Login

Seravy · September 5th, 2017, 08:08

Well, okay, let's assume most of the skill does go into unit buffs, city buffs and creatures - the other categories are marginal. (City buffs can be expensive because they are cast on all the cities which can mean 30+ times)

Creatures - we already know the AI is inefficient with these. I think it's a very conservative assumption to say at least 50% of their creatures won't be at a relevant location and/or get wasted. Either they stay guarding a city you don't care to destroy for a long while (yes that can include the capital, sometimes there is no need to attack it first, or at all) - or they get included in some random stay along with 8 halberdiers and get killed easily and they aren't good enough if not in a doomstack. So for this alone we probably need more than 50% discount and not on the highest level but everywhere except normal.
Unit buffs - the AI sucks with these. It's better now, but I risk saying much more than 50% of the buffs will still be on units where it won't make a difference, like that one strong unit which gets killed anyway because it's in a weak stack. Or simply, it's a buff not relevant to the game - Regeneration if its stack is losing the battle, Magic Immunity if it gets killed by melee units, Resist Magic against a Life wizard and so on.
Cit buffs - The AI is fairly good at these because "put all on every city" is not a bad move. However, there are city buffs that are completely wasteful to cast in certain cases, such as city defense buffs if your territory is not being attacked. Probably about 40% of the spells here are wasteful.

(note that all of these are random guesses)

So we have two major categories where the AI has a horrible (worse than 50%) efficiently, one that's slightly better, and then there are all those other categories where it probably is more or less efficient such as curses and global enchantments. that puts the overall efficiency somewhere in the range of 30-60% but the AI can't prioritize between the goals either, so while it has that much chance for a spell in a given category to be relevant to the game, that still doesn't mean it's the best option.
(as a human player, I often pick one spell and cast nothing for dozens of turns if that's my best option. The AI doesn't do this).
Overall, I believe not even the Lunatic 40% would be enough to make up for how poorly the AI picks spells, but fortunately they also get other advantages to fill in for that.
Finally, we can't improve the AI's overland spell selection in a way that changes this - the best play is deterministic and deterministic play can usually be exploited by the human player, so any "improvements" would actually make it easier to beat.

(either way, we have no way to measure how much casting advantage the AI "needs", so we have to keep guessing.)

Seravy

Btw a fairly simple solution to improve the balance of Time Stop would be to add "During this spell's effect, overland casting skill is reduced by X%", that way pushing the point where you get massively ahead on casting skill by using it higher. If anyone is interested in this solution, I can calculate the numbers.
(tho with Doomstacks around, being able to take 20 consecutive turns and destroy 10 cities in a row before the player can do something about it might be bad enough regardless of skill - not much can be done about that except higher maintenance.)

Arnuz · September 5th, 2017, 08:29

Agree overall but for one point. Creatures garrisoning a city are hardly useless. They prevent attacks from other AIs, resulting in the sensation of
- pointlessness of AI-AI wars
- making it seem that the game is rigged with all the AIs Vs you from the beginning
I find that unit buffs aren't that bad nowadays, the priorities are quite well placed.

On the skill: Nelphine, with that amount of cities that'll come from amplifying towers in large part... I am quite convinced that making them feed the skill point growth mechanism as spell power, rather than the current linear solution despite the maintenance change, would solve a lot of issues.

Seravy

Okay, so unless there are any objections, I'm going to start implement these changes :

1. The AI can't use Spell Blast randomly if the primary target has below 200 overland casting skill.
2. Final Wave will cost 1000.
3. The AI will pay 40% more for listed spells. This puts the required skill to instant cast them at 700 on Master which is unlikely even at Huge land and Max power except maybe at the last few years before SoM, at which point it's acceptable.
This puts the AI's effective casting cost of globals at
140% for Easy, 112% for Normal, 92.4% for Advanced, 70% for Expert and Master, 56% at Lunatic, so on all except the beginner levels, the AI is still casting them cheaper than the human player, but the reduction is not as much as on other, repeatable spells.

These are the spells to be affected (mostly globals) :
-Survival Instinct
-Clairvoyance
-Herb Mastery
-Nature's Wrath
-Spell Binding
-Great Unsummoning
-Suppress Magic
-Time Stop
-Meteor Storm
-Doomsday
-Armageddon
-Chaos Surge
-Crusade
-Charm of Life
-Eternal Night
-Evil Omens
-Zombie Mastery
-Final Wave

4. Haven't done the revision of the AI spell blast list yet.
This is the current list, suggestions are welcome.

Nelphine

Why not simply increase the cost of all globals plus final wave and great unsumkoming - I'd also include disjunction in that list. Also I like this solution. Its effectively a compromise between keeping high overland skill, and reducing it as I want, for the one category where high costs are meant to be a huge factor. Which is why I want to include disjunction and all globals - the whole category is meant to be low use high cost. Might also want to include disenchant area to make human curses more effective.

Seravy

(September 5th, 2017, 09:31)Nelphine Wrote: Why not simply increase the cost of all globals plus final wave and great unsumkoming - I'd also include disjunction in that list.

Can't increase the cheap ones because that would make them act functionally different against Suppress Magic and/or Nature's Wrath. Disjunction falls into this category as well, having a base cost of 375, but it's worse - being a slider spell, the strength is calculated from the total cost set which we are raising, so it'd make it stronger.
Also, I'd like to keep this effect to the minimal possible, we are making Spell Blast more powerful by it afterall. Globals that aren't directly harmful should slip through.

...in fact, I should probably remove Life Force, Enlightmentment as they have no military or immediate harmful economic effects.

Note that high cost and low use have no inherent correlation - we can reduce Disenchant Area/Disjunction use without raising their costs and have the exact same outcome (ok, the AI has a few percent extra for other spells but that's about it)

Edit : we might want to adjust the recast rate on "high recast rate" globals to prevent the case when the AI is not casting anything else because the player can spell blast them every turn on Armageddon. (realistically it's loser to "casts 40% fewer spells because those are blasted Armageddons" but I think you get what I'm saying. Not saying to make them low, but from the current 200, maybe make them 100 (equal to summoning) instead?

Nelphine

I know enlightenment and life force are huge targets for me. They should be on the list. I'd want things like aether binding on there as well. Far more than something like zombie mastery. Economy things ARE things I want to deny the AI. But I was thinking you were able to modify their overland bonus for these spells, instead of changing the cost for the spells (that would allow for sliders to be affected).

And by low use, I mean things that ate low use, regardless of cost. If the AI is only going to use just cause once, it can afford to not have such a huge advantage over the player while cssting it. It becomes a 'choice' - or in the AI case, an opportunity cost.

This is why I think all globals, disjunction, final wave, great unsummoming and disenchant area should be in here. Spell blast too, but that's only like 20 more cost so who cares.

Seravy · September 5th, 2017, 10:17

Quote: But I was thinking you were able to modify their overland bonus for these spells, instead of changing the cost for the spells (that would allow for sliders to be affected).

I can't. I mean I can do that but Spell Blast will not be aware of it (and there is absolutely no way to change that, cram in the current effect was already a huge task) so it'll calculate negative amounts of mana as needed to blast the spell because there would be more remaining cost than what it assumes to be the total and that probably crashes the game - even if not, we don't want spell blasting to produce mana :D

Nelphine

I'm confused. How does spell blast work currently with overland casting skill bonus of the target?

Edit: either way, I'd still include all globals even if you can't do the slider spells.

Seravy · September 5th, 2017, 10:37

When casting a spell,

Total cost = base cost
Cost remaining = base cost * AI cost multiplier

Then,
Spell Blast cost = base cost* AI cost multiplier-Cost remaining

Final Wave (new) on Lunatic

Total cost = 1000
Cost remaining = 400

Spell blast cost = 1000*0.4-400 = 0, correct, as the AI made no progress yet.

If the 40% modifier is added only to remaining cost then

Cost remaining = 560
Spell blast cost = 1000*0.4-560 = -160.

If added to both then
Total cost = 1400
Cost remaining = 560
Spell Blast cost = 1400*0.4-460 = 0, the correct amount.
The 0.4 is a constant in a table so it can't be a function of the spell type.

Navigation

Extra Menu

About us