AI overland stack/unit movement strategy

[Nelphine]

I'd actually make the minimum movement speed for the doomstack 3. Speed is very important for doomstacks.

[Seravy]

Not muchmoves 3 until the midgame so that's not an option.
I'm not worried about this, as soon as the AI has enough of the strong, fast rare creatures, slow crap won't get a chance to be included.

[Seravy]

   
First new type stack that actually reached me in a game. 1406 March on hard. Either my longbowmen in my capital can stop that or I have a problem. This thing doesn't stay in conquered cities, unless there is a bug...and 4 more turns until my first Efreet is ready.
...well, they went directly for the capital. And it held. 8 longbowmen are not just for show. An unprepared player would have been knocked out though.

[coldsteel]

It is amazing to witness such marvels

[Seravy]

Not sure if it's caused by the new procedure or it's just pure coincidence, but I feel the incoming lizardmen stacks are better organized than usual. Stacks of 6 units appear out of nowhere periodically and I'm losing cities one at a time because they didn't have a strong garrison - a single gargoyle can't fight 6 lizardmen units. I have to stop attacking the High elf player (green) for a while and focus on the lizardmen because they are a threat.

I think the graph speaks for itself :

   

That comes from the 4 cities I lost so far. It seems the human player can no longer get away with the usual "I have 1-2 units everywhere" tactic on hard - if there is a lizardmen player.
Haven't seen any such stack from the green wizard - elves don't move on water, and he didn't summon enough sprites to build a stack from those it seems. Hard AI casting bonus is not high enough to have 9 spare sprites while the wizard also knows War Bears - or maybe they prioritize buffing too high.

...but there is more. The AI seems to be able to get the amount of units they need to attack my longbowmen right. They send 2 lizardmen javelineers against 2 longbowmen instead of the usual 1 - and two can actually win the battle, as the AI gets a turn to summon centaurs.

Meanwhile the Life wizard who only has their capital left declared war because they had more military power then me instead of being scared of my superior forces. How, you wonder? Well, sometimes these things appear now..

   

My Efreets can deal with them but without that, I'd be in big trouble.

[Seravy]

One odd behavior I'm not sure we want to keep or change.

When the doomstack conquers a city, next turn it leaves it, leaving the conquered city completely or mostly undefended.

PRO
-The doomstack keeps moving and conquers more and more cities if there is no interruption from the player.
-Allows the AI to keep attacking at a fast pace.
-The doomstack is still within range - if the player recaptures the city immediately but uses insufficient troops to hold it against it, the doomstack might turn back an annihilates them, causing the city to lose buildings three times total which is a lot of damage - but only if it's a better target than the player's other cities.
-If the doomstack is top tier (great drakes and the like) the raze decision will have a 100% chance to raze the city anyway, so this only applies to rare or below doomstacks.
-If the doomstack is recapturing, it will be "busy" for the turn, allowing the AI to move the next 9 best units towards building a new doomstack - tho if they have any other target to attack, the same happens anyway.

CON

-Leaving then recapturing the city multiple times gives the player a lot of opportunity to hurt the doomstack with spells, while they only lose one city - albeit that one gets wrecked badly, it's sill better than the stack going through all of them
-The AI damages the human player's cities but it's easy to take them back later once they leave the area
-If the doomstack doesn't have anything to attack (unlikely, unless it's weakened badly), it'll be oscillating between leaving the city, and going back to it to garrison until new garrison units are produced.
-If the doomstack was killed, or is busy heading towards a target, and the next 9 best units the AI has are all garrisoning the same other city, that one will be left undefended as the units leave to act as the next doomstack.

[Seravy]

There is a lizardmen opponent again in the early game.
While their offense is good, and I've lost a few cities here and there, I still gained territory overall and am in the position to start invading their main continent.

At which point the drawbacks of the new algorithm start to surface :
-The AI controls a Sorcery node but has no units protecting it. Nothing at all, it's regularly melded by other wizards. Lizardmen units are all intercontinental so they all get used to build stacks except the swordsmen and spearmen, which never leave their cities so don't end up getting placed on nodes. If they had better luck and used some war bears to conquer the node, those could have garrisoned it but it didn't happen. (or they got buffed by water walking and left - I've seen ww bears attacking a few times.)
-Likewise the AI's cities have no javelineers at all (more expensive unit) and only a few halberdiers (less expensive unit but still valid for the doomstack)
-The "doom" stacks aren't all that much of a threat anymore. While they seem to manage to build up to a fairly large size (especially now that my cities have larger garrisons), my longbowmen+gargoyle+city wall combination is pretty much invincible to stacks of javelineers, sprites and halberdiers. It was dangerous before I had my longbowmen ready though - lost about 3 cities that way.

Overall, I think the tactic is performing fairly well, but does leave lizardmen (and draconian) players somewhat poorly defended (in exchange for being much more aggressive and powerful in offense) - at least on hard or lower difficulty where the AI doesn't have an endless supply of new units in the early game.
The problem is, if any minimal garrison requirements are defined and observed by doomstack building, that will break apart the doomstack as soon as it conquers a city instead of letting it continue.

[zitro1987]

So far, I like the new algorithm (not playing a very high difficulty). Even outside doomstacks, a doombat fairly early in game came out of nowhere and attacked me.

Offensive AI is relatively more entertaining than slowly defeating town after town. While towns appear somewhat easier to conquer, sometimes i have to do it for my own towns.
*Inspires me to cast 'cloud of shadows' and possibly pick 'guardian' retort in future.

[Seravy]

While I believe there is no problem with other races - and yes, the game is a lot more fun to play now, at least on Hard -, I think we need to do something about lizardmen (and draconian) throwing all their non-swordsmen units into the war instead of leaving better garrisons.

A few possible ideas :

-Implement a restriction based on unit type. For example, Javelineers and Draconian Halberdiers can't participate in doomstacks, leaving those for garrison duty. Unfortunately, as there are only very few units in these races, this reduces the versatility and diversity of doomstacks a lot.
-Implement a restriction based on unit experience. For example, lizarmen units that have at least 50 EXP can be used for doomstacks but below cannot- Having only a barracks, this enforces the units to stay as garrison for at least 20 turns before they can move out, ensuring an acceptable supply of defenders. The problem with this one is, it breaks if Altar of Battle is cast, and even if there is a unit with less EXP that is "free" and not in a city (such as the 10th unit produced), it can't participate in the doomstack. On the other hand, the doomstack is guaranteed to have at least a level on each unit which is not bad.
-Implement a required garrison system that forces only lizardmen and draconian normal units to stay behind in cities/nodes. This way doomstacks won't leave the freshly conquered city empty, and other cities will have defenders, plus other, more valuable units can stay on the offense, but in exchange the wizard has to build a new doomstack every time it conquers a city using lizard/draconian normal troops. Assuming their main cities are of the race, there is plenty of availability to do so, but it'll take time - so offense becomes a lot slower, but the conquered cities are more likely to hold.
-Do nothing and hope the AI will have at least one common/uncommon land based summon that stays in cities, or produces so many units that the doomstack(s) can't pull out all the units from the garrison. (more likely true for impossible only)
-Do nothing - one race per plane having a unique fighting style make the game more interesting, and weak garrisons might not be relevant if the offense keeps the players busy on self-defense. Questionable, lizardmen have been over-performing under AI control in (older) tests, so if the offensive strategy is better, it might make that problem even worse - on the other hand if the weaker garrisons make them somewhat vulnerable, it can balance out their above average performance.

[Nelphine]

I think I am confused. When the human player makes a doomstack, you make A doomstack. (Yes you can get more later, but generally you make a single one that is amazing, then while it rampages, you build defenses, until you are finished, then you start building more doomstacks.)

Therefore, if the AI is only building one doomstack, they shouldn't have any Garrison issues. If they're building more than one stack, they are NOT building doomstacks. They are simply building offensive stacks.

If you are actually talking about doomstacks, there is a fundamental problem with what you are building, or you wouldn't have these problems.

If you are simply building offensive stacks, then you should have a garrison requirement (say 4 units), and then offensive stack creation, then further Garrison building.

Offensive stacks SHOULD be broken up when a new city is conquered. A DOOMSTACK should not. These are very different concepts.