AI defensive targetting overland

[Nelphine]

I wonder if number 5 is getting picked. Obviously not if the AI tries my own continent (hello tiny spearmen garrisons!) but if a stack of hell hounds moves to an another arcanus' wizard continent maybe the hell hounds find no targets? Doesn't seem very likely - especially since the ai should keep trying new continents till it finds mine. Then again, I also see this kind of congestion with lizardmen AI and they shouldn't even have the shipping problem.

But at least number 6 explains some cases of it when I let the last AI go crazy.


If this is really unexpected behavior, maybe I'll go watch for it in my next few games. I know my current game, the myrran AI (Sss'ra) had about 4 stacks near his fortress. But he also had an enemy AI city within 6 squares, as well as a neutral troll city within 5 squares that I took. (Pretty sad for 1408). I'll check the 2 arcanus wizards shortly.

[Nelphine]

@Arnuz: I don't you'll be able to get the AI to consider multiple stacks. If you go through the AI thread you'll see the decision process for stack targeting, and it's huge, and it literally doesn't know about the existence of other stacks it owns. You'd have to rewrite everything, and space is already a huge issue. So you can't make it bigger than it us, which I think multiplr stacks would require.

[Arnuz]

Mmmm... Agreed.Don't try to calculate the multiple stacks, too much work.

How about this instead. Add a step that considers the presence of other stacks around the stack currently being decided, without considering what's inside them. Add a randomisation step to the decision on whether to attack, that basically checks the previous decision, then overwrites it with an attack on a low percentage roll which depends on the number of surrounding stacks that could also reach the target. If there's one, then 3%, 2 6%, etc.

I'm coming from the consideration that the game is ruthlessly random, so throwing in even more randomness might well turn out to work

[Seravy]

Maybe a solution would be to consider garrisons as stacks and re-evaluate the target of them as well each turn?

I don't understand. This is exactly what is being done. Every turn the stack blocking the tower has 1000 strength and the AI's best own stack on the continent has 700 so it's not attacking.

The problem is not the AI blocking the tower for itself - they don't do that unless they have no way to leave the tower on the other side. The problem is Nelphine putting units into the tower so the AI can't attack it either due to army strength or due to an Alliance.

Now that I think of it, considering groups of stacks vs stacks, despite the sea limitation, would allow the AI to also go for stronger nodes than the strategic combat allows, simulating the users' trick of whittling the node down. Strategic seems to prevent that somewhat if I get it right as it minimises losses, this might need to be changed.

It doesn't work because it indeed minimizes losses and that's intentional - we don't want the AI to lose more troops against neutrals or other AI - and the units in nodes heal to max after each battle (this also cannot be fixed - the game simply doesn't store the health of units in lairs : it literally only stores "3 units of Sky Drakes" and nothing else.)

[Arnuz]

1 - good, I misunderstood then.

2 - maybe strategic combat might be different against nodes and lairs? That'd help the AI, by emulating the whittling down approach, in combination with the multi-stack randomisation. And make use of the capital superfluous stacks. Yes, I understand that this is starting to be a lot of work...

[Seravy]

2 - maybe strategic combat might be different against nodes and lairs? That'd help the AI, by emulating the whittling down approach, in combination with the multi-stack randomisation. And make use of the capital superfluous stacks. Yes, I understand that this is starting to be a lot of work...

Making neutrals suffer maximal losses while others suffer minimal is probably not hard to do.

Problem :
This assumes the AI will be allowed to send "too weak" stacks against neutral targets. Which means, it's allowed to intentionally sacrifice troops for a future gain.
The AI isn't smart enough to do this in a "safe" way like a human player.
It'll lose things such as :
-the doomstack, yes, the whole thing, regularly
-all of its heroes and their items, so they never level up and the treasure it earns will do nothing

-whatever else good troops it has and would need to fight an enemy

Furthermore, after weakening the node with 2-3 such stacks, the human player or another AI can get the now much weaker node using any nearby random forces.

Ultimately, the lairs would act as black holes eating every single AI unit until they are cleared.

[Nelphine]

To completely derail this whole conversation, what makes the AI 'forget' weak lairs or nodes on it home (or second) continent? Or like in my game, neutral cities near its capital? Presumably they tried to attack them early on and found them too powerful - but why don't they get cleaned up later? (Note, im specificaly not talking about strong lairs, like several death knights and half a disen wrairhs, that the AI just never gets strong enough to attack. Only ones that seem too weak.)

[Seravy]

I don't think the AI can "forget" anything. So that means it never had a stack with a larger strength on that continent, at least not during the turn(s) when continent evaluations were reset, or that it always found a higher priority target for those stacks. (war with another wizard, or better lairs)

...the other possibility is those targets were unreachable due to units or other nodes in the way.

[Nelphine]

Right but a nature node with 7 great lizards and 2 sprites, 2 squares from its capital, or a lair with few unicorns that isn't on a shoreline? Or a troll neutral with 5 shamans and 2 halberdiers, that only had one ocean square adjacent? These all seem very unlikely to have been continually too strong, and all of them had at least 7 squares around - and no lairs or nodes adjacent when I arrived.

But the higher priority targets is a possibility. Since I see this regularly, can targets on another continent be higher priority?

[Seravy]

7 Great Lizard? That's a very hard node.

The troll one sounds unlikely. The few unicorns as well. For these I have to assume blocked path or other targets. Although if it was a race that produces weak units, such as Dark Elves, it is possible. Best is if you observe if the AI really had a stack capable of attacking or not.

Since I see this regularly, can targets on another continent be higher priority?

No, they can't.

There is one more possibility - if the race was Draconian or Lizardmen, those units will leave towards the main selected continent before stacking up, so if the home continent wasn't selected (and it probably won't), then there won't be large stacks capable of attacking there, unless they accidentally move over the land tiles when moving from one place to another. The Doomstack is an exception, but it has to actually step on a land tile by chance to be able to target a lair on that continent - intercontinental targeting recognizes units and cities only.