very tough and very very much improved!
It could just use my "Learning AI" idea to be even better!
-
@Captain-Crunch Did I call this AI bad?
-
@alkexr said in Improving AI:
But then we are trapped in the cycle of not having enough players, because not having good AI
this dude is classic
-
Outside of the intricacies of the debate... I think the one real question becomes how feasible is any of this with only 1 guy working on the AI at his leisure. The leaps and bounds that have already been achieved seem mind boggling to me for a single guy trying to patch together improvements on his own in his spare time.
Any grandiose plans we might have would likely need more resources to be achievable.
"A joyous heart sours with the burden of expectation"
Hepster -
@Captain-Crunch "But then we are trapped in the cycle of not having enough players, because not having good AI" - I was following up on his point.
-
Were all of my posts in this thread so bad as to deserve a downvote, whoever it was?
"For the world is changing: I feel it in the water, I feel it in the earth, and I smell it in the air."
-
@Hepps ya I know I can't believe how tough this AI has gotten from just 1 dev working on it (redrum) despite the criticism from these other 2 lol
But I already said I would wait for more of the priorities for this game to be finished and fixed before I really ask for someone to look into my very easy idea on how to improve the Ai by "learning" from an implemented "WinnersCapturedTerritories" saved file that would help the AI better decide where to attack
A perfect example would be: on WWIIClassic map ...
let's say the AI wants to attack Western Europe or Southern Europe BUT THE TUV and Battle Calculator show that both battles are equal in gain and cost ... so perhaps the AI would use the "WinnersCapturedTerritories" file and see that past winners more often captured Southern Europe over Western Europe and thus you have data that helped the AI choose its target and should therefore be a better decision (the majority of the time)
Hello! It looks like you're interested in this conversation, but you don't have an account yet.
Getting fed up of having to scroll through the same posts each visit? When you register for an account, you'll always come back to exactly where you were before, and choose to be notified of new replies (either via email, or push notification). You'll also be able to save bookmarks and upvote posts to show your appreciation to other community members.
With your input, this post could be even better 💗
Register Login