AoS Army Harmonizer

[samheim]

I had this idea for a software solution for AoS and 40K's terrible balancing issues.

Basically, it's a website. What happens is you feed in your army list before the game, and your opponent does the same.

What happens then, is the software recalculates the warscrolls and adjust's their stat lines to harmonise the lists.

So say you have a cav unit, that unit would be given a label like "Fast".

The cav unit would then be given a move characteristic randomly between 7 and 9.

The software would use keyword labels, points and machine learning to formulate more harmonised lists.

You would then print off the newly generated warscroll cards and use them in just that game. It would be aimed at people who were trying to run competitive tournaments.

 

[LizardWizard]

I had this idea for a software solution for AoS and 40K's terrible balancing issues.

Basically, it's a website. What happens is you feed in your army list before the game, and your opponent does the same.

What happens then, is the software recalculates the warscrolls and adjust's their stat lines to harmonise the lists.

So say you have a cav unit, that unit would be given a label like "Fast".

The cav unit would then be given a move characteristic randomly between 7 and 9.

The software would use keyword labels, points and machine learning to formulate more harmonised lists.

You would then print off the newly generated warscroll cards and use them in just that game. It would be aimed at people who were trying to run competitive tournaments.

In my experience the competitive scene in AoS is pretty balanced. Most players at the top tables are skilled at finding the most optimal units from their battletome and matches are rarely decided before the end of turn 3 or 4.

 

[samheim]

In my experience the competitive scene in AoS is pretty balanced. Most players at the top tables are skilled at finding the most optimal units from their battletome and matches are rarely decided before the end of turn 3 or 4.

Fair point. To be honest, the whole reason I play AoS is to get away from video games and technology in general.

The only thing I use like that is the AoS list builder.

I am on the fence though, with a game as complex as AoS, maybe the rules writers should be using AI to help balance the game. AI is no big deal nowadays, and it's ideal use case is for doing something like AoS.

 

[LizardWizard]

I think there is definitely room to use meta data taken from tournaments.

 

[Canas]

Using machine learning to balance a game like AoS is (currently) a terrible idea. A balanced (and fun) game isn't a purely mathematical equation that needs to be solved and as such the current methods of machine learning are not fit. Especially not the way you are proposing. And even if you would do it properly most likely you're going to end up with a complete incoherent mess of a ruleset that isn't necesarly fun to play with as it's super difficult to represent that to an AI. Furthermore, although it seems "easy" to balance it this way cuz you got a load of numbers to tweak it'l be nigh on impossible to get the AI to properly take into account special rules and weird synergies and last but not least, it'l utterly fail at taking into account player enjoyement.

Anyway, tl;dr; it's a terrible idea to make a game with AI, most likely you'l end up either with weird stuff like a model that has a 1" base movement next to one with a 15" base movement, or with a bland homogenous mess.

Also, the point of armies in games like AoS is that they're assymetrical. Harmonizing them would defeat the point of having different armies.

 

[samheim]

Using machine learning to balance a game like AoS is (currently) a terrible idea. A balanced (and fun) game isn't a purely mathematical equation that needs to be solved and as such the current methods of machine learning are not fit. Especially not the way you are proposing. And even if you would do it properly most likely you're going to end up with a complete incoherent mess of a ruleset that isn't necesarly fun to play with as it's super difficult to represent that to an AI. Furthermore, although it seems "easy" to balance it this way cuz you got a load of numbers to tweak it'l be nigh on impossible to get the AI to properly take into account special rules and weird synergies and last but not least, it'l utterly fail at taking into account player enjoyement.

Anyway, tl;dr; it's a terrible idea to make a game with AI, most likely you'l end up either with weird stuff like a model that has a 1" base movement next to one with a 15" base movement, or with a bland homogenous mess.

Also, the point of armies in games like AoS is that they're assymetrical. Harmonizing them would defeat the point of having different armies.

Non-sense! using Domain Driven Design we can model anything in software.

Class Stormcast(effects Effects) {

private $weaponSkill;
private $attacks;

public function calculateTotalEffects() {
foreach($this->effects as $effect) {
$this->totalWeaponSkillIncreasePotential;
}
}
}

And so forth. Its totally possible to build a system that factors in all the potential synergies. In fact its trivial. I am sure Hearts of Iron 4 is a million times more complicated then AoS.

Also on the movements etc.

I said to set a range. So the movement for a "fast" troop type is toggled between 6-10 range. not 1-15 range.

The point is by harmonising lists. I don't mean to make the lists the same. I mean to balance lists using logical parameters. So it can tweak lists from old tomes to more fairly compete against new lists.

Also it will make list building more diverse, and not just limit peoples army choices to a few strong units.

 

[samheim]

The machine learning part, is to analyse the collected dated and tweak subtly the selected parameters. Not set the parameters.

I mean Totalwar models warhammer fantasy battle is software. It has all the effects and synergies.

Only that is a billion times more complicated, because it is an actual video game.

This is just some data in and data out, pretty basic stuff.

 

[samheim]

Class WarScroll;

Class Buff;

Class Spell;

Well, its pretty easy to do. You model the war scroll card, and then model all the spells.

You then have a Class Unit Balance, that takes a Class WarScroll and Class anySpellEffects or Class AnyMagicItem Effects etc.


Its not even a lot to do really. Its not hard to do from a technical perspective. Its laborious typing in all the warscroll data and relevant battle tome data. And that's about it.

A reasonable challenging problem would be workout how to balance the summoning. The buffs are easy because you can see how they modify stats in number terms.

Summoning is harder because its at first hard to grasp how to objectively define its value to the game.

 

[Aginor]

No. Absolutely not.

You can do some of the stuff you mentioned for computer games, that's correct. That's why your examples are computer games.

For a tabletop game you have to use techniques that are usable for humans.

 

[GreenyRepublic]

You can do some of the stuff you mentioned for computer games, that's correct. That's why your examples are computer games.

This, and even then the act of balancing any game of some non-trivial complexity via software/machine learning is nowhere near 'trivial' - if that was the case then there would never have been a single balancing patch for Starcraft, Hearthstone, DOTA, etc, yet there have been many years of many many balance patches for each of those games.

 

[Canas]

Let me put it this way. To even begin using AI to balance a game you'd need to make a complete simulation of that game. This goes far beyond simply putting warscrolls & abilities in a database. You also need to implement realistic player behaviour which is not something we're anywhere near capable of doing. Nor is it something we'l ever truly be capable of doing as we're not mindreaders and can't predict all the weird stuff players will come up with. You will inevitably forget to model something in a domain that complex.

No. Absolutely not.

You can do some of the stuff you mentioned for computer games, that's correct. That's why your examples are computer games.

For a tabletop game you have to use techniques that are usable for humans.

Balancing a videogame requires largely the same techniques as any other game. After all the players are still human and what's ultimatly most important is that it's fun for the players. Using algorithms to balance stuff will at best result in a bland game of nothing where everything is equal, and at worst it results in an overly complicated mess filled with anti-patterns and toxic gameplay as an AI simply doesn't care about those things.

 

[Aginor]

Balancing a videogame requires largely the same techniques as any other game. After all the players are still human and what's ultimatly most important is that it's fun for the players. Using algorithms to balance stuff will at best result in a bland game of nothing where everything is equal, and at worst it results in an overly complicated mess filled with anti-patterns and toxic gameplay as an AI simply doesn't care about those things.

Yes and no. You can get away with some stuff.
Like values not being in the 1d6, 2d6 or 3d6 range, actions requiring dozens of variables and so on. You can make it vastly more complex in the background because a computer keeps track of the rules for everyone. That's a pretty fundamental difference IMO.

 

[LizardWizard]

I agree with you guys. AI isn't there yet.

I think there is a lot of potential value in looking at the tournament scene because most of that data is already made public. Using computation to isolate most winning units, Allegiances, Artefacts, Allies, ect, could be useful in helping target areas of focus.

This data could then be aggregated to look for common trends in a units effectiveness. For example, perhaps units with a movement of 14" are vastly out preforming otherwise better stated units with a movement of 10". This could then be used to scale back that units movement instead of increasing the point cost.

However, for all I know GW already does this.

 

[samheim]

Guy's don't the worry!

I have hardcore experience of building an online game system, that was based on warhammer before.

I have fought these software dragons and won before. I say I can do it again!

What your visualising is a perfect end solution that works 100% and is perfect.

I am not visualising that at all. I am seeing a tool that warmly "harmonises" 2 factions against each other.

It will use subjective values as well as binary ones.

One thing I notice as I am just beginning to prototype this god software in my brains is this.

The points values in AoS are basically all made up bullshit.

In software I want the points to work like this.

Movement 5 = 4 points.

Movement 6 = 6 points.

Movement 10 = 15 points.

I want hard points values for everything, not just some "I feel like this is the value, therefore it is."

Then we can set a ratio so that if something has movement 10 and 5 attacks. Those attacks get progressively more expensive.

Don't try to fight against the army harmoniser, be at peace with the army harmoniser.

 

[samheim]

The machine learning part...

First of all it doesn't work like the way you think it does. You think the Ai is going to write the application.

That's not how it works.


The machine learning part, analyses the data, and lightly tweaks the varible's ranges.

Or better still.

I am going to generate a random number. On a 4 + I will give something 1 extra weapon skill.

Well the machine learning could set the 4 + to a 5 + if its deemed to be to low, based on reported data.

Everything else is down to my masterful system design. Not the AI which is probably really just there, because what the hell everybody else is doing it.

 

[LizardWizard]

The points aren't made up malarkey even now. Wounds, save, movement, ect all impact the units cost. It is why Liberators are so close in cost to Chaos Warriors. Or why all lesser demon battleline line units are all 120 per 10. These are obviously more than just random. Most of the time I can guess a units points cost within 10-20 points just based off knowing the unit size and reading the warscroll.

There are also factors that influence a unit's cost outside of the individual warscroll. Like maybe GW wants to kill a particular list. They might target an 80pt hero that is used in the list to push the point total over 2k.

So while we may not know if GW uses a hard point assigning system, it is obvious that there is some system already in place.

Edit: I did a swear. Caught it in a re-read. My bad.

 

[LizardWizard]

I have a few questions:

• How do you expect a list harmonizer program to be used?
• Does each player bring all their models, input them into the program, and the program writes a list?
• How would this work at a tournament?
• How would this work for summoning?
• Are you going to use a different list every match of the event?
• Do you input all players list at the start of the event or would this be done match by match?
• Do you feel list writing isn't a skill that needs to be showcased on an individual basis?

 

[samheim]

The points aren't made up malarkey even now. Wounds, save, movement, ect all impact the units cost. It is why Liberators are so close in cost to Chaos Warriors. Or why all lesser demon battleline line units are all 120 per 10. These are obviously more than just random. Most of the time I can guess a units points cost within 10-20 points just based off knowing the unit size and reading the warscroll.

There are also factors that influence a unit's cost outside of the individual warscroll. Like maybe GW wants to kill a particular list. They might target an 80pt hero that is used in the list to push the point total over 2k.

So while we may not know if GW uses a hard point assigning system, it is obvious that there is some system already in place.

Edit: I did a swear. Caught it in a re-read. My bad.

Yea there are a few units like you say. But what about 230 points fiends?

Its easy for someone to say, these look basically the same as the Chaos Warriors lets point them so.

But the whole point of the army harmoniser is that it will take into account the maximum potential of units with buff's and then harmonise the other list to offer a fairer challenge.

So someone would be able to field a power DoK list against someone playing 2 start collecting boxes of Slaves to Darkness.

The army harmoniser would tweak the warscroll's to balance these forces more closely without doing stupid stuff like making the Chaos Warriors move 10 or something.

Basically the idea is to remove the list building element, and make the game more about skill. You can use all your favourite models and they won't be wasted or pointless to take.

They will work in harmony with the rest of your force.

 

[LizardWizard]

Basically the idea is to remove the list building element, and make the game more about skill. You can use all your favourite models and they won't be wasted or pointless to take.

So you don't view army building as a skill?

Fiends are 210, btw. And they're comparable to Plague Drones which are 200 per 3 models. One has fly and is tougher, the other has more damage.

 

[Canas]

Yes and no. You can get away with some stuff.
Like values not being in the 1d6, 2d6 or 3d6 range, actions requiring dozens of variables and so on. You can make it vastly more complex in the background because a computer keeps track of the rules for everyone. That's a pretty fundamental difference IMO.

yeah, but ultimatly the same principles apply. I mean you can base attack values on super complicated mathematical equations if you wanted to. Especially if you keep all of that hidden in the background. The player doesn't care if the damage was calculated rolling a dice or by solving several differential equations, he cares that he did 3 damage.

Mostly the advantage of videogames is that they can take the bookkeeping away from you. Which has little to do with actual balance for the most part. The only advantage it can give is that you can buff & nerf with smaller increments, as you're not chained to say a D6. Ultimatly the principles stay the same though.

O, and of course that's ignoring mechanics that can only be done on a virtual machine (see the TGC hex, completly virtual and takes full advantage of that with its mechanics resulting in some neat things).

I agree with you guys. AI isn't there yet.

I think there is a lot of potential value in looking at the tournament scene because most of that data is already made public. Using computation to isolate most winning units, Allegiances, Artefacts, Allies, ect, could be useful in helping target areas of focus.

This data could then be aggregated to look for common trends in a units effectiveness. For example, perhaps units with a movement of 14" are vastly out preforming otherwise better stated units with a movement of 10". This could then be used to scale back that units movement instead of increasing the point cost.

However, for all I know GW already does this.

They do, to an extend. There was a white dwarf article a while back in which they explained how they determine pointvalues. It seems to be one of the few aspects they actually have company-wide guidelines for how to determine these. It's the one thing that isn't super suspectible to powercreep. The main issue is that their testing is rather limited, despite it being absolutly vital to the way they determine points. So they simply end up missing things, especially when an army can do something new and fancy...

Basically the idea is to remove the list building element, and make the game more about skill. You can use all your favourite models and they won't be wasted or pointless to take.

At that point you might as well just go and play with whatever nonsense you've made up yourself as you're no longer playing AoS.