H5 AI harder than H3 AI?

[Kristo]

Chess AI's use opening books and endgame scripts because it's faster, not because it's hard to program. Lots of really smart people have extensively studied the opening and closing moves of a chess game. For certain game states, there exists a provable absolute best next move (or at least they've narrowed it to a few choices). Why bother searching for the next move when the work has already been done?

AFAIK, no one has studied a Heroes battle to the point where you could script such behavior. What recourse do we have but to examine every possible move combination (to a reasonable search depth) and pick the best one? Basic chess AIs are not smart. They run a relatively simple recursive search algorithm to find a path through the game that leads (hopefully) to a win. If you give them enough memory and processing power, they'll play a pretty good game. And that's all we need. Why try to be smart when brute force is good enough?

[ThunderTitan]

Can you guys explain what you mean by "move patterns"?

The pieces move only in certain ways and thus are easier to model mathematically.

And you actually don't think being able to stand still, defend or move in any square in the move range add too many possibilities?! Maybe the actual algorithm would change all that much (debatable), but the possible outcomes get multiplied... http://en.wikipedia.org/wiki/Computer_c ... ving_chess

Plus, the biggest problem would be actually determining what moves are best, which would require ppl playing the game for a while... and who's gonna design an AI for a few years after launch?

[Kristo]

I agree the branching factor is much higher for a Heroes battle. But I don't think you have to search very many ply (turns of lookahead) to be effective. Most battles aren't fair fights anyway.

As for what moves are best, I'd naively order moves in this order:

1. Moves that cause damage, in decending order.
2. Don't move.
3. Moves that don't cause damage, in decreasing order of distance.

This ordering puts stupid moves like "walk next to a strong creature but don't attack" in the third group. Pruning optimizations (see Alpha-beta pruning or Negascout algorithm for example) would hopefully toss out moves like that before going too far because moves from the first group would likely have all produced a better score. If "don't move" is a good move, then you have fast creatures wait and slow creatures defend.

[ThunderTitan]

Well sorry, you're right, a decent (non-retarded) battle AI isn't that hard... i was thinking you mean one that can truly challenge a good player like the chess AI can...

[Vitirr]

If you give them enough memory and processing power, they'll play a pretty good game. And that's all we need. Why try to be smart when brute force is good enough?

The shame is, the average player doesn't have a Deep Blue machine available at home.

So no, I don't think only brute force is enough.

[Mr.Wonderwall]

No i don't have any difficulty in normal,it's very easy,because you start with a lot of money and resources,i can't understand why you face difficulties!

[lumpoor]

AI is better in H5. Still sucks, but in H3 they always let the melees charge as far as possible on the battleground, leaving them very vulnerable.

In H5 they've learned how to defend