Could we make T4 dwelling and summon only please ?

[Bob5]

If you're just looking at present board states you're missing all information from prior turns. Unlike games like Chess or Go where you can see everything at all times, Age of Wonders needs scouting, it needs players to remember where they saw the enemy five turns ago, how big their armies were, what kind of units and with what movement options, and so on. It can only go on what it sees and what it has seen in the past, which adds an extra dimension of complexity to the game. The input is a lot larger than just the current board state. It's the combination of all board states in the previous turns it has seen.

 

[Jolly Joker]

The discussion doesn't lead anywhere. Fantasizing about neural network AIs isn't going to help.

AoW 3 AI is fine - with a couple of points to improve.

1) Movement routines; very important, the AI makes itself too vulnerable IN TRANSIT, it is "sloppy" how it ends its stacks movement and prone to let itself attack piecemeal.
2) AI players needs to be more customizable with a view on setup. Starting conditions shouild be customizable for each AI/player SEPARATELY, so that you might be able to set a game, where opponents start with a town and a village and the player with an outpost (for example).

For the issue of this thread, I suppose this will already be solved by command points. In AoW 3 every stack has a leader, which is the unit with the highest leader "rating" (which in turn depends on the price of the unit. You only need to transfer this value into a command point value and assign each unit a command point value, if it is lead by another unit.

 

[Gloweye]

Also, every turn is unique. Every game is on another map. Every map is unique. AI's learn by watching results of multiple choices and then grading them on which are the "best". But in AoW, since every turn is unique, there is only once choice it can see.

There's a ton of reasons AI can't work for AoW. And someone who does AI for a living would know that.

 

[BloodyBattleBrain]

So what is the current system used for the AI?

And thoughts about my T4 musings?

 

[Gloweye]

IM guessing same as AoW3 - but probably improved.

 

[The Mentat]

I think the only realistic way to train an AI for a game like AoW would be to create first a conventional AI and let it then play against the learning AI. So basically whole games would be played similar to the auto comabat in AoW just with different AIs.


Btw there seems to be some progress with a trained AI for Starcraft 2, its still not as good as the regular AI though:

 

[Jolly Joker]

So what is the current system used for the AI?

And thoughts about my T4 musings?

Don't lead anywhere. In a system where you have to put an effort into getting something what you get must be worth the effort. A higher tier must be worth researching and building otherwise it won't see use. I also don't see any sense in adding "limiters" like the need for a special resource, because it just tends to reduce the number of real decisions you have (no resources - why research tech?).

AoW 3 favors numbers because numbers mean action points and action points are tectical options. So, say a T4/Hero has 80 Command points, and the command value of a unit is given via AoW3 upkeep cost. So 80 CP means, you can lead 2 T4s and 1 T3, which means you have 3 T4s and 1 T3 in the stack. Another option would be 1 T4, 2 T3s and 2 T2s, having a 2-2-2 formation.
The system also allows to
1) Give lower tier units high command value
2) Add command point costs for same units in a stack (2 Dragons in the same stack get alpha male problems with each other)

Also, the game may feature "combined arms boni" - stack boni (like +1 melee strength for every unit in the stack) that are applied due to combined arms, giving mixed stacks an edge.

 

[Nerdfish]

If you're just looking at present board states you're missing all information from prior turns. Unlike games like Chess or Go where you can see everything at all times, Age of Wonders needs scouting, it needs players to remember where they saw the enemy five turns ago, how big their armies were, what kind of units and with what movement options, and so on. It can only go on what it sees and what it has seen in the past, which adds an extra dimension of complexity to the game. The input is a lot larger than just the current board state. It's the combination of all board states in the previous turns it has seen.

Not quite.
Scouting information is not valid for a long time. Rarely is the position of enemy units 3 turns ago of any use. In the simplest case, it only takes a one layer LSTM to keep track of the LAST known location of enemy units, regardless how long ago they were seen, and treat turns independently otherwise. This increases training time but not amount of data required.

 

[Leyrann]

Not quite.
Scouting information is not valid for a long time. Rarely is the position of enemy units 3 turns ago of any use. In the simplest case, it only takes a one layer LSTM to keep track of the LAST known location of enemy units, regardless how long ago they were seen, and treat turns independently otherwise. This increases training time but not amount of data required.

You know, if they don't manage to create a learning AI better than a conventional AI for a powerhouse like Starcraft, then for Age of Wonders it's certainly not a good idea.

1. Starcraft does not have the complexity of Age of Wonders simply by virtue of turn-based allowing for more complexity.
2. Starcraft has much, much, muuuuuuuuuch more money behind it than Age of Wonders.

 

[Nerdfish]

You know, if they don't manage to create a learning AI better than a conventional AI for a powerhouse like Starcraft, then for Age of Wonders it's certainly not a good idea.

1. Starcraft does not have the complexity of Age of Wonders simply by virtue of turn-based allowing for more complexity.
2. Starcraft has much, much, muuuuuuuuuch more money behind it than Age of Wonders.

Playing better than a hard-coded AI is trivial.
The Starcraft learning platform exist because Google want to limit the AI's APM.
It's pretty easy to have an AI that micro so well that'd it'd win with nothing but zerglings but there is no challenge in that.

 

[Leyrann]

Playing better than a hard-coded AI is trivial.
The Starcraft learning platform exist because Google want to limit the AI's APM.
It's pretty easy to have an AI that micro so well that'd it'd win with nothing but zerglings but there is no challenge in that.

What are you even talking about? I'll admit I haven't watched the entire 24 minute video myself, but if that video indeed says that the trained AI for Starcraft 2 is worse than the base game's AI, then it's most definitely not trivial.

Start making sense, please, this discussing is getting very tiring.

 

[Gloweye]

Eh, it's not worth it. He's got an idea in his head and won't listen to reality.

 

[Nerdfish]

What are you even talking about? I'll admit I haven't watched the entire 24 minute video myself, but if that video indeed says that the trained AI for Starcraft 2 is worse than the base game's AI, then it's most definitely not trivial.

Start making sense, please, this discussing is getting very tiring.

The Speed and awareness of the NN AI for starcraft is constrained in order to make it a fair challenge.
For instance, the NN AI does not have direct access to the game state (the hard coded AI does), it must perform image analysis of a screen cap in order to obtain the information. it must scroll around the map like a human player, etc.
The goal was force the NN to play like a human, rather than to play at thousands of APM that it's really capable of, so it can't kill 20 siege tanks with 20 lings by splitting off individual zerglings targeted by the siege tanks. or kill 40 stalkers with 4 thors and medivac by picking up the thors while the stalker attacks to make them miss.
If you remove the speed governor on a AI in a real time game, you get a micro bot that does not need strategy because it out-micro humans to the extend where strategy don't matter.
And that doesn't make the game a very good AI testing platform.

 

[Bob5]

Regardless, not very relevant as we're talking turn-based games here, not real-time based.

 

[Nerdfish]

Regardless, not very relevant as we're talking turn-based games here, not real-time based.

As you say, the Starcraft AI testbeds are not very relevant comparison to a potential adaptive AI in AOW.

 

[sikbok]

Hi everyone,


On a NN based AI. Please keep the discussion civil.
Its a pretty complex topic, so while it boils down to a 'will this work now [true/false]' we need a whole lot of text more to get to the reasons why it will, or won't.

If someone starts a new topic on this I'll chime in there too.
I made some NNs before I got into game development, so I hope I can add to the discussion.


On T4s - and back on topic - I know this was a thing when AoW3 had just launched.
However we did address this in subsequent patches, so I wonder if this is still a thing that happens in the current game.
Has anyone run into this issue recently? If so, can you tell me a bit about the circumstances (number of turns played, number of players left, are those players humans or AI, etc.)?

Then for Planetfall. We changed a couple of things - not unit tiers though - that will have quite an impact on how the game plays.
While not directly addressing T4 amounts, I wonder if this will actually still be an issue in Planetfall.

 

[AwesomeLion]

On T4s - and back on topic - I know this was a thing when AoW3 had just launched.
However we did address this in subsequent patches, so I wonder if this is still a thing that happens in the current game.
Has anyone run into this issue recently? If so, can you tell me a bit about the circumstances (number of turns played, number of players left, are those players humans or AI, etc.)?

Then for Planetfall. We changed a couple of things - not unit tiers though - that will have quite an impact on how the game plays.
While not directly addressing T4 amounts, I wonder if this will actually still be an issue in Planetfall.

Maybe I am misreading this, but if I understand you correctly you are looking to have a "soft" cap for T4? By soft cap I mean a cap that limits initial T4 spam, but the cap itself can increase as the game goes on.

Analyzing the screenshots we have seen I assumed the last "resource" was a T4 cap "resource". Ref:

Speculating now but I imagine the resource being from left to right: Energy, Ore, Research, Influence, Morale/Happiness, Operation/Command points, Tier 4 cap. Personally I am all for having some sort of limiter on T4 units as they have a tendency to get spammed and that super awesome T4 unit you were excited about suddenly becomes less and less special and everything below T4 becomes inferior. I remember seeing T4 spams all the time in early AoW3, but this got gradually better and better with updates which was great.

I look very much forward to seeing more of this game - as it has become my number 1 most expected game <3

 

[Jolly Joker]

I'd be very much against such a cap because - as mentioned - they tend to take decisions out of your hands. In the worst case scenario - why bother with even researching/building when you can't guarantee a supply that would justify it.

I can only speak for AoW 3, not for planetfall, but T4s are NOT all-conquering; when they are produced/summoned there isn't that much XP to be gained, except on very rich very large maps, and a Champion 7 Troll you nursed right from charming it in turn 10 will beat every T4 you just produced.
Mass-producing T4s in AoW 3 is neither possible nor advisable, since it's much better and practical to mix stacks, especially since MCUs and RG tend to give lower-tier units a lot of oomph - plus, you NEED support units anyway.

If you can mass-produce T4s and spam them all over the place, you are playing with the wrong settings/mods.

 

[Leyrann]

I would be very much against any kind of artificial cap. There are enough gameplay mechanisms to limit tier 4 spam without it:

- Upkeep becomes too high.
- Lower tier units can be upgraded or gain experience until they can compete at least to a certain degree.
- Cost can be made expensive enough that you just want a few.
- Tier 4 units get stronger from non-tier 4 units, like the Shrine of Smiting in AoW3.
- Tier 4 units can give an aura bonus to the entire stack that doesn't stack; so after the first every Tier 4 unit wastes it's aura and therefore part of it's power.
- Tier 4 units can be made to fulfill a specific niche, while sucking if you spam them, like transport units in earlier AoWs or a unit that requires setup or protection to work properly.

And I could probably think of a bunch more options.

 

[AwesomeLion]

Fair enough, you both make good points - I'd be okay either way they go with it =)