/sci/ - Science & Math

>>7644113
If you have a defined ideal, with no regard for internal ambivalence, then it is always possible to do something perfectly.

>>7644113
I saw this on sale on steam, did expect so many posts about it

>>7644113
Answer from an actual game programmer. Look at it this way AoE2 runs at 30 frames per second. So it's actually a turn based game with a finite grid. So you could solve it for each turn/frame in the same way you would solve chess. You could give all the variables and relation between all variables a cost. Then recursively you could iterate between all possible outcomes of the turn. Then choose the best outcome. For that frame and calculate this for all upcoming frames. It would be computationally very heavy but not impossible.

There is. People playing it are, in their little, using the most effective combinations and whatsoever. There are a lot of variables: every step would be taken into consideration if you want the absolute perfection, also it would be different map by map. Possible, but you won't see it coming.
Maybe with AoE1 it could be a thing.

>>7644132
That would be an alright heuristic, but wouldn't solve for the best game, and you couldn't necessarily have a cheap way to evaluate what is a "good" move.
Some turns it might be best to do nothing. In the long run, what seemed like a bad move might be good.

You'd have to calculate out all possibilities and see which one ends first, or loses the least pieces. Near impossible with present technology.

>>7644113
Starcraft + Korea = Perfect game

>>7644146
No one said anything about heuristics,
>recursively you could iterate between all possible outcomes of the turn
it's bruteforce. Yes it's not possible and won't be for very, very long.

I wonder what a perfect player would build.
Composed armies?
Built to shit armies?

It would have to be perfect in a vacuum to where it would win every single time with the same strategy. That is decided before the game even starts.

>>7644158
a perfect player would probably scout the enemy base ASAP and keep the scout gathering information about the enemy

>>7644158
The perfect player is executing perfect Dark rushes while walling enemies from their resources.

>>7644157
Guess I don't know what recursive means.

I'm becoming more braindead every day. Feels good.

>>7644167
I dunno, this is a pretty big assertion to make. Are dark rushes objectively good? Picture two of these perfect players, maybe there's a way to perfectly defend from a dark rush and make the other dude waste his resources for nothing and destroy him in feudal.

>>7644118
>implying there is always internal ambivalence

>>7644188
That was not implied.

>>7644189
then, your definition of perfect refers to relative evaluation of perfect, OP's however refers to perfect in the sense of optimal

>>7644195
All evaluations of perfect are relative to the system generating it.

Not everyone wants to win the same way.
Some people might want a win to contain certain elements.
Some people might not want to win at all, and are searching for the inverse; the absolute worst decisions possible to be made. Despite inherently being dualistic, it then becomes perfection.

>>7644195
No, you just want to win.

>>7644186
Yes, dark rushes are good. We're talking a perfect player here. It slows down the progress of your enemy, while allowing you to create your own economy while the enemy is dealing with your constant stream of militia and random structures blocking them from their resources. It's pretty obvious you've never played real Age of Empires 2 friend.

>>7644224
Everyone knows what dark rushes are supposed to do, the point is whether a perfect player can easily defend from it and make the attacker lose more resources than he does.

You're too focused on the attacker being perfect, you're not considering the defender to be perfect as well. Home advantage matters, and even more for a perfect player.

For a game to be solved, and there being a 100% certainty that someone is going to win. Things like first plays come into effect the most.
Chess isn't a solved game, but if there was a perfect way to play it. White would be the winner. Because it gets the 1st move advantage.

In Age of Empires there is no advantage to this. You start at perfect level ground. Now, building a strategy that is frame perfect and extremely efficient, you have one hell of an opponent. But on perfectly level terrain with the same resources given to both the game cannot be solved.
There is no natural advantage to being on either side.

>>7644261
>first player advantage

this is terribly wrong in general, many kinds of fair-ish games actually favor the second player

>>7644263

The point still stands. The only thing that claims advantage over the other is when the computer registers hits.
If two perfect players dark age rush each other. With the same amount of militia, villagers, whatever. Since they're both perfect they're the same.
The only advantage one has over the other is when the first attack is registered or not.

>>7644113
23 civs in random match-ups (about to get more):
529 possibilities.

So first, pick a civilization. This will narrow it to 23 possibilities.
You'll still need a hell of a lot of practice if you want to get proper optimized builds for each match-ups.
So I just do whatever, since there's no way I'm gonna remember all that stuff.

Starcraft 1 had the most developed strategies as far as I know (as well as only 3 races) and over time they started to converge to an equilibrium where 90% of games involved both players following the same standardized strategy.

>>7644336
Competetive Age of Empires 2 is mostly dark/feudal rushing and if there is water, galley fights.

>>7644132

Maps are randomly generated and there is fog of war requiring exploring to find resources.

That amount of randomness can never be accounted for.

I forget what name they'd given it, but there used to be competitions between coders who'd make custom AI's and then make them battle each other. Quite interesting.

The fact that you can't see the opponent means that the game can't be solved. Even a minimax type algorithm to play it wouldn't work if you can't anticipate the opponent.

TLDR if there was no fog of war and you could see the whole map it would be possible

RTS games don't have perfect information so there could never be one strategy that dominates each time.

If you knew your opponent was following a certain strategy, you could always find a superior strategy in some way. For example, doing the exact same build as the opponent but producing one less soldier in favor of one more worker/harvester and playing relatively defensively. You would gain an economic advantage while having no military disadvantage due to reinforcement time.

However, there could be a mixed strategy equilibrium where players are using a set of strategies with a fixed percent probability of using each strategy. i.e. each time you would randomize builds using fixed odds. This mixed strategy equilibrium would be the "perfect" way to play a game in that deviating from the set of strategies (and probabilities of using each strategy) would lower your chance of winning, but there is no one strategy that would let you win 100% of the time.

If we assume that there are a finite number of strategies (which we can get just by assuming the number of player actions per minute is limited), then we know that such a mixed strategy equilibrium DOES exist by Nash's famous proof (see http://www.cs.ubc.ca/~jiang/papers/NashReport.pdf).). That doesn't mean we know what the equilibrium is though, chess also has one or more equilibria but they are unknown.

>>7644132
There are cases where backwards induction can't find any equilibrium, even when we know that it exists.

>>7644477
>Maps are randomly generated and there is fog of war requiring exploring to find resources.
>That amount of randomness can never be accounted for.
True, but, the other player has to account for it just as much as you do; ergo it cancels out.

>>7644769
No it doesn't. The algorithm is simply worthless with so many unknowns.

You'd need a big fucking neural net to achieve this with all of the unknowns.

Just look at pro players like daut and viper

Aoe is not a puzzle, it is about speed.

>>7644772
>No it doesn't. The algorithm is simply worthless with so many unknowns.
People play AoE all the time, and humans definitely seem to develop a certain level of skill at the game, meaning it's in theory possible.

>>7644784
Yes, but it would need machine learning. An algorithm couldn't do it realistically.

>>7644769
the "randomness" isn't so random, since it uses peudo random generated numbers; so.. based on that...
or just set the knwledge of the map == 0 and start with exploring and making decisions from there on; maybe having a machine learning base behind the decision making

>>7644113
No one in the AI community cares about age of empires 2. There was a starcraft AI that did pretty good though:
http://overmind.cs.berkeley.edu/

It wasn't that interesting in terms of the AI approaches they used.


>>7644772
well if you aren't an idiot then you'd give each unit a simple neural net and train them using difference rewards. Classic agent based shit.

Can't build TCs until castle age seriously limits the game IMO.
Fast expand in SC is a big strategical part of the game.