Warzone 2100 Forums

I agree with that ! Didn't your patch was about the targeting AI more than the "shoot behind" bug. Did you solved that one too ?

raycast wrote:it raises the need for rebalancing

Again, this weird idea that balancing at large scale is even possible...

I'm sorry but any change, as small as it can be, "echo" on any research that lead to it AND any research that come from it AND every research affected by the prior or the later. In other word, the subtlest change affect at the least 1/3 of the tech tree. On long term, the only way to "absorb" the change and keep the balance is to change the very form of the tech tree.

Assuming someone knows how to do it (because I don't), applying a global change of that scale would requires hundreds of very specific move in the branches of the tech tree.

I'm just saying, because people tend to believe the balance is some easy action that require no time. But actually, it's about 1/3 of all the development effort.

I'd prefer to be greatful for having theory on subtle stuff that is not easy to notice instantly but may break things in the long run.

You know, we're not talking about stuff that can be tested by just opening the game and killing a few AI tanks in 10 minutes. Observing our community, i have to admit that balance is something that takes many months and often years to explore; and it's months and years of the whole community, not two or three players who are clever enough to apply the patch, or even a dozen of players who can load a mod. That's why we absolutely need to theoretize.

I'd also not complain on the speed of patch review. I'm having a very curious 3-months-old patch in the tracker now that is still on review and i see nothing wrong with it; i just let *them* decide on priorities.

The only thing i see there is that this thread is taking too long, so people become impatient.

As obviously our current balance is already broken, we will need to rebalance anyway.

E.g. flamers are too strong, yet also behave irritatingly because they randomly fire 1 tile off, even at short distance (= 45 degrees off!). Artillery overkilling is another example.

Since we need to fix the balance anyway, we may as well try to do a couple of changes that will also affect balance, and then fix it just once, instead of over and over again.

Plus, I don't see any proof of your claim that this means redoing the tech tree. A lot of balance changes can actually be done on a micro level such as reducing the initial accuracy and damage of machine guns and flamers, for example. Plus, it is already broken nevertheless...

So IMHO, we should try to make all the changes that will require rebalancing sooner than later, then do the rebalancing. Maybe also pay some effort into tools for rebalancing, such as e.g. collecting statistics such as damage-dealt vs. damage-taken vs. cost for different units across a game. Or maybe actually try the free-market cost model for some time (we can use this for beta versions only!) to find a good balanced static cost set for the next release.

That's why we absolutely need to theoretize.

No. On the contrary. The point is: our system is too complex for us to be able to find a complete theory. The models we have discussed here were WAY too simple, and useless as such. So instead, we should A) start with our own intuitions, because the human intuition is often quite good at finding tradeoffs and B) try to employ some rapid-turnaround incremental approach to fine tune balancing (could be market, but also just frequent beta releases along with a feedback form and maybe statistics on the popularity of weapon combinations).

Either way, not doing anything clearly is not an option. We have pending gameplay changes that we need to try, refine and deploy; and then we need to fix balance again, but also find a way to perform future balance corrections easily.

The 3.2 release round will require quite a lot of play-testing of balance. There is already the $100 cost for derricks and increased cost of power generators. I will be pushing an accuracy change soon. (As soon as I've finished the upgrades stuff for javascript - that stuff took a long time to come to fruition... And once you see what can be done with the upgrades changes, I am sure some interesting suggestions will be made.) We will probably need to do something with the targeting code (that raycast has been working on). Oh, and the really cool ECM tech is now added in 3.2, which might whack medium to late game balance all over the place.

Thanks, Per. Good examples on why we need to rebalance anyway.

Do you remember the sonic tank in Dune II? That could be an interesting addition to the tech tree,too: sonic/shockwave weapons for late game? They also had a super heavy slow tank, which would cause an explosion when destroyed. Plus, I would love to see radar jammers (e.g. to protect artillery batteries).
Plus, we also should make commanders more useful for multiplayer. Right now they AFAICT die too easily to become useful.
In some other post I shared some more ideas such as shield generators that could provide some extra cover to nearby units, comparable to having some repair trucks in the second line.

If we ever want to add any of this, we'll need a streamlined process of rebalancing the game, such as having balancing servers that keep track of statistics and gradually change balance; which is loaded by the users on connect. If we reach a stable balance, we can then use this as default for the releases.

raycast wrote:No. On the contrary. The point is: our system is too complex for us to be able to find a complete theory. The models we have discussed here were WAY too simple, and useless as such. So instead, we should A) start with our own intuitions, because the human intuition is often quite good at finding tradeoffs and B) try to employ some rapid-turnaround incremental approach to fine tune balancing (could be market, but also just frequent beta releases along with a feedback form and maybe statistics on the popularity of weapon combinations).

Nope. You can't have a balance out of just trial and error.

If you have some experience in balancing, you would easily see that some games were designed with balance in mind, while some others weren't. I can mention Battle for Wesnoth as an example of a well-designed game; with enough math put into the early development, and providing simple yet sensible autobalance techniques, now they enjoy having as many as 6 fairly balanced and completely unique factions, none of which has any useless redundant units and needs to carefully consider terrain, enemy choices and even time of day mechanics. This is not something that is achieved by simply tweaking stuff until it gets balanced. You need to be sure it's balance-able before you start. This thread is not about balance, but about balancability.

I think both are right. You need theory to have a place to start, and you need practice to see if your assumptions are correct. Often some things that seem really clever in theory turn out to be awful in practice. But without theory giving you a solid foundation to practice on, you'll get nowhere with just trial and error.

Actually, wesnoth is in some points simpler than WZ. First of all, it is turn based, then it is tile based, too.

Yet, to quote from the wesnoth forums:

There is no way to just have balancing done by some magic formula. This simply does not work due to too many factors being involved.

From there I also found the link to these slides about balancing OSS games:

https://archive.fosdem.org/2012/schedul ... Slides.pdf

I do agree that we need to have some theory and some plan of how balance should be like. For example, we may want to give certain weapons a slight early game advantage, but make them a dead end on the long run. Because that makes gameplay interesting. There should be a chance of winning with an early MG rush, but it shouldn't be a sure thing against players prepared for this. Which is why I believe in "crowdsourcing" balance, improving over a manual design (which should involve some theory).

And last but not least, we can't just do random changes and see which one "improves" balance (how ever we would measure this) - trial and error indeed will not work well, and take a long time. Instead, we should maybe try to work on a theory of automatic balancing, i.e. how to setup experiments and incorporate collected feedback to improve balance; or even just to produce balance *suggestions* (and then review them manually).

raycast wrote:Actually, wesnoth is in some points simpler than WZ. First of all, it is turn based, then it is tile based, too.

Yet, to quote from the wesnoth forums:

There is no way to just have balancing done by some magic formula. This simply does not work due to too many factors being involved.

From there I also found the link to these slides about balancing OSS games:

https://archive.fosdem.org/2012/schedul ... Slides.pdf

I do agree that we need to have some theory and some plan of how balance should be like. For example, we may want to give certain weapons a slight early game advantage, but make them a dead end on the long run. Because that makes gameplay interesting. There should be a chance of winning with an early MG rush, but it shouldn't be a sure thing against players prepared for this. Which is why I believe in "crowdsourcing" balance, improving over a manual design (which should involve some theory).

And last but not least, we can't just do random changes and see which one "improves" balance (how ever we would measure this) - trial and error indeed will not work well, and take a long time. Instead, we should maybe try to work on a theory of automatic balancing, i.e. how to setup experiments and incorporate collected feedback to improve balance; or even just to produce balance *suggestions* (and then review them manually).

You don't understand what NoQ just said. It's about balanceability. If you guys push a realistic accuracy model, the mean deviation won't be close to the mean accuracy. It's like trying to balance a marble on top of a windy peak. If we both know it gonna fall but can't predict on which side, we can't balance the game accordingly.

In order to balance the weapon we NEED a general idea of the relative accuracy of the weapon compared to the others. That will fluctuate wildly in the model you guys proposed and voted. The relative accuracy wont be stable and therefore the balance point wont even exist.

I understand that you believe we can't reduce all to mathematical models, but we surely can for THAT particular part of the problem and the result is that the resulting balance point require both players struggle to keep all their army in a 1 tile wide area. If, at any point in the game, one fail to do that, the lost will be incredible for him. Somewhere in the magnitude of losing 33% of his asset per minutes.

I strongly doubt that anybody will find that situation fun.

Iluvalar wrote:In order to balance the weapon we NEED a general idea of the relative accuracy of the weapon compared to the others. That will fluctuate wildly in the model you guys proposed and voted.

You keep ignoring that I tested the accuracy of the model and found it to be stable. I found no wild fluctuation. For all intents and purposes, using physics based only for direct fire weapons means that for approx 90-95% of the time hits will be automatic, and I did not find any strategy you can use to vary that percentage in any meaningful way. Hits are just too fast to dodge.

So why don't you explain or show me where it breaks down and becomes random? Then I can test it again, and see if you have a point. Better yet, why don't you test it yourself?

I already sent that but anyway...

Here you see a graphical, scale accurate picture of the problem.

Situation A - Enemies are far enough, both weapons accuracy matter. Here, the comparative accuracy of both weapon make one player hit 50% more. (150% total). And, of course, it would require us to take that comparative accuracy in account in balance.

Situation B - Both weapons are now close enough that they hit 100% of their shots. Balance must NOT take accuracy into account.

Situation C - Somewhere between A and B and fluctuating in 1 single tile between 100% to 150%.

If we chose situation A or B one of the 2 weapon will be balance on it best scenario and will suck in the other. Which will make all weapons except the base one unusable.

The only way we can balance that is by taking the mean of the 2 extreme and assuming (quite stupidly) that both players will work together into staying flat on that equilibrium line.

What Per is trying to say is that enemies are never far enough (just due to limited weapon range). We, sort of, are always having Situation B or something very close to it.

@Iluvalar I do not understand you either. I do not see any instability, which would prevent balancing. All I can see (in your second post) is that the game is too complicated for your theoretical model. Which is a failure of your theoretical model for balancing, not of the actual game. Yes, your model needs to take distance and accuracy into account.

And yes, it is realistic that weapons are more accurate at short distance than at long distance, which will likely show up in a physics based model automatically. As of now, the physics based models all will give you near 100% accuracy at a short range. The old code didn't - it would allow flamers to fire 45° off at a chance to enforce low accuracy at 1 tile range, too. Which looks really odd in gameplay, when a bunch of flamer cyborgs doesn't manage to hit a wall, but fires flames all over the place... Yet, at long range, in current code, the shots will still be exactly 1 tile off, even when you miss. So you can't really miss a factory (we're talking 1 tile from the center!), and you can't really miss a battle group when you aim at the central unit.

When switching to a physics based model (or actually anything except the absurd 1-tile-off code we currently have) we will need to rethink accuracy and the effects of accuracy upgrades.

Which is actually why I suggested we could also change accuracy from "shooting accuracy" to "chance to do damage on impact".
So at accuracy 50%, 50% of on-target shots of a MG (which is pretty much 100% at short range!) would not do damage, despite being a hit. Just because, say, they get reflected of some metallic part of the target. It's actually realistic that not every MG bullet pierces the armor the same way. If you fire a MG at a tank, how many shots do you think do actual damage, and how many get deflected in one way or another?

So I suggest that we break accuracy actually into two parts:

• A chance-to-damage ratio, giving how many shots on target actually do damage.
• A standard deviation (in tiles) at maximum weapon range (which will be linear with actual range! - we may want to use the "base maximum range", to have range upgrades not reduce this precision!)

The first value, we may want to keep high (say 90%+) for long range artillery weapons with area effects. It's the "dud ratio", and we should maybe not display an explosion on impact. It's not as if a tank or building can deflect a missile explosion. For MGs and short range, this will be the most interesting value for weapon upgrades.
The second value is what one would think of as "accuracy" for long ranged weapons. It's how good the weapons aim, an expected deviation from the target. At long range, targets may move - so the shot may still hit the target, it's not guaranteed to miss 50% of shots if we have a standard deviation of 1 tile. For short range weapons, we should keep this value low. It's hard to fire a MG 1 tile off at 8 tiles range, that is like 10° off or so.

If you want to mathematically estimate the total chance - it's something like (1-cdf(stddev*distance/maxdist))*chance-to-damage, where cdf is the CDF of the squared uniform distribution.

I have refreshed my patch (but which just modifies the shot deviation, largely interesting for artillery) to the recent changes in the combat.cpp file, see http://developer.wz2100.net/ticket/3989
but I did not yet implement this two-accuracies approach.

NoQ wrote:What Per is trying to say is that enemies are never far enough (just due to limited weapon range). We, sort of, are always having Situation B or something very close to it.

If that's the case, this mean he implemented the "startcraft" option instead while pretending it's the other one. Btw. "close to" make no sense at all here. With 50% accuracy at max range, assuming the unit have a radius of 1 this mean all the bullet will fall in a radius of ~1.41 tile around it. Which look close, but still miss half of the hit.

Raycast, you might be interested to read Deflective armour, or why the accuracy is working like this.. But this is not an option in this post, so you are going of topic a little.

NoQ wrote: What Per is trying to say is that enemies are never far enough (just due to limited weapon range).

It kind of makes me wonder what a mod would be like where you have Per's simulated physics combined with a much less limited range for direct fire weapons and fog of war. I imagine preferred weapons, units and tactics would be very different between the open and hilly parts of a map.

Iluvalar wrote: If that's the case, this mean he implemented the "startcraft" option instead while pretending it's the other one.

Simulated physics without accuracy is not the same as starcraft. Because those 5-10% of situations where misses both visibly should happen and actually do happen keep the player's suspension of disbelief intact. With Starcraft, weapons would do strange things fairly often, like a Yamato blast tracking and hitting a retreating Scout at three times its maximum range.