Caster of Magic II Brainstorming Megathread

[Seravy]

Yeah but the AI is currently getting exactly the things it needs for the diiffuculty to be as intended. That's important, the AI can't function properly if it is merely getting random stuff, it has to get specifically the things they are inefficient at using.

[massone]

On the difficulty and retorts thing, I think it would be better and consistent for the AI to simply create a set of difficulty based profiles for builds. Don't make it randomly pick (except maybe keep a few "random" profiles to keep things interesting), but a pool of more advanced builds for it to choose from, that the AI can definitely handle. Easy to test, easy to implement, results in AIs that are challenging, and easy to update in new versions if players figure out better strategies that are simple for the next version's AI to copy and execute if it had the same build.

[prokolyo]

I was thinking about difficulty levels. Wouldn't it be a good idea instead of giving the AI numeric bonuses at the higher levels, to give them more book/retort picks (or less to the human for that matter). This way it would feel more natural.

I don't think so, books/retorts are both not enough advantage to be worth as much as the resource bonus, and a too large advantage in the sense an enemy that has more options available is much harder to play against at the same time.

Packing the resource bonuses into a few "AI only retorts", from which each wizard picks a few may have two advanteges:
1. Increasing versatility, because each enemy could have different bonuses
2. It is more transparent to the player, what bonuses the enemy gets, because they would be visible in the magic screen as retorts.

You took the words out of my mouth

[Seravy]

Higher difficult levels do increase the chance for the AI to pick the "best" retorts for each profile already. (already meaning CoM I)
That has absolutely nothing to do with -for example- the AI needing a high bonus discount to their maintenance costs because they are unable to even comprehend what maintenance is and keep building troops as long as there is production capacity to do so.
The bonuses are transparent, the full table is in the documentation - all effects, not only the resource percentage ones.

[massone]

Higher difficult levels do increase the chance for the AI to pick the "best" retorts for each profile already.
That has absolutely nothing to do with -for example- the AI needing a high bonus discount to their maintenance costs because they are unable to even comprehend what maintenance is and keep building troops as long as there is production capacity to do so.
The bonuses are transparent, the full table is in the documentation - all effects, not only the resource percentage ones.

I've never seen the AI stack retorts. I tested it with the CTRL+RVL command, and they seem to always pick extra books, usually of the wrong realm, taking at most 3 retorts, but usually sticking to just 2 or even just 1. Also, according to the table, they only double chance of picking the "top 4 retorts" at Lunatic, but that's only relevant for a very small minority of players. Even on lower difficulties, like Advanced and Expert, the AI seems to pick more books a lot more often than more retorts, sometimes even taking 0 retorts in exchange for more books.

Also, what exactly are the "top 4" retorts? Are we talking about individual retorts, or specific combinations known to be good?

[Seravy]

Each AI profile has 2 retorts assigned to it. It seems this isn't difficulty scaled though, the percentage to pick them is a flat (fairly high) number. We can make them difficulty scaled but I see no need, the base percentages are reasonably high.
What is difficulty scaled is Specialist for 1 realm wizards, and the 4 generic "this is good for the AI' retorts which are Alchemy, Warlord, Guardian and Tactician.
There is a 50% chance to pick a retort for each "leftover" pick which is very high as well.

[prokolyo]

Yeah but the AI is currently getting exactly the things it needs for the diiffuculty to be as intended. That's important, the AI can't function properly if it is merely getting random stuff, it has to get specifically the things they are inefficient at using.

The bonuses are transparent, the full table is in the documentation - all effects, not only the resource percentage ones.

I see what you mean - if it ain't broke, don't fix it, and agree with you.

But it'd be much nicer to say, hey, I beat this alchemy/warlord/channeler/archmage/sage master/cult leader wizard (just throwing those randomly, no idea what works best with what), rather than, hey I beat this wizard that had 200% production/research bonus Or does it... :/

[Seravy]

But there's also a required garrison number based on personality, right? So we can use that number instead. When there's 3 Colossus in the garrison for a Peaceful wizard, just push out all the rest to the field.

That doesn't work like that.  We have 3 colossus and 6 magicians, and we are Peaceful.
If we summon a Hell Hound (or build a swordsmen) then the magician gets sent out and the hell hounds/swordsmen stays  but we want the Hell Hound to be pushed out and the magician stay (because the hell hound has the lower garrison priority). 

The two functions pick from the opposite end of the list, stackbuilding takes the BEST unit except the first few, while "over9" takes the WORST unit. We can't use these two at the same time, we have to pick one or the other each time a unit is taken out for any reason.

Basically, we have over9 which is responsible for culling the weakest units while "stackbuilding" picks out the best ones (except mandatory best that has to stay anyway). We need to balance these two as they are both necessary, we can't replace one with the other completely.
And we need to balance them in a way that "stackbuilding" is used less often than "over9" otherwise the garrison cannot improve because we take from the top of the list more often than from the bottom.
The question is, how quickly do we want the garrison to improve and/or in which cases.

[massone]

But there's also a required garrison number based on personality, right? So we can use that number instead. When there's 3 Colossus in the garrison for a Peaceful wizard, just push out all the rest to the field.

That doesn't work like that.  We have 3 colossus and 6 magicians, and we are Peaceful.
If we summon a Hell Hound (or build a swordsmen) then the magician gets sent out and the hell hounds/swordsmen stays  but we want the Hell Hound to be pushed out and the magician stay (because the hell hound has the lower garrison priority). 

The two functions pick from the opposite end of the list, stackbuilding takes the BEST unit except the first few, while "over9" takes the WORST unit. We can't use these two at the same time, we have to pick one or the other each time a unit is taken out for any reason.

Basically, we have over9 which is responsible for culling the weakest units while "stackbuilding" picks out the best ones (except mandatory best that has to stay anyway). We need to balance these two as they are both necessary, we can't replace one with the other completely.
And we need to balance them in a way that "stackbuilding" is used less often than "over9" otherwise the garrison cannot improve because we take from the top of the list more often than from the bottom.
The question is, how quickly do we want the garrison to improve and/or in which cases.

I don't understand your example. What function would cause the magician to get sent out if it has the higher garrison priority? The over9 units should be pushing out the lowest priority unit. Stackbuilding can't draw it out because it would only be allowed to use field units, not garrison units.

It should work like this:
Summon a hell hound --> over9 pushes it out of garrison because it's weaker than magician --> run stackbuilding and it takes the hell hound
Summon a 4th colossus --> over9 pushes it out of garrison because there is already 3 colossus and we are peaceful --> run stackbuilding and it takes the colossus.

The only place where I see there might be a problem is if the colossus has higher garrison priority than magician. In that case, all we need to do is do a check for if the required garrison is already met. If it has, just push it out. Any other units weaker than Colossus but higher priority than magician can go into garrison, and push the magician out.

[massone]

Each AI profile has 2 retorts assigned to it. It seems this isn't difficulty scaled though, the percentage to pick them is a flat (fairly high) number. We can make them difficulty scaled but I see no need, the base percentages are reasonably high.
What is difficulty scaled is Specialist for 1 realm wizards, and the 4 generic "this is good for the AI' retorts which are Alchemy, Warlord, Guardian and Tactician.
There is a 50% chance to pick a retort for each "leftover" pick which is very high as well.

I think 50% is too low, and 2 retorts assigned is also too low. Because the AI removes the base profile retorts and rerolls new ones, doesn't it? And all of the standard profiles only have 2 retorts maximum, which means the average becomes just 1 retort, and the other gets switched for a book, if I'm understanding this correctly.

It only takes the 2 assigned retorts together only 25% of the time. 50% of the time, it takes 1 book + 1 retort, and 25% of the time, it takes 2 books instead. On Lunatic, it only has a 25% * 25% = 6.25% chance to take 4 retorts, and even when it does, 2 of them are random and not comboed.

But in fact, it's usually better and easier to play 4 stacked retorts, and there's specific combos that are much better than others.

Adding 1 or 2 books doesn't make that much of a difference, but stacking Specialist on top of Spellweaver makes a world of difference, especially on a Life Wizard that has Divine Order. This is an effective 43% overland casting cost, which is equal to the Lunatic bonus for AI with resistance to dispelling.

In mono-realm plays, it's much stronger to use 10 or fewer books, and use all the remaining picks for retorts, whether that's 2, 4, or 6 picks (assuming you take the right ones).