New experimental version

[Seravy]

Honestly, I'd rather the AI kill my outpost. It will be destroyed, rather than spend 2-5 turns getting to a pop 20 27 building city. If you can teach the AI how to do turns, that's finw, we can change it then. But in 5 turns, I can make that pop 20 city so well defended, the doomstack can't even attack it as a viable target.

It's not that simple, since we are talking about a LARGE distance here that takes 5 turns, you won't know which city it is targeting, as you generally have more than one in an area. And the stack can reroute to your other large cities if one gets too protected.

I have to weight two risks against each other here.

Risk 1 : the stack gets torn apart when fighting for the small city, so it fails to do any real damage later
Risk 2 : the stack gets turned back by a large defending force before they reach the "big" target.

Now, in case 1 the AI failed to do significant damage (No an outpost is not a significant damage and you know perfectly well you can lead your doomstack to the enemy fortress without any risk of losing it on the way so summoning closer to them is just for convenience, it's not a major factor.) and lost the doomstack (or at least part of it - it's out of action for a while and has to regroup, or will attack things while having fewer units and get obliterated soon enogh. )
In case two the AI also failed to do direct damage but still has the doomstack AND damaged your economy indirectly - you had to rush buy units you weren't planning, you have to move stacks into garrisons that you wanted to use elsewhere and what's worst : the doomstack is still there so not only do you need to keep those units in your city to protect it BUT you also need to fill all nearby cities with units otherwise those get attacked. You don't only have one big city in an area in most cases, you have several. Also the stack is now in your main area and will probably steal some nodes and lairs from you there - you can't garrison those.

I think it's obvious that risk 1 is greater and hurts the AI more, while risk 2 is actually helping it indirectly. It's like chess - if you can prevent the enemy from moving 3 figures by keeping one in a key location, that's better than if you destroy a worthless pawn but move it away.

Now, if the AI gets lucky and neither risk comes true, in case 1 it merely gets an outpost, in case 2 a big city. It's obvious which is better.

Also, I don't know if its there, but nodes should have a minimum garrison size of 1 (you were talking about increasing it to 3, I don't think 3 is necessary.)

No, it's currently 3, I considered increasing it to 4 or 5. It used to be 5 for a very long time but was decreased to 3 eventually. This doesn't apply to units participating in doomstack like usual - if that conquers the node, nothing is left behind to garrison it.

[Seravy]

1/2x?

It's just 1/2*(1/x), same thing, the relative priorities of the distances are unchanged.

Ah so the priority happens on every turn?

Yes, every nonsettler unit forgets their destination and picks a new target every turn, which usually is the same one but sometimes not.

[Nelphine]

Right, and what arnuz and I are saying I'd that in most cases, destroying hat outpost is a good move. If the human is still settling,
then that outpost is IMPORTANT. And just like you've said, you know perfectly well, it usually isn't an outpost. And a small city is amazing for the AI, because it can rush buy buildings, and use cheating growth, to grow it into a big city, and most importantly - IT can summon things into that city, and end up with 2 doomstacks both close to that big human city instead of 1.

And remember Garrison bouncing and mobile defensive stacks? Yes if the AI gives me 5 turns, I'll have 1 stack that can defend all 4 or 5 cities the ai doonstack is threatening. So your comparison to chess pawns is wrong. Instead the cases are this:

Case 1: the AI goes after the big far city. The human has enough time to put together a counter doomstack, defends the big city, and has enough to either destroy the AI doonstack out right, or at least defend both the big city and the outpost. This is the most common outcome in my experience.

Case 2: AI goes after big city. Nothing the human has can ever to hope to stop it, so the AI doomstack takes the big city, then spends more time going back to tke the outpost (and it might take a few turns summoning replacements). This is the most common outcome (when the AI successfully takes any cities at all) in my experience.

Case 3: the human has nothing that can stop the doomstack. The AI takes the outpost then moves on (even if it needs to spend a turn summoning replacements) and takes the big city. This case is flat out better than case 2 due to not wasting time going back and forth.

Case 4: the human can scramble to stop the AI in a few turns. The AI attacks the outpost immediately and wins, then moves toward the big city (possibly summoning reinforcements if the outpost is actually a pop 1+ city.) Then the human doomsrack gets there and the AI cannot take the big city and maybe even dies. This case is still better than case 1.

Case 5: Your case. AI takes big city, human scrambles, and prevents the AI from doing anything more, but cannot take back the big city. This case DOES favor taking the big city first, but this case is FAR less common than case 1 or 4.

Case 6: the AI takes the outpost but the human scrambles and stops the doomstsck from doing more, yet the human cannot retake the outpost. This is worse thanncsse 5, but again, occurs far less often than case 1 and 4.

Therefore, the only time its preferable to take the big city first is if the AI can win either fight, cannot continue, but also will not lose whatever city it takes from the human. This is simply too rare a situation to be worth going after the big city first.

[Seravy]

First, the small city has a much better chance to be one taken from an AI instead of one being built by the human. So it has a low chance of being important, at least I don't build my outposts in enemy territory where I can't defend them. I build them on the best place around my capital. Even if I do build something far away, if it's not in a place where I can defend it, I won't consider it relevant.

If I want to summon units near the AI's fortress, I conquer one of their small cities because the AI is guaranteed to have some and they are guaranteed to be poorly defended - it takes well over 20 turns even for the AI to turn a settler into a city with 9 halberdiers garrisoning it. (and if they do there is the next one, they keep building more)

Next, the 5 turns is unrealistic. The posted save file allowed the doomstack to reach in 3 turns out of which the last one you can't use to increase your defenses - produced units aren't ready til next turn. And that city was quite far away : 8 tiles. I don't think the AI will pick a city 15 tiles away vs one only 3 away, to get the 5x multiplier you'd need to have 25 total buildpop to outweight the base priority alone and then the small city might have more than just 0 pop. And that also assumes there are no other similar value targets in less than 15 distance, unlikely. By the time there is a 25 buildpop city around, the map will be filled with cities almost everywhere. There will be like a dozen of them to pick from in that area - 15 distance is a 30x30 circle, which is almost a quarter of the world.

You're also assuming the two cities belong to the same people - there is not guarantee for that. In your scenario either the big, or the small city might be someone else's not yours. Fortunately at least being at war with both is now guaranteed.

You also ignore that the AI gets to build a second doomstack while the first is heading towards the target - and that will likely attack the small city immediately, probably not waiting until 9 units but already going for it at 3-5 - that is if a land stack haven't already done that.  Or, at the very least, the doomstack procedure is not being idle while it happens and prepares a second doomstack so the time is not wasted at all.

Case 3: the human has nothing that can stop the doomstack. The AI takes the outpost then moves on (even if it needs to spend a turn summoning replacements) and takes the big city. This case is flat out better than case 2 due to not wasting time going back and forth.

If it's a doomstack and if you have any units nearby (for example those you moved outside the small city because they can't defend it), the doomstack will leave the small city empty. You can reclaim it next turn forcing the doomstack to turn back and attack it again. You can keep doing that until the AI eventually picks "Raze" and destroys it completely, and no real damage is done to the city.
If this happens to the big city, it'll be reduced to nothing eventually and massive damage is being done, and eventually the city is too small and the AI moves on and hits the next big city it can find. That's what we wanted a doomstack to do and doing it to an already small city just locks the doomstack in place, achieving nothing.

Case 1, while bad for the AI, rewards good gameplay so it's good for the game overall. Stopping a doomstack is not trivial.
Case 2 is unlikely, the human will probably get their now undefended city back, turning the doomstack back towards it, escalating the damage, or it'll head toward the players other big city as it's now closer to the heart of their empire.
Case 4 is definitely better than case 1, I have nothing to add here.
Case 5, Agreed albeit I often end up with this so I'm not sure it's still more often than case 1. The AI is quite careful and not likely to mindlessly rush into 9 adamant longbowmen with a stack of nagas anymore.
Case 6 seems identical to case 5 to me.

I think our greatest disagreement is on case 2 and 3. If we want a doomstack to act doomstack-like then it has to go for the most valuable target first otherwise it'll get stuck pounding on a pop 1 city with zero value instead of destroying the entire empire until they are all pop 1 cities.

(note that this is generic attack targeting for any stack, not just doomstacks. But land bound stacks are limited to one stack per target so they'll attack both the far and near city assuming the AI has at least 2 stacks which they usually do. So doomstacks are the important case here.)

Edit : Case 3 is actually a fairly big argument against having any base priority at all - that prevents the AI from hitting the units that can reclaim the empty city after the doomstack moves away, as long as that stack has below 100 total military value. But fortunately, if it goes after the big city first, then hitting the units is unnecessary - reclaiming the big city causes even more damage to it so the AI will no longer care about staying there and can hit another one to destroy that as well as now the big city is a small city instead.

[Nelphine]

Ok, so I think what you're saying is that you're concerned that if a doomstack is going to target small cities that are adjacent, then it will end up in a vicious cycle where a) it takes the small city b) it then leaves the small city and heads towards the large city c) the human takes the small city back since it was left empty d) the ai is still so close to the small city that it turns around instead of going to the big city.


I think this is a valid concern. However, note that the argument I'm personally trying to make is that if the AI can attack something THIS TURN. Which means, usually, it's going to have to be adjacent (otherwise, since it doesn't know what kind of movement its own stack has, it doesn't know if it can reach it this turn.) Also note that the current distance formula you are using accounts for this. (If they're 2 squares from the small city and 4 cities from the big city, they're going to go after the big city. It's only when the doomstack is at least 3 times closer to the smaller city than the big city that your concern comes into play.)

Which means, in step b, the AI had to go only one square with its doom stack. IF that is the case, then it's perfectly reasonable to assume its going to take the doomstack a minimum of 3 more turns to reach the big city. 3 turns is a really long time. Therefore, I think it's actually reasonable for the AI to turn back and take the small city.

In this case, the problem is actually that doomstacks can leave the city empty.


Therefore, yes, I believe my reasoning is still sound. If your concern about ping ponging the AI is that high, then I think we have to rexamine the rules of doomstacks leaving cities/nodes completely empty. Yes I'm aware this is a whole can of worms that we've discussed quite a lot, but that's the source of your problem in the scenario described - not the targetting priority itself.

And super idealistically, I'd want to solve it by having the summoning circle go to that city (which coincidentally, actually takes advantage of the fact that doomstacks are stuck in a city for a turn; that gives the AI time to summon in reinforcements to help prevent leaving the city empty.) Something like 'after conquering any city, immediately cast summoning circle on that city' even though that's not during the normal spell phase.. no you can't do that since it will already be midcast, unless you could somehow recall what it was casting previous to that. Which is almost certainly not possible. So, while idealistic, it's probably completely unrealistic.

[Arnuz]

Heated debate <3

Personally I was more peeved from the war target vs neutral target thing. I am holding off commenting on the distance until I test a bit. Just one thought before I forget, that occurred to me night time:

make distance weigh more

Is there a formula that works better than (1/X)? And it better be not any more complicated to calculate than that, in intercontinental we have space but in land attacks we don't.

1/2x?
[/quote]

Or also 1/x^2, 1/x^3/2, or 1/x*log x depending on what we want to achieve if the distance factor shouldn't be linear.

[Arnuz]

I've tried RC5 on the savegame n.3 from yesterday and now yellow's nagas do a perfect purple's fortress strike :D

The fortress has only sprites... purple gets defeated and there's a lot of empty neutral cities all around, funny situation. So much for nature being strong at the beginning though. The two fortresses are at distance 12, and the doom-stack started around half way between the two.

War surely has bigger consequences now. On the other hand... I don't know, might we be opening yet another can of worms with this change? Is it less fun if the game plays itself? I know - I'm the one that advocated this. Still, I'm a bit concerned. It might be that the issues stemming from the no need for scouting for the AIs are magnified by this set of priorities. Argh...

[Arnuz]

Is there any way to fiddle with the parameters? Would you be able to make them loaded from file?

This is gonna be interesting:

[Seravy]

if the AI can attack something THIS TURN.

Then you are arguing for something impossible as we have already discussed - the AI can't know how many turns it takes to reach something.
1 distance vs 3 distance and 10 distance vs 30 are the same thing for the AI.

which coincidentally, actually takes advantage of the fact that doomstacks are stuck in a city for a turn;

Not anymore, I fixed that yesterday.

having the summoning circle go to that city

The AI has no idea what a doomstack is outside the doomstack procedure so it doesn't know which is the "doomstack city".

If it's about an exception that distance 1 targets should have more priority than what the formula yields, I find that acceptable but we I don't think it's necessary. We should try this version first and only change it if it does not work well.

[Seravy]

Orcs hate Elves...but this much? O_o