(BTW seravy, yes, I'm saying an outpost that you can summon to, is literally as important as a population 7 city with the buildings I listed, that somehow was not ever allowed to have a summoning circle in it no matter who owned it. Obviously no city is 'immune to summoning circle', I'm just trying to make my point clear.)
Quote: No no. The base priority is due to summoning potential. Summoning is as good as having those 7 buildings.
Wait what? As far as I know cities contribute nothing to your summoning capacity, aside from their Amplifying Tower.
If you mean the location that's not a resource and we can't raise the base priority because "hey this city might be closer to the capital than the other one". What if it's the other way around? Imagine the Yellow capital is next to that big neutral city and not the small purple one. Would raising base priority help or hurt then?
btw intercontinental currently prioritizes nodes and lairs equal to the strength of the forces guarding it so it's pretty much equal priority to random units. Land has different priority so this is also inconsistent. Maybe land attacks should also prioritize nodes and lairs equal to their military value?
Yes land attacks should prioritize lairs and nodes based on Garrison strength. That explains random ones remaining so long
And, yes arnuz and I are both saying cities base priority SHOULD be based assuming its in a good location. That is exactly what we want. We WANT the AI taking out cities close to the stack in most situations. And only when a city is supremely developed should it have even a chance of attracting the AI away from a closer city.
For humans, we can judge if its worth diving deep into the enemy territory to attack a given city. The AI has no way to really judge that, and arnuz and I both see (in actual play) the AI skipping close cities to attack a farther city far too often. Conversely the number if times I think 'why didn't the AI go for that distant city' I can count on one hand. It virtually never happens.
The AI is all about size. It wins by being huge and summoning tons of things. It has such high resource bonuses it can buy all the buildings it needs - so its priority is to get more cities, not necessarily get better cities.
I agree that the AI should prefer closer targets, which is why there is a base priority. What I don't agree with is the amount. Assuming a distance modifier of 3x (1 tile instead of 3, this is the smallest it can get, and honestly a distance difference below this is not very relevant unless we are talking 10 tiles vs 30 but then neither will get picked in 99% of the cases) a "100" base priority with 10 buildpop means the 0 pop outpost has 100 priority, while the 8 pop, 10 buildings, developed city has 280, not enough to outprioritize it, and we are looking at the smallest possible distance here. If it was 6 tiles vs 1 tile, a 20 pop, 27 buildings city would still fail to get picked because 570<6*100. I think this would push things a bit too much in favor of near cities - they'd almost always get picked, no matter what. And that can actually hurt as much as help - if the AI loses their big city and the doomstack is further away, it won't care to reclaim it and would keep demolishing small outposts nearby.
Also, what I suspected turned out to be true, the AI does leave units that just finished attacking something marked as "attacking" so they can't participate in doomstack action for a turn (tho they might be able to participate in other forms of attacks which do not care about it).
Anyway.
I've uploaded the current state of the game, including all changes discussed above and in the previous days. The file is Caster5RC.zip because this is not really a release more like a tool for last minute testing before one. Feel free to use it in the small amount of time we have left before the main release and tell how it works now.
This is the full list of changes
-Fixed : maintenance cost of Seismic Mastery and Fairy Ring were swapped.
-Fixed : Units with Life Steal were not valid AI targets for Blood Lust.
-AI can now target units that have a touch attack even if they aren't a priority choice for Focus Magic
-The AI is now not allowed to cast Focus Magic before turn 10 - it's too likely to find no targets and waste the spell, as well as significantly slowing down early expansion (80 can be used to make two hell hounds or a ghoul instead!)
-The AI is now allowed to cast Earthquake when Seimic Mastery is on if they don't have an expensive summoning spell available to trigger it.
-Fixed EXP14C bug : intercontinental attacks read the enemy land power from the matrix incorrectly, resulting in failing to attack targets sometimes.
-Fixed EXP14C bug : intercontinental attacks read the enemy strength of nodes/lairs from the matrix incorrectly.
-Fixed EXP14 bug : intercontinental stacks with a power divisible by 256 couldn't do intercontinental attacks.
-Each unit has a minimal AI rating of 1. This allows the AI to attack units with zero attack strength such as settlers or floating islands.
-Fixed EXP13 bug : Dispelling Wave still removes combat global enchantments even though it's not supposed to.
-Fixed bug : AI fails to leave city walls if enemy obtains ranged superiority during combat instead of starting with it.
-When the AI calculates the ranged capacity of armies for wall decisions, it'll now use (1+figures)*attack*ammo instead of figures*attack*ammo - this should let low figure units with high attack compare more fairly with high figure units.
-Adjusted the effect of mana spent in strategic combat for each realm.
-Fixed : Flame Blade and Mystic Surge weren't swapped for overland Dispelling Waves, causing the wrong name to show up.
-Fixed memory corruption bug that happens when banishing wizards and finding spells in the ruins (spell name was stored at a null address)
-Units that require both Stables and Smithy now only require the Stables because that already implies a Smithy is built.
-Hydras now have 1 less resistance.
-Great Drakes now have 1 less resistance.
-Demon Lords now have 1 less resistance.
-Colossus now has 2 less resistance.
-Behemoths now have 1 less resistance.
-Time Stop maintenance is now 30*turns instead of 100*turns
-The AI will now leave at least 7 units in the city when using Astral Gate instead of 6.
-Fixed diplomacy bug : When the AI offers a treaty, and wants to also offer gold, it fails to do so, or offers 0 gold sometimes.
-Fixed : The text displayed when casting Drain Power is wrong.
-Supreme Light now costs 6000 to research.
-Fixed minor bug : target war priority variable is not set for neutral targets on AI land attacks.
-AI intercontinental attacks will now ignore unit based priority and only use city based priority on city targets, similarly to land attacks.
-AI intercontinental attacks now have the same priority on city targets as land attacks
-AI land attacks no longer have a 1/3 multiplier to the priority of enemy stack targets outside the home continent.
-Doubled the granted priority of buildings/population in cities for both land and intercontinental attacks. Doubled the priority of fortresses (land attacks only).
-Halved the priority of neutral city targets compared to other war targets.
-Lairs now have priority equal to their unit value for land attacks to match intercontinental priorities. Note that while Node and Tower priorities are not changed, these also count as lairs as long as they are not cleared, so if their “lair priority” is higher, it'll be used instead of their other priority. So in this case they also match intercontinental.
-Fixed bug : AI doomstacks fail to move next turn after attacking a target.
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. Then the AI doomstack spends 3-4 more turns getting back to the outpost, by which point its grown into a city, I've bought city walls, and I've reinforced IT and now the doomstack fails to kill either. The AI, should NEVER choose a target more than 3 squares away if it's adjacent to a target that it can attack this turn.
(If both targets are AI, its not so bad - the targeted AI won't be able to drastically change it's own defenses in response to he attacking doomstack. But the human? If the ai can attack a human city this turn, it MUST do so. Yes, occasionally this will hurt the AI - far far more often it will help the AI. Until you can teach it the difference it needs to do the one that comes up more often.)
Reuploaded for one change which was still inconsistent between land and nonland attacks :
Quote:-AI intercontinental attacks now use the same extra priority for fortresses as land attacks
While I thought it's a minor thing, in reality it seems to be major - said yellow stack went immediately for the purple capital and defeated the wizard. Which, btw, is a far better move than trying to reclaim the pop 3 city :D
However...this might change the game a lot - both more eliminated AI in the early game, and...the human player will actually get murdered on turn 30 now if they don't have a full stack in their capital and anyone nearby has a doomstack. Which doesn't mean only Sorcery...I've seen 9 waterwalking war bears or 9 wraithform ghouls before...
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.)
The fortress attack problem is a symptom of cheating being too high to sustain for early games, not a problem with targetting. AI should be trying to do fortress spikes, which this could do.
I like the continent logic removal: it always sounded like a bit of a patch from the old programmers because they couldn't manage to implement functioning intercontinental doom-stacks.
Thanks for doing these changes! I like them! I think that now the cheating will become overpowering, as the targets will make much more strategic sense. I hope that we can be able to reduce it by the next RC, and I really REALLY hope that we can make it scale with time, or number of cities, or anyhow not be so front-loaded as now.
(October 5th, 2017, 14:13)Seravy Wrote: Checked twice, being hostile without a war does not have the 10x modifier of wars. Do note however that in your version intercontinental doesn't have any of these modifiers yet. I've already implemented that while we were discussing this.
[...]
Because we want neutrals to be preferred over a hostile but nonwar target. If both are equal it won't be preferred. However you are right and I already made neutrals use the war modifier and then get halved on top of it. So it's nonwar<neutral<war now.
Fantastic, thanks.
(October 5th, 2017, 14:13)Seravy Wrote: Even on the highest difficulty setting, building destruction rate is only 45%, and there the AI builds that much faster to get them back anyway. But whatever the destruction rate is, the small and the big city would lose the same percentage so their relative priority remains the same aside from the "base value".
As Nelphine pointed out, the AI building faster is a reason to consider buildings less important. That said, there's the damaging the opponent consideration, but that sounds like it should be a consideration more for war targets than for opponents to me.
(October 5th, 2017, 14:13)Seravy Wrote:
Quote:Idea: would it be easier to calculate city targets distance from the fortress than from each stack?
Land targeting is not calculating any distance at all when putting together the priority matrix. In the area where it adjusts this with the distance, there is no space for anything detailed.
Ah so the priority happens on every turn? I was thinking something more sparse, maybe calculating the priority modifier based on the distance from the fortress each time a city is created. Not for this release obv
(October 5th, 2017, 14:13)Seravy Wrote:
Quote: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.
[/quote]
1/2x?
And finally...
(October 5th, 2017, 14:13)Seravy Wrote:
Quote:* reduce initial cheating a lot (whatever means, it doesn't matter if it's skill or mana so long as the result is having a limiting factor, so just do the easiest for you)
* raise cheating based on number of cities
Not relevant to stack movement whatsoever and too large scale to do directly prior to a "stable" main release. If anything initial cheating is what allows the AI to not lose the game immediately, as early game is far too important as is. And I think our last conclusion was that's unfixable.
Agreed on not doing this hurriedly, but I hope that you will consider the point for the next release. As you know I am much more interested in the initial phase of the game, so I give some different input than other players. I don't mind the cheating per se, but right now it's so much front-loaded at the beginning that it makes anything you do matter much less. If the AI gets locked against some unbeatable obstacle you reach a point where you overcome the cheating and then you win, like the first example with an island locked Tauron full of hellhounds, otherwise you lose. I'd much rather have the cheating scale more gradually, and why not, even surpass the current amount later in the game eventually.
