AoS Army Harmonizer

[Canas]

I think you underestimate the power of machine learning, with the proper system in the back and "AI" can evolve quickly "skill" wise due to the fact it can learn as quick as you throw money at it (i.e. hardware). Basically as long as you just throw enough resources on it, it could play millions of games in a matter of days, and through that, you get data you can use to improve the army rule set for human players.

After all, we are not looking for someone to challenge a human player per say, but to optimize the game rules and make them as fair as possible.
Being able to run X model, Y army against Z over and over again would be faster to do with a system like this in place, than through traditional means. However, the initial cost would be massive, which is why it is not utilized.

Issue is that to make the rules fair we need to take into account how they're used. An AI's can play in super weird ways. There's some playthroughs of super mario bots where you can see this. And I think the starcraft bot had similar issues where they could make it beat a human, but it'd also be playing in an absolutly insane manner (e.g. clicking 1000's of times more often than the humans, and limiting the amount of clicks quickly made it a terrible bot). Similarly, in abstract games where the AI at first appears to be playing near randomly, only for it's strategy to become apparent 20 moves in when it wins the game seemingly out of nowhere as it had just been playing an optimal path from the start.

This all depends on what you allow the "AI" to do and know, it would be very simple to add a rule on a specific "AI" model where it randomly does a slight misplacement like a human player could do.

To an extent yes, but it'l be difficult to figure out and cover all of these.

​

In the scenario/case we are discussing, this would be a rather trivial matter compared to the other parts. These should be created similar like army rules in Lua (or another side language/format), both to remove complexity away from the main software, but also to make it possible for someone with no knowledge of the actual software to add content for the simulations.

Adding them isn't too difficult. But it does make for some wildly varying enviroments causing the attributes of a unit that matter to change drasticly. The easiest exampels are skinks; great to hold objectives, borderline useless when you need to assasinate the enemy general. It'd be interesting to try and figure out. But I wouldn't be surprised if you'd get stuck swinging back and forth between local optima due to the vast differences between scenario's.

 

[samheim]

The Red Devil has some expert knowledge, that I must admit I don't have.

I don't know anything about ML. When I said ML, I also said that I was going to use it as a buzz word because everybody else is using it at the moment and that's what seems to get funded.

To be honest I am using stone age technology. I don't even want to say what stack I am using because it's funny. Not really into development in a big way anymore.

But with that being said. My idea was to kludge it together. I have worked on some low budget online games before. And sometimes you use tricks to make people think they are playing against something more intelligent then they really are.

I think the ML option would work and probably be better, but of course its cost prohibitive. My idea is to try to make something that I can do on my own that kind of works, and we see how it goes from there.

Also a lot of the things like battle plans. I am thinking that, that could be accounted for by applying penalty to fast moving/horde armies if the game is about capturing many objectives. Or a penalty to tough armies if the game is about killing all your opponents.

I am hoping the old fashioned approach with some heavy tweaking, may possibly be able to offer something in the end not so different to a perfected ML version.

 

[The Red Devil]

Issue is that to make the rules fair we need to take into account how they're used. An AI's can play in super weird ways.

I am not disagreeing that there is a large number of challenges, but the software as described is possible with today's software and hardware possibilities. I might be looking at this from a larger scale than you, my point of view at this is from an enterprise-level and having at least several hundred thousand euros to develop the system, at this point handling the issues you describe is possible.

Though in reality, we are discussing something that will never become the reality, or if it does it will be a major scaled-down version of what we are discussing, mainly due to the market is so small that you will spend several times more money developing it than you will ever be able to earn back.

I am hoping the old fashioned approach with some heavy tweaking, may possibly be able to offer something in the end not so different to a perfected ML version.

An scaled-down version of this, that would be much simpler to implement, could be an application where you are able to pin one model group against another. At this stage to run the simulation, you only need to implement unit/model rules.

Though this would not give you a good simulation for points, it would be able to give you a good idea of what models are viable against X, and also how cost-efficient your choices are.

Note. The hardest part here would be to simulate the dice throws. Just using a cryptographically secure pseudo-random function to base this of would be incorrect, since you, in reality, do not have an equal chance to get every number on the dice.

 

[Canas]

I am not disagreeing that there is a large number of challenges, but the software as described is possible with today's software and hardware possibilities. I might be looking at this from a larger scale than you, my point of view at this is from an enterprise-level and having at least several hundred thousand euros to develop the system, at this point handling the issues you describe is possible.

Meh lemme put it this way. The biggest issue is getting humanlike bots to balance it on to ensure they don't start balancing around "weird" AI playstyles and end up stuck in a, for humans weird, local optimum. Essentially the bots need to be able to pass a turing test for games. And this is not something that has currently been done, neither in general, nor for specific games. I guess this is currently at least theoretically possible, but it's definitly a hellishly difficult problem and solving it will get you quite a lot of praise. And solving it, as well as a couple other issues related to healthy game design, is going to be needed to have any hope of making an actually fun and healthy game.

If on the other hand, you don't care how weird the rules potentially may end up then it is indeed fairly standard software. Just getting the AI to tweak stats till armies & units all have a ~50% winrate is doable if a massive undertaking. It'l probably result in an absolute mess of a game though (hell, obsessing over ~50% winrate already makes humans make questionable designchoices). But it's more than doable.

 

[Aginor]

An scaled-down version of this, that would be much simpler to implement, could be an application where you are able to pin one model group against another. At this stage to run the simulation, you only need to implement unit/model rules.

Something like that has been done, by our very own @GreenyRepublic:
http://www.lustria-online.com/threa...ry-age-of-sigmar-battlesim.19812/#post-285055

And that's where we all could see that something seemingly trivial can in fact be quite hard. It is a good start but some things that are trivial for humans are actually pretty hard to put in numbers for computers.

 

[samheim]

Something like that has been done, by our very own @GreenyRepublic:
http://www.lustria-online.com/threa...ry-age-of-sigmar-battlesim.19812/#post-285055

And that's where we all could see that something seemingly trivial can in fact be quite hard. It is a good start but some things that are trivial for humans are actually pretty hard to put in numbers for computers.

I didn't see Greey's program before. I am really impressed he had a go.

The only critique I will give him is, work on a nice user interface so that people can follow on with what you are doing. People are impatient and get easily frustrated with things that are not presented in an easy to use way. He would get a lot more people interested in his idea if they could easily understand and interact with it. That aside well done.


Yes its true things can get complicated very easily and can easily spiral out into a never ending job, that will never be finsihed.

That's why I said my idea at first is to balance the start collecting Seraphon box against The Khorne blood bound.

Also there is not one way to do things, a lot of software complexity can be tamed if you get your initial design right.

And by right I don't mean you do things the right way, I mean you work out what you can fudge and what can't be fudged.


Trying to perfectly emulate AoS when you are just starting off, is never going to work. You will die from straining your neck to look at the top of the mountain.

The trick is to try to make some good, that won't be perfect but will showcase what your idea is.

 

[samheim]

Also it's my opinion that the problem is not so big that it needs the ultimate solution to solve.

We already know, that Age of Sigmar is a good game that people like playing. The game designers know what they are doing. They know how to make a good game.

The problem we are facing, is that after you have played any large board game for a while, it starts to become more about certain stand out units or lists that seem to have an unfair advantage.

So from this we can deduce that the pointing system is less then optimal, in taking into account the different kinds of battle plans and match ups.

This even makes sense, when you consider the first version of AoS didn't have any points values at all. The original idea was just to play for fun and experience the AoS world. It works very well like that. You have a fast army beat a slow army. Whatever. Its all about narrative.

The problem occurs when people want to play the game competitively. Or they want their games to be evenly matched up for both players to have an even chance at wining the game.

As AoS is a game of dice, we could ask how important it is to play thousands of games with ML. As the dice are always going to be random.

GW at the moment does a form of machine learning, where they gather data from battles and use that to adjust armies.


My opinion is, the problem is to do with non standardised pointing and lack of compensation for mismatched match up's.


That's why I said the need to harmonise the armies in real time by offering bonuses to the weaker fellow.

Could this be carried over to non-digital format? maybe, it would mean having a stack of cards that granted attribute's to one players force when he was facing to much of a challenge from the other players army/battle plan.

 

[samheim]

In my limited understanding of ML. I think its a tool that helps to identify problems. It doesn't solve problems.

My idea for a solution is how the Stomcast Thunderbow and Great Mace do D3 hits against Nighthaunt and Chaos instead of 1 against everybody else.

Maybe in the battle tomes they could have a section where your army would have different attributes against each other faction.

So say when Beast claw raiders face Slaneesh, they get a more powerful storm that slows 1 wound infantry by 2 inch per turn, and then deals d6 mortal wounds to them, to compensate for Slaneesh bloody over powered summoning!

Conversely it would also say when Beast Claw Raiders are matched against Maggot Kin the storm is colder making the BCR have a + 1 to save or something.

----------------------------------------------------------------------

Obviously this is just a vague idea, and doesn't provide a solution to a lot of the redundancy in the game.

In an ideal world, every magic item/spell would have a purpose and a good reason for taking it sometimes. Maybe by providing different bonuses or penalty's to different match up's, it would create more incentive to take magic items that are otherwise overlooked.

 

[samheim]

So that was it. It's probably my fault for not being able to crystalize what I wanted to say in simple enough terms. Its this though.

"Provide matchup specific, buffs and debuff's".

The good news is that I don't really need to build the army harmoniser anymore. The same thing can be achived with 22 sheets of A4 paper.

I think by adding this 1 layer of detail, games can be greatly harmonised.

The only difference was that, the software I was going to make, would be able to balance points mismatched armies. But people have accurately expressed that no one would be interested in playing home brew rules.

 

[Dan32234234234324]

I would still love to see the software as I think it would be really cool, and you could use it to highlight statistical imbalances between units.

It would be compelling to be able to show that Unit A is mathematically overpriced compared to Unit B, and that could spawn some really interesting discussions in the AoS world!

Side note, I prefer your now-pleasant demeanor compared to the way you behaved before. Thank you for turning it around

 

[Erta Wanderer]

I would still love to see the software as I think it would be really cool, and you could use it to highlight statistical imbalances between units.

It would be compelling to be able to show that Unit A is mathematically overpriced compared to Unit B, and that could spawn some really interesting discussions in the AoS world!

well it's not automated and it takes a bit fo personal leg work but i use
http://tools.druchii.net/AoS-Combat-Calculator.php

 

[samheim]

I would still love to see the software as I think it would be really cool, and you could use it to highlight statistical imbalances between units.

It would be compelling to be able to show that Unit A is mathematically overpriced compared to Unit B, and that could spawn some really interesting discussions in the AoS world!

Side note, I prefer your now-pleasant demeanor compared to the way you behaved before. Thank you for turning it around

I should probably do it, because it's something that will make me want to program. As I am older finding the motivation to sit in front of the screen for long hours, trying to fix small errors when your brain went to soup 5 hours ago. Its really hard.

I will try to make a start.

Another point was, that I foresaw as other's did how hard it would be to point for some of the abilities. Anything is possible with software. If you told people when they were playing Mario Bros, one day Star Citizen would be a game they could play. It would be impossible for them to believe.

Anything is possible, but so is climbing Mount Everest. You still have to go out and climb the mountain.

I think we can work out how many square inch's are on a table. 4 x 6 feet.

So that would be a total of 288 square inch's.

Somehow this helps to work out the effectiveness of the abilities radius of effect. But I don't know how yet.

So to me that is a big problem to work out how to do that. It would take some time to figure that all out.

Actually building a pointing system, is probably not that difficult in comparison to factoring in these things.

And it might even/probably is just some mathematical equation you can do to work out all this stuff for you, without needing to build complex models.

Then there is the other problem, this is something where you could get deep into it, and then be faced with another big unforeseen problem, that you had no idea about when you began.

At the moment, the time I would be willing to invest, would probably take me up to some basic units and basic abilities. The Old Blood's snap to it ability to me at the moment, looks like it could take more time to work out then the rest of the application.

That said there is probably someone out there, who could do it in 5 min, because they already have experience of solving those kind of problems.

 

[samheim]

COMMAND ABILITY Paragon of Order: The Saurus Oldblood gives an unspoken command, causing its warriors to snap into a new formation. If a Saurus Oldblood uses this ability, each Seraphon unit from your army within 10" can immediately reform around one of its models. That model must stay where it is, but each other model in the unit can move up to 3" so long as it does not end this move within 3" of the enemy

---------------------------------------------

Even that's hard, they had to make it within and not wholly within.

So how would you point this ability?

My only guess at the moment, is that you have a separate points cost for out of movement phase movement.

Each unit, that means really your whole army could re position.

That's sort of why I don't want to start typing code. In real life AoS, how are they working out the points value for this ability?

Should it be in the game if no one understands the value of it's effectiveness?

If it was just one unit, you could apply a different points cost for out of phase movement.

Dice are random, but you know they go from 1-6. This is just random, if you split it in the middle and charged movement for half the points value of the army, it would probably be over priced. There probably is no right cost for this ability.

The question is, how many of these extremely difficult or impossible to point abilities are in AoS?

And really you need to have a precise system for pointing values. The game would still be hard to balance, because different factions have different abilities that don't match up well.

But when you move into the realm of making things with a subjective value, rather then a value based on anything else, all your solutions will just need to be things that make the game subjectively feel more fair.

And that's what people want from any AoS software calculator. Precise points values. Which until you solve this problem, is impossible to do.

 

[Erta Wanderer]

COMMAND ABILITY Paragon of Order: The Saurus Oldblood gives an unspoken command, causing its warriors to snap into a new formation. If a Saurus Oldblood uses this ability, each Seraphon unit from your army within 10" can immediately reform around one of its models. That model must stay where it is, but each other model in the unit can move up to 3" so long as it does not end this move within 3" of the enemy

---------------------------------------------

Even that's hard, they had to make it within and not wholly within.

So how would you point this ability?

My only guess at the moment, is that you have a separate points cost for out of movement phase movement.

Each unit, that means really your whole army could re position.

That's sort of why I don't want to start typing code. In real life AoS, how are they working out the points value for this ability?

Should it be in the game if no one understands the value of it's effectiveness?

If it was just one unit, you could apply a different points cost for out of phase movement.

Dice are random, but you know they go from 1-6. This is just random, if you split it in the middle and charged movement for half the points value of the army, it would probably be over priced. There probably is no right cost for this ability.

The question is, how many of these extremely difficult or impossible to point abilities are in AoS?

And really you need to have a precise system for pointing values. The game would still be hard to balance, because different factions have different abilities that don't match up well.

But when you move into the realm of making things with a subjective value, rather then a value based on anything else, all your solutions will just need to be things that make the game subjectively feel more fair.

And that's what people want from any AoS software calculator. Precise points values. Which until you solve this problem, is impossible to do.

in the current aos this ability is worth about 20 points when compared to other abilities that do simelur things and units that are about the same and scaled for price inflation.

 

[samheim]

Maybe charge points for each inch of the abilities range x the amount of movement it offers.

So 10 inchs x 3 x what ever points cost.

 

[samheim]

+ 10 points tax for within, not wholly within.

 

[Erta Wanderer]

Maybe charge points for each inch of the abilities range x the amount of movement it offers.

So 10 inchs x 3 x what ever points cost.

your thinking about this the wrong way you have to see what it does not how much 3" is about the avrage you would get if you ran so this ability alows you to spend a command point to allow a bunch of units to run and charge with some clunky wording. and better consistency

 

[samheim]

your thinking about this the wrong way you have to see what it does not how much 3" is about the avrage you would get if you ran so this ability alows you to spend a command point to allow a bunch of units to run and charge with some clunky wording. and better consistency

Sure,

As we are planning software though, we want to have solid values. We want to price for movement and range, things like that. So we can apply the same cost to everything, over the entire game.

So we are looking for ways you would point something like that objectively.

I understand when you say how powerful it is, and that makes it more difficult, because its becomes a subjective measurement, we can't can't apply a logical value to in software.

Also I was thinking, 8 inch radius covers less square inch radius then 10 inch radius.

So maybe it would be, better to point for the total amount of square inchs covered by the radius of the ability.

I am definatly open to any suggestions on this, as you say, its very powerful, how do be objectively quantify that power though?

 

[Erta Wanderer]

Sure,

As we are planning software though, we want to have solid values. We want to price for movement and range, things like that. So we can apply the same cost to everything, over the entire game.

So we are looking for ways you would point something like that objectively.

I understand when you say how powerful it is, and that makes it more difficult, because its becomes a subjective measurement, we can't can't apply a logical value to in software.

Also I was thinking, 8 inch radius covers less square inch radius then 10 inch radius.

So maybe it would be, better to point for the total amount of square inchs covered by the radius of the ability.

I am definatly open to any suggestions on this, as you say, its very powerful, how do be objectively quantify that power though?

right witch is why we divide movemet abilities in to categories =1 to run and charge, +2" movement, it is rairly more then this if it is a strait buff the few that are are 4", run and charge, retreat and charge, and teleport. we don't base it of of " we bais it of of cattagory
if you are talking radios they are divided into 8" 10"(for old abilities not used any more) 12" 18" 14" 32" and a few skaven ones that are 13 for flavor just sort into range category and point appropriately

 

[samheim]

right witch is why we divide movemet abilities in to categories =1 to run and charge, +2" movement, it is rairly more then this if it is a strait buff the few that are are 4", run and charge, retreat and charge, and teleport. we don't base it of of " we bais it of of cattagory
if you are talking radios they are divided into 8" 10"(for old abilities not used any more) 12" 18" 14" 32" and a few skaven ones that are 13 for flavor just sort into range category and point appropriately

Yes, sure.

I was thinking like this. And I could be wrong.

We want to work out the points value of the smallest unit. So for movement,

1 inch of movement costs 5 points (for example)

With a spell, we want to have a points cost for the range of the spell, the amount of damage it can do. The type of damage it does, and the radius it covers.

Its true they do have categories they are all fixed. So maybe that's ok to point for just the radius types.

This is what I mean though, you want a formula that never changes. The formula always stays the same.

So this is what I think. Someone has made a subjective judgement on the spells. Maybe they said, this big powerful spell can only have a short range, because its big and powerful. Then they thought, well I think a good points value for this is 25 points.

The problem is then, someone else from a different army has said, this spell is big and powerful, I think I make it medium range, and price it 20 points.

If you have 1 formula, then you can say, this spell costs, y amount for this range, it does this much damage which will cost z amount. The total cost is etc.

I am not saying that GW does not have a formula. I would say, from my observations I don't really think they do. I think its based on how they feel it is. At that given time.

That's why I don't think its possible to brake down the points costs of the models. I think the only way is to try to build up the points cost of the models, as there was never a cohesive system for pointing models in the first place.

Not saying I right, maybe they do have a formula and they are making all their pointing decisions based on that. I do find that really hard to believe though.