October 3rd, 2020, 18:04
(This post was last modified: October 3rd, 2020, 19:15 by Seravy.)
Posts: 10,463
Threads: 394
Joined: Aug 2015
Quote:But actually, you only need a spell that any one wizard doesn't have, and for that wizard to have a spell that some other wizard doesn't. You can then take the newly traded for spell, and trade it again with the 3rd wizard. And again with the new spell...
Except, each time you can only get a spell of equal or lower value.
So assuming spells 1 to 10 have values equal to that number, you need to trade spell 10 for 9, then 9 for 8, then 8 for 7, then 7 for 6 and so on.
Furthermore, the AI with spell 9 cannot know spell 10, the AI with spell 8 cannot know spell 9, then AI with spell 7 cannot know spell 8 and so on.
That's already extremely specific but in addition to this, all players will research spells in the order of 10,9,8,7,6,5,4,3,2,1 so a player that satisfies the previous conditions can only happen if the spell with the higher number showed up on their research list too late (or they can't research it at all).
Of course that's only theory, in practice the research priority and the trade value only has strong correlation, they don't have the exact same order.
The AI can't trade rapidly for the obvious reason of game balance - there is a chance the human player is playing a realm no one else plays so they can't trade at all, or have significantly different amount of trading partners. Trading rapidly also breaks the game pacing among other problems.
So assuming trading is a problem (I don't think we can actually prove it but I'm willing accept it's somewhere between "major problem" and "not a problem at all" based on experience from the two tests) we do need a solution that slows trading down.
Quote:-just make the AI trade as often as possible so the player doesn't have much opportunity to broker anything
See above this is not an option in a game where the number of trading partners is a random amount between 0 and 13 for everyone including the human player.
Quote:-limit the number of times you can trade in one turn, with other wizards refusing to deal with you
We can't do this either because it does nothing. There are 72 turns to research a spell tier and each tier has roughly 12 spells (in 1 realm+Arcane), so the player only needs to trade one spell every 6 turns to have all spells without researching anything to keep up with enemy research speed. I'm obviously not going to tell a player "You traded with Ariel 5 turns ago so even though there are these 6 other players you could trade with, you just can't because the game doesn't allow it. Wait another 5 turns, the rulebook says it's 10 turns between two trades..."
Quote:-make the AI more reluctant to trade spells of the current highest tier. Want any of the good uncommon spells? AI won't give it to you until they start researching Rares, unless you give them something they really, really want.
This can work, probably even better than what I added today, but where do we draw the line?
"Never trade spells with trade value>=(best known spell trade value+4-Number of AI wizards)"?
I do see one major flaw in this though. If the AI finds a spell of a higher trade value or the player gives them one intentionally, they basically break the seal. And the player is trying to trade away said higher trade value spells.
So basically, if I have 3 good spells I can trade away, I am forced to give the AI the best one for something not so great, but I can trade the second and third spell the normal way.
...actually there is another flaw. The AI will research that high trade value spell first on their own because it means it's a very good spell. For example in Life, Prayer has trade value 33. The next best uncommon is 24 and uncommons in general are in the range of 12-25. But the AI will research Prayer first. If they did, they won't hesitate to trade away those value 24 uncommons because they are so much worse than Prayer. And of course the player would never be able to trade for Prayer either way because 33 value is the highest among all uncommon spells, there literally is nothing they could offer for it, they need at least an above average rare.
Anyway, I am going to need a formula or something to evaluate otherwise this idea only sounds good but is too generic to implement.
For reference, we had until today
"If AI spell trade value<=Human spell trade value then trade is ok."
and now it is
"If AI spell trade value<=(Human spell trade value)-Max(0,(Number of wizards-4)/2) then trade is ok."
and this is only for trades the human offers.
For trades the AI is offering, the rule is
"If AI spell trade value<=Human spell trade value + Min(6,Relation/10) then trade is ok."
Quote:You can see the full table here : https://www.realmsbeyond.net/forums/show...?tid=10059
There is a -5 modifier for spells a player could research on their own however, so offering a spell the AI cannot get for one you could get can help you trade more. So maybe considering that a +4 is too little.)
To better understand the difficulty of designing a formula for this :
Assuming the AI has access to all spells of the current tier and next tier to pick from (in reality it's random of course), and also assuming none of the game state specific modifiers change it, in case of Life, the AI will pick spells in the following order of trade value :
19, 15, 15, 9, 6, 18, 3, 8, 20, 23, 5, 2, 33, 18, 15, 9, 14, 21, 38, 35, 29, 24, 11, 31, 28, 30, 24, 31, 34, 27, 21, 22, 30, 50, 34, 40, 42, 52, 45, 43, 36, 37, 36
As you can see the AI will usually research several high trade value spells before going after the medium value spells in the same tier, and leaves the low value spells to the end of the tier or even wait and skip them entirely until the middle or end of the next tier.
So basing a formula on the best spell the AI has is difficult because research does not happen in the order of increasing value, far from it. (and of course finding an early rare in treasure can let the AI have even a 35-40 value spell before they have anything else over 20....)
Maybe if we used "average of the trade value of the best 4 spells the AI knows" in a formula, that could work and smooth out the massive jumps in the value of the best spell known?
For example :
"Never trade spells with trade value>=(average of best 4 known spells' trade values+4-Number of AI wizards)"?
Edit : But this doesn't make sense from the AI's perspective if I stop and think about it.
I mean, obviously, not trading my best spells for something less valuable is correct strategy but the AI never trades for a less valuable spell in the first place at least not when the human is asking for the trade.
In other words this feature makes the AI not trade their best spells for even better spells. Which makes no sense. Accepting that trade would raise the AI's advantage towards everyone and reduce my advantage against them.
Posts: 343
Threads: 4
Joined: Mar 2020
I will try to see how trade brokering works out on a test game.
I'm running Maximal map 13 players with Large continents right now, but haven't encountered anyone for 23 turns yet and have magic spirits moving in 4 directions, going as far as the edge of the what I think would be a normal COM I game map on all 4 sides (one screen on the Cartographer) and I've only seen 2 tiny continents including my home continent, which can only fit 4 towns each. The distances between continents seems excessively long even though I chose Large continents. This is quite strange.
October 4th, 2020, 04:00
(This post was last modified: October 4th, 2020, 04:13 by Seravy.)
Posts: 10,463
Threads: 394
Joined: Aug 2015
It's most likely because you chose large continents. If you put fewer but larger continents on the same size map, the distances to travel to find one will be greater than if there was a hundred tiny islands.
(it's also possible you simply had bad luck and the continents were in different directions than where the spirits went.)
But I'll try generating a few with that parameters, just in case. (oh, Maximal map...those are huge, it's very possible you have no land in that area through sheer bad luck.)
...looks like it's working correctly. Yes, there are a few near-empty areas that are the side of the area visible on cartographer. But there are also places where the same amount of map space has 7 large continents.
Posts: 10,463
Threads: 394
Joined: Aug 2015
In automatic combat, I modified something that might be related to the AI's ability to banish other AIs or hunt for treasure.
In the previous version, if one side dealt 20 damage capable of hitting flying units to an army with 100 HP out of which 20 was flying HP then that was calculated as "20% damage dealt that damages flying units" so both HPs took 20% damage - normal HP went down to 80 and flying HP went down to 16, assuming the damage is distributed equally between targets, basically assuming the attacker didn't care to deal damage against the flying HP pool any more than the non-flying one. Same applied to invisible, magic or missile immune HP pools.
I changed that today (not yet uploaded) to apply the full damage to all HP pools. So if an army deals 20 damage that can hit flying enemies, than that example army goes to 80 HP out of which 0 flying. Basically, it assumes the attacker cares to target things with attacks that cannot be hit by other types of attacks.
The big inaccuracy in this is that it applies to all special HP pools, so if the attacker launches an attack that deals 20 damage that can hit flying, invisible, missile immune, and magic immune units, and the enemy had 20 of each then all four types go down to zero. In other worse we get the outcome that would happen if the same unit had the four abilities. So when they are four separate units, it's much easier to defeat them than it would be real combat - you only need to do one attack that is able to damage all four of these things instead of four. I think this is acceptable as that situation comes up rarely. (The same unit being buffed with multiple immunities is more common.)
October 20th, 2020, 15:40
(This post was last modified: October 20th, 2020, 16:45 by Seravy.)
Posts: 10,463
Threads: 394
Joined: Aug 2015
In my current test game I have Chaos books and I noticed Chaos is even less effective than it used to be mainly due to AI improvements.
In particular, Wall of Fire was the best common Chaos spell, allowing you to destroy entire enemy armies for 20 MP...but it doesn't do that anymore. The AI is smarter and won't attack into the wall with units that have no realistic chance to survive and damage the unit on the other side of the wall. Which means the wall only buys time, no longer kills enemy units. This doesn't seem like a large change but I estimate at least half the enemy units I killed in my Chaos games at common tier were killed by the wall so it IS a big deal. On the upside, it does make casting the wall as an overland spell worth doing because that way you don't need to worry about the enemy retreating and coming back the next turn - they still cannot enter so you can slowly kill them a few units at a time with damage spells or ranged attacks.
Of course, putting Immolation on the unit that is guarding the gate still works for the same effect, but that spell is not available early enough.
I haven't played most other realms yet so it's very well possible the improved AI simply makes all realms feel weaker but it might be worth thinking about how to improve chaos just in case.
One thing I do know is Misfortune is a great spell, so we don't need to worry about that, it works very well for a common.
Edit : Anyone remembers if we wanted Inner Power to be +2 or +3 defense? Helps says 2 but effect is 3, not sure which is wrong.
October 20th, 2020, 16:52
Posts: 343
Threads: 4
Joined: Mar 2020
It seems to me that either AI improvements, or some yet unidentified issue with the game has made some AI perform much better than expected on high difficulties and others much worse.
Some of us mentioned in the other thread that AI Gargoyle stacks are really dominating in the hands of AI, so I don't see Chaos as being particularly less effective.
I've also found the AI's use of Volcanos now incredibly annoying. They are spamming it on the same cities, covering every tile in the city catchment except the coastal ones that they can't. They're even casting on the city tile itself, and they didn't use to do that much in COM I. I think there ought to be a limit on adjacent volcanos, otherwise the +3 Power should be lowered, or the limitation on Volcanos being unable to be raised on existing Volcanos or certain types of terrain should be brought back.
This shouldn't be viable as an economy spell and also have the same effect as the much higher level curses when spammed, AND be undispellable/irreversible, at only the Uncommon tier. Which is better? Pestilence for 250 MP and slow killing the pop and pay spell maintenance, or 5 volcanos to permanently destroy 5-10 pop cap/food bonus cap/targeted minerals, while returning 15 Power/turn?
October 20th, 2020, 18:11
(This post was last modified: October 20th, 2020, 18:16 by Seravy.)
Posts: 10,463
Threads: 394
Joined: Aug 2015
There is nothing wrong with volcanoes, and yes they are intended to be a viable economy spell. I recommend reading the old discussions about it in the Chaos thread.
One thing I noticed in the current game is the AI already has a few other rares but not Flame Strike even though they prioritize researching that very high. I peeked at the wizard data and it's because Flame Strike hasn't shown up on the research candidates. (They have already researched Inner Power, Elemental Armor, and are researching Chaos Rift.)
The following spells are clogging the list up :
Nature's Curse
Chaos Spawn (this should actually be picked before those the AI already picked I think, as the AI has no rare summon yet)
Warp Lightning (this is lower priority so that's fine)
Chaos Rift (currently researching)
Disenchant Area
Detect Magic
Create Artifact
Summon Champion
Now the rares are fine since they would be expensive to research but maybe we should push the commons and uncommons to higher priority to open up more rare choices for the AI to pick from. Detect Magic in particular as it's common and cheap to get but was pushed up to rare tier in the research priority due to Spell Blast moving up. This also shows how critical trading with the AI is - most of the spells on the list are things I wouldn't hesitate to trade to the AI, yet giving them any 2 of those would likely allow them to get Flame Strike several years earlier as those spells no longer block their list from getting more good spells added.
this could be one reason for the random performance of the AI - those who get lucky and get the good spells on their list first will have a major advantage over those who get the weak rares first and can only progress by researching those weak rares, as the remaining uncommons are rated even lower priority.
Edit : they didn't pick Chaos Spawn because they consider it the same "tier" for research choices as Chimeras. In general the weak rare summons and the good uncommons are the same category, so the AI will almost never prioritize Chaos Spawn because missing Chimeras is unlikely. Maybe this should be split up to two sub-categories? Although Chaos Spawn are pretty horrible for the AI to use so maybe not...but Angel and Stone Giant should be rated a higher tier than the good uncommons? I dare say Stone Giant isn't much worse than Gorgons considering Gorgons are slow and easy to kite but the giants can shoot and have better armor and resistance.
October 20th, 2020, 22:27
(This post was last modified: October 20th, 2020, 22:30 by massone.)
Posts: 343
Threads: 4
Joined: Mar 2020
(October 20th, 2020, 18:11)Seravy Wrote: There is nothing wrong with volcanoes, and yes they are intended to be a viable economy spell. I recommend reading the old discussions about it in the Chaos thread.
I did a forum wide search and the only posts I found after it was changed to +3 Power have all been negative, in the Chaos and Research Costs threads, and I found nothing explaining why they're fine. I can see that some of this is probably because players don't like getting hit by volcanos, but it wasn't this bad in COM I. The AI is actively spamming the volcanos in a manner that bombs specific cities, and they're doing it in a single turn because of cost reductions for high difficulty AI. It's not being used for the purpose of destroying minerals anymore, and I'm fairly sure this is another reason that Chaos AIs are growing their Power at ridiculous paces (this and the Gargoyle taking nodes issue).
And frankly, I think they're unbalanced even in COM I, for the reasons I mentioned in my last post, and there a host of other reasons it's bad:
-Destroying Adamantium or Orihalcon on Arcanus is ridiculous and extremely unfun. There is usually only 1 adamantium ore at most on the map unless you're playing Nature, and often none when not playing Rich. With Raise Volcanos, what is a wondrous find or a major strategic objective becomes worthless. Even if the ore doesn't end up making that much difference (which I think it can, for certain races and setups, even a single ore can be incredibly powerful because you can essentially skip casting 9 Heroisms on your doomstacks, and I'm always planning my cities builds around that whenever I see it), it reduces strategy and variety and forces players to focus more on the numbers game, and is overall just a source of frustration.
-Generating 3 Power on 50 Casting Skill is almost worthwhile on its own, considering that Heavenly Light is 3 on 60, and Sanctify is 3 on 70, and the latter two can be dispelled or killed.
-Every other curse in the game can be dispelled, resisted with nightshade, or blocked somehow. Volcanos can't, and it's only Uncommon. This should make it comparatively not as good for the effect, but it's actually also arguably better than the Rare and Very Rare curses. I personally think that Pestilence is strictly worse and Drought is situational for when you want to bring down prod bonus. Drought's pop lowering is efficient at 6, but it also caps at that and can be easily dispelled before pop even goes down. Volcano is much better for destroying pop and keeping it low, especially in combo with other chaos spells, and high speed flying invaders, AND it provides an almost unlimited expansion of Power.
October 20th, 2020, 22:59
(This post was last modified: October 20th, 2020, 23:48 by Seravy.)
Posts: 10,463
Threads: 394
Joined: Aug 2015
In theory the AI formulas for casting curses and targeting volcanoes should be the same. The only change is, the AI knows the difference between different types of ores and targets the better one first. They shouldn't concentrate the spell on one city, if they do that's either a bug or a special circumstance, like it being the only city they scouted so there is nothing else to target. If it is a bug, I will fix it but for that I need more information.
-The economy effect on Volcanoes is fairly balanced. It's definitely not bad but not great either, and generally only worth it if you are not fighting anyone, otherwise you'll need to cast summoning spells, walls of fire, or unit buffs instead.
-The "curse" effect is bottom tier. You need to cast the spell 21 times to cover a city entirely which costs 1050 mana AND it doesn't work on a whole lot of terrain types that produce high food. Add that to the fact curses are rarely a good strategy for the human player and the AI is restricted in their usage by their personality, diplomacy, and scouting, none of which applies to the majority of non-curse spells in the game.
-Reducing maximal population does nothing to a city that already reached that population. You need either Doomsday or Drought in play for the population to naturally shrink, otherwise the city needs to be unable to feed itself which almost never happens. (buildings produce enough food for 8 units of population and then you get overfarming which allows you 1 food per farmer even if there is no actual food on the map.)
-It's powerful at countering ores which is intentional. It reduces the luck factor of starting near good ores and raises the relevance of diplomacy and scouting denial which are the two main ways to prevent Raise Volcano.
-The economy effect suffers from one factor no other spell does - it has a diplomacy penalty. You either target your own land, land no one is using at all, or land that belongs to someone you are already at war with, but then you're ruining the cities you're going to conquer.
Edit : If you still believe Volcanos are too powerful, prove it by playing an at least Expert difficulty game that you win by spamming volcanoes.
...or maybe Master difficulty? I don't know. Most spells should be good enough to be playable on Expert...can't even say it's a problem if that's the only spell used, I mean, spamming werewolves, spiders, shadow demons, etc can also win the game if done well. So it's a bit hard to draw a clear line between fair and too powerful. Well, if something beats Lunatic effortlessly that is clearly too powerful but I don't think Volcanoes are anywhere near that. I most definitely played a test game when they were increased to 3 power and they weren't all that great, just "playable".
Edit 2 : Found a bug that might be responsible for the inconsistent AI performance :
"-Fixed bug : the first AI player still gets a turn even after being eliminated. As a consequence, the first AI player who actually remains in the game executes their "start turn" phase twice, gaining income twice as well."
October 20th, 2020, 23:45
Posts: 736
Threads: 50
Joined: Jul 2020
Actually I've never had any success using Sanctify as an economy spell, and I was surprised too. In theory, making a unit cost no gold while getting 3 power for 1 mana upkeep should pay off handsomely. In practice, you need to burn a lot of mana with multiple casts before it starts having an appreciable effect, and your power sources are vulnerable to being murdered during wars. This makes it a bit of a wash overall.
|