As a French person I feel like it's my duty to explain strikes to you. - AdrienIer

Create an account  

 
Myrran AI Performace

I decided to run some tests with the self-playing patch to see how much the AI expands on Myrror. Always huge map and impossible.

These are the screenshots at turn 100 :

   

The Myrran AI clearly didn't do well expanding here (purple). Playing Dwarves who grow slowly and have expensive settlers which might explain it.
Red has nearly as many cities as Purple even though they came from Arcanus - Barbarians break towers easily and spread quickly as well. They are at war with purple, which probably slowed them down further - in fact some of the red cities might have been purple originally.
Yellow sent a lot of units through Astral Gates, but no settler managed to reach Myrror among them, as they were produced in cities not equipped by the spell or blocked by ocean. If they hit the neutral on the continent, they can start spreading.


   
This time it's much better - using Draconians-, but they still managed to fill only about 35-40% of the plane.

   
About the same as the previous, but a lot of space taken by an Arcanus wizard.

I'm not so sure if the AI is doing a bad job with settlers - sure, it didn't fill the entire plane but is the expectation to do so even realistic? 15+ cities in 100 turns is a lot. Either way, the next 3 runs will be these exact saves, but with a higher settler limit to see how much difference it makes.
A more alarming thing I see here is how 2 out of 3 games broke a tower before turn 100. We might want to strengthen those further? (Impossible of course means they do it earlier, but still it seems to early to me?)


With 4 settlers per continent and 6 per plane

   

Don't see any significant difference here. While Red hasn't taken away nearly as many cities from purple yet, the war has been declared and they'll be taken soon, resulting in the same game state as with fewer settlers.
It seems, for dwarves, the speed of producing settlers is more limiting than the number they can have at the same time - not that surprising.

   

Some improvements here - the territory filled is the same, but there are no gaps between cities, bringing the total cities to 5 higher than before. Unfortunately, the quality of these spots are horrible - they were left empty for a good reason, so it's not a real improvement.
...and now I realize draconian settlers fly so the continent limit wasn't affecting them much in the first place - it's unlikely for them to head to destinations on the same continent at a rate larger than 2 at a time anyway. ehh. Might as well not bother running the third one as that wizard was also a draconian. What a waste of time.


Instead, some dark elves, with the raised limits


..in the next post as it seems I can't attach more then 5 in the same one...
Reply

   
Dark elves with 4/6 settler limit

   

And the same with the original 2/4.
7 fewer cities, so I guess the difference is significant, especially if we assume some of the remaining cites probably got built earlier, which means they might be more developed.

I think it's worth adding the following AI rule :

If turns<100 and difficulty>=Extreme and AI Myrran and human player is not Myrran and land size is Large or Huge, max settler limit +2 (both on continent and on plane)

However I think the real limiting factor here might be the "no more than one settler produced at a time", maybe that should be dealt with as well? Count produced settlers as though they were already made, but still check towards the total limit, not to "one at a time"?

...also the AI is forced to build a fighter's guild and 4 units in any cities if they have 2 or more active settlers, which slows down expansion - I believe we should keep this to make sure the expansion doesn't turn into overextending with no military to defend.
Reply

How does the AI defend cities? Does it rely on units produced in the city, or can it reinforce from other cities?
Reply

(February 21st, 2017, 01:38)Catwalk Wrote: How does the AI defend cities? Does it rely on units produced in the city, or can it reinforce from other cities?

Units produced in the city, plus any idle units on the continent not already defending something else or heading towards a target or rally point can be pulled into the city if garrison size is too low. Intercontinental units probably cannot, as those will head towards the destination continent instead - no flying or waterwalking reinforecements - but there is an exception, if the continent is "type 2" meaning it has significant enemy presence, then intercontinental units won't leave it.

"too low" is less than 4+(population/2) units, or 9 for the capital.
Reply

How about giving city defense very high priority when choosing stack targets, to free up new cities to focus on development instead? That also allows for more potent city defenses.
Reply

(February 21st, 2017, 07:08)Catwalk Wrote: How about giving city defense very high priority when choosing stack targets, to free up new cities to focus on development instead? That also allows for more potent city defenses.

That breaks the system, as it results in a stack already heading towards a military target to split up halfway there and send some units back to a city. I tried it and it doesn't work well.
Reply

As far as I know, the decision to build military or buildings almost doesn't care whether the city has units already. Basically, city production decisions have no knowledge of any units already in play. Thus the cities have to build troops often enough that IF there are no other units available, then the city will end up being defended adequately. Which means all cities have to build a fair amount of units - even if they already have invisible sky drakes defending them. So the only thing that could change is how many troops it wants to build, in a generic situation that is meant to cover all possible ingane scenarios.

This is the biggest reason the AI still needs huge bonuses to resources and production/casting skill. They have to be able to build/cast so many things that they happen to build/cast the right thing, along with 3-8 things that might be totally useless.

This is also why things like changing settler priority/amount has to be so carefully considered. The AI literally has no idea what is going on, so if its building too many settlers, then it might not have enough military and it won't know it.

Similarly, the sawmill question. The AI doesn't actually 'pick' things the way a human does. It has to use an algorithm, for all cities, that somehow works for all situations in the whole game, do it needs to have some high priority things that ensure, no matter what else it picks, it will do so efficiently enough to make us think it actually is an AI 'picking' like a human would.
Reply

(February 21st, 2017, 07:47)Nelphine Wrote: As far as I know, the decision to build military or buildings almost doesn't care whether the city has units already. Basically, city production decisions have no knowledge of any units already in play. Thus the cities have to build troops often enough that IF there are no other units available, then the city will end up being defended adequately.

This isn't true for the city itself - it knows its own garrison size (and will 100% make a unit if below 4). It's true in general however - the AI has no knowledge of how much units it has overall on the continent or world -instead 3.2 AI is aware of enemy cities on the continent and will make units more often if there is a large enemy presence, or if they have better resources available for making the units (better military buildings already built, mithril, adamant, orihalcon, etc).

Settlers however override the "unit or building" decision and count as neither - they can be made even if the AI wanted a unit. Same for ships and engineers.
Reply

Ha! Its getting better!
Reply

(February 21st, 2017, 07:42)Seravy Wrote:
(February 21st, 2017, 07:08)Catwalk Wrote: How about giving city defense very high priority when choosing stack targets, to free up new cities to focus on development instead? That also allows for more potent city defenses.

That breaks the system, as it results in a stack already heading towards a military target to split up halfway there and send some units back to a city. I tried it and it doesn't work well.

How about this:
When a stack doesn't have a target, it first checks to see if any cities need to be reinforced. Higher priority given to cities closer by. Units en route to defend a city count as defenders. Would that work? Aggressive actions aren't aborted that way, unless I misunderstand.
Reply



Forum Jump: