AI overland stack/unit movement strategy

[Seravy]

The AI is bad at this, we all know that.
If rare+ creatures are more accessible and relevant, the AI has to improve their ability to use them.

I think there are several large problems here (aside from using the wrong kind of unit which I can't do much about):

-These creatures are helpless while the wizard is banished - without spell support they are no better than a random node garrison. This will be fixed by changing banishment rules.

-They aren't careful at their attacks and get killed too easily - like entering a city full of ranged units alone and getting killed before even getting a turn. If I can do that, changing stack strength calculation to count only the defender's ranged power at double weight can help this. If invisible units were worth double on the attacker's side, even better (as those counter it). This should at least prevent the AI from throwing away these valuable units without a real fight.

-They aren't used in large enough stacks to matter. At the new reduced cost, the AI will now have enough units to be able to build doomstacks from them in theory - but in practice, that means we need a new procedure that makes them do that. I'm considering to make the procedure pick the 9 best idle (including nonfortress garrisons) intercontinental units, and pull them all towards the same spot unless at least 6 of them are already together - but I don't know where to insert this step. Before attack targeting means units will prioritize this even if they could attack an adjacent unguarded city or other target. After attack targeting means, well, it means it just plain does not work because the units are already taken by the "send to main action continent" procedure which affects all idle units. There is no place in-between, as they are done per-continent, and we need this to be done regardless of continents. Unless...I suppose there is a workaround. If the new procedure considers units heading to the main action continent's focus point idle units then it'll work. In theory it's also possible to then select the next 9 best units and build another stack and keep going until there are no more units - but this is most likely excessive and drains the garrisons of cities completely. Building one stack at a time should be enough.

[Nelphine]

The main problem with an AI doomstsck is that the human can't see it coming. The AI doesn't know a human doomstack is coming, and currently, to counter this, the AI is permitted to target the human doomstsck outside of sight range.

In this way, the humans can form doomstscks, and the AI cannot. The humans cannot target doomstacks from far away, but the AI can.

There's a natural balance.

If the AI is permitted to form doomstacks outside of human viewing range, then they should not be able to target human doonstacks outside of viewing range.

I'll come back to more on topic in a bit.

[Seravy]

On viewing range :
Doesn't matter. The human player can cast the spell immediately when it appears on their screen. The AI cannot, it has to go through random rolls still. There is no change to that and until the AI works on a "hey there is a doomstack, I'm going to Firestorm it right now" basis instead of the usual "hey, I want to cast a firestorm now, ok, done, now I need a target, is there any?", basis, viewing range is unfair to the AI. The only spell the AI casts that way is Extreme Stasis in the new procedure, but that already has a range limitation of 7 - if it even works, haven't tested it...
Also the human player can (and will, see Hadriex) split their doomstack into two halves to avoid getting targeted while the AI cannot.

anyway, back to the topic.

Doomstack algorythm preliminary testing...

-The AI succeeded in building a 9 stack of sprites in 1403 through the procedure.
-Previous attempts weren't wasted either - while the stacks didn't reach 9 units, the 5-8 of them attacked stuff and conquered/cleared them.
-However, the AI's cities aside from the capital contain close to no "intercontinental" units - they all got pulled out to build the stack(s).
-Unfortunately the stack didn't find anything useful to do and got called to the main continent where it kept breaking apart then reassembling (as the continent focus point and the stack assembling dragged the units between each other), but if there was a target to fight it would have stayed together. Maybe increasing the range intercontinental stacks check for a target would be a good idea? idk, probably not, there would a target in range if there was an actual war (or these were stronger units)
-After the "top 9" units were out busy forming stacks, the other sprites and nagas summoned to cities managed to stay inside so this doesn't eat up garrisons unless all the built stacks are busy attacking targets still. (in which case offense instead of defense is probably a better plan anyway)
-As this is limited to intercontinental units, walking units were kept as garrisons as expected.

-While the stack seemed to be busy doing nothing, then a single sprite split off for seemingly for no reason, what actually happened is the AI summoned a gargoyle, which joined the stack as the 9th unit, being more powerful than a sprite. While not originally intended, this algorithm is capable of replacing parts of the stack with better units as they become available o_O ...and to confirm it, in just 2 more turns another sprite got replaced by the next gargoyle. (lucky the summoning circle was so close to the stack I guess)

Hard to believe but it seems this is working and aside from the AI having much less flying and waterwalking units in garrisons, I see no downside yet.

[Nelphine]

The problem with viewing range is that a good flying stack can attack from out of viewing range. If you want to limit the doomstack to speed 2 or 3, that would be fine. But that would be against the very purpose of a doomstack.

And yes, teaching the AI more anti doomstsck tactics (similar to stasis) is a good idea. If you're going to teach them to use them, then you should give them the same limits as the human, and teach them how to act within those limits.

[Nelphine]

On another note, I'm not sure this is a good direction to take the game. Better doomstacjs is undoubtedly smarter, and more dangerous, but is it actually as much fun for the player?

[Seravy]

The problem with viewing range is that a good flying stack can attack from out of viewing range

That's what your navy is for, to intercept stacks coming from the sea, or at least spot those too strong to kill by ships.
There are million ways for the human player to see a map area if they really want to (put a lizard spearmen there, cast nature's eye, etc...) while the AI is incapable of doing that.

And no, the AI will never learn how to kill a hero doomstack or a shadow demon doomstack with Fire Storm and Blizzard because those spells are incapable of doing that. Stasis and Dispelling Wave are the only things remotely capable of stalling a doomstack in the game.

...more fun to the player? I don't know. I do know I don't find it fun when the AI keeps sending millions of weak crap at my cities and only has a chance to win if they have unstoppable superiority in combat spells no matter how much stronger they are because there is no way they can build a stack that can kill my <insert 9 normal units here> unless they are like 5 times my power level.

If needed we can always adjust difficulty modifiers again, but an AI that knows Efreet or Great Drake before you do should be a threat, it shouldn't only mean their capital is slightly harder to beat.

(if you're worried about the sprites and nagas, they actually managed to form large stacks anyway - there were plenty of them in the early game so it happened naturally - the difference will be in the BIG units.)

[Seravy]

One more positive side effect : The AI will keep their stronger heroes in these doomstacks, as they have a very high "gold cost" - as long as they were enchanted by some sort of flight or waterwalking spell. This helps the heroes gain experience, stay alive, and it's harder to steal the items from the AI.
   
One doomstack the AI built. These are their very first doom bats actually.
About 10 turns later it looked like this :
   


(the sprites and gargoyles are probably there by accident - they stepped on the same tile but are heading elsewhere)

...but it's still not bug-free, now it keeps going back and forth in an area without getting more units.

[Seravy]

Two turns after the bug was fixed, the doomstack improved to this :
   
for reference the image also shows the forces the AI has in their capital.
It's up to pure chance whether the first few units go to the capital or other cities but the odds are 1/1/1 for the capital, a random city, or the frontier city, out of which 2 are used for building the doomstack so it's 2 out of 3 unints for the doomstack.

Note that only intercontinental units can be used, so if the AI summons walking units and does not buff them, such as Great Wyrms, they stay where they were summoned until they exceed the required garrison quota.

...due to this, it might be worth enabling the AI to use the "intercontinental" spell group even if they have ships?
Or not...such stacks would be extremely vulnerable to drowning by dispelling wave. Let's not overdo that kind of strategy.

[Nelphine]

Absolutely the human CAN do things to see a doomstack coming. Put a unit out on the ocean though it'll probably get hunted down. The AI is very good at controlling the seas. The AI, aside from putting tiny units out far away, is just as good at scouting as the human. And the AI already has swarms of units more than the human. They are naturally scouting all the time.

I'd be fine with the AI getting 'vision' in squares from cities. I think it should be able to cast within 11 squares of its capital, and 5-7 of any other city. It doesn't need to be true vision for the AI, just an approximation. 'But that's a huge distance - that's probably its entire empire plus a big chunk'. Exactly. The doomstsck should always be formed in a controlled area far from the enemy. Then it zooms in for its target. As soon as it gets close, boom, valid target. 20 squares away from the enemy, on another plane? No. Not valid. Just like the human will never see it that far away. Does that lead to some nastu situations, like planar travelling to attack? Of course.

But right now, planar travel means absolutely nothing. Control one tower, that's good enough. Astral gate and planar travel are almost obsolete aside from earling settling.

[Seravy]

First downside of the new procedure found. Lizardmen and Draconian basic units are intercontinental, so are the swordsmen spawned from settlers enchanted by waterwalking. In the early game those can get dragged into the doomstack, leaving smaller cities completely empty.
Probably a check needs to be done to make sure if the units do come from a city, it has defenders remaining...but I have doubts if there is enough remaining space for that.

...tho I still don't understand how a pop 3 lizardmen city has both no guards and no sawmill. Weird. Maybe a neutral attack?

...another empty lizardmen outpost. I think I know an easy solution for this though - make the minimal cost used for the doomstack 55 - that way swordsmen can't quality.