Quote:
MaxWilson said:
3.) the strategic space is *huge*, with many different and interesting counters possible to any specific tactic, so even a fairly dumb hill-climbing search may stumble across a good counter to, say, E9N9 Niefelheim (e.g. race for Alt-6 and Iron Bane).
|
Dominions' complexity is what I'm worried about with this method. The tests they ran were relatively simple (see section 6.2 starting on p26), but yet the AI took on average 20-30 battles against unchanging opponents to beat a static AI (table VI, p38). Given the much larger number of categories of spells, I'm not sure the AI could learn to counter players' tactics.
Still, since it looks like a genetic algorithm, it's probably the best approach there is for Dom3. I think it would be worth trying for some simple cases to see how well it does against the current tactical AI. (If the game has the right hooks, of course...)