Accuracy straw poll

[Per]

Do not vote unless you have read viewtopic.php?f=42&t=8964.

[Originway]

Do not change anything / I do not understand this poll / No opinion
should be split since the first two mean two totally different things

[Iluvalar]

Changed my mind and voted for real physic stuff.

Seems like you devs want to waste time on that epic fail idea of yours. I won't stop you.

[aubergine]

I don't fully understand how the different methods would affect things, but I can say that the Starcraft "always hit" would be horrible as it would remove some of the gameplay dynamics.

[NoQ]

We absolutely need inaccuracy to decrease the importance of fire concentration. It has already gone "a bit too high" with the invention of the new netcode (that is, now it actually works on both computers). With inaccurate fire the middle unit only receives part of the enemy army's fire.

By "too high" i mean that anybody who doesn't use it already dies horribly. No need to increase it further, i think.

P.S. Voting on balance? Bad idea.

[Nameless]

Stupid forum was stupid. Something bugged out and I forgot what I was going to say earlier..

Personally, I voted the starcraft option even though as aubergine said it would remove some of the gameplay dynamics but I'd halfta agree with Per more.

[raycast]

Can't we use a accuracy definition that takes range into account?
I.e. accuracy "1 at 10" means that the shot will have an aiming (!) standard deviation of 1 tile at a distance of 10.
After the shot has been fired, it is up to the physics to deliver.

i.e. Gaussian error on aiming, physics only after shot was fired?

One thing I'm still a bit undecided with is movement prediction. As of now, IIRC the code just assumes that a unit will continue to move linearly with the current velocity. However, for short range this does not make much of a difference, while for long range it IMHO is off most of the time (unless you play boring NTW maps). Because sooner or later there will be a feature on the units path, which will thus change. It does not make sense for artillery to fire at a mountain, just because the unit was moving that direction.
On contrary: more often than not, there will be a bunch of units moving. By firing at the position where the unit was - even more so with incidentary mortars and such - you get a good chance of hitting other units that follow. By setting up some mortars you could this way bombard the exit path of a base quite nicely, and force the other player to do some "batch exits", i.e. collect some units, wait for the exit being clear, then head out.
I don't have an idea for the ultimate solution yet. The attacker shouldn't know the path found by the moving unit. Maybe a flood fill path find in the movement direction is the way to go. But maybe it's just tracing the linear movement path to the first obstacle, then aiming at 1 tile before the first obstacle in current movement direction...

[Per]

Seems like you devs want to waste time on that epic fail idea of yours.

Which idea is that?

[raycast]

Attached is a patch that will change the accuracy substantially. Note: this will really need rebalancing.

Previously, a missed should would always miss by "2 * (100 - resultHitChance) + minOffset" units. Assuming an artillery with 50% accuracy, this would be 105 game units, almost one tile. This value only changes with the weapon specification. So for 50%, you would see 50% that are dead on target, and 50% that are exactly 105 units off.

Attached patch will vary the missing radius much more, via a random offset and by taking the distance into account.

With 50% accuracy, a shot will still be perfect with 50%.
25% of the shots will be 3 tiles off target at a range of 10, very roughly distributed like a gaussian (actually Uniform^2 -- Gaussian like in the sense that there are rather many low deviations and fewer large deviations, but bounded). They will be a rare maximum off of ~12 tiles.

At a range of 5, 75% will be <1.5 tiles off, with a maximum of 6. At a range of 1, the Q75 is then just .3 tiles.

A ripple rocket has an initial accuracy of 80%. At maximum range 86, this means that 80% of shots will still be full on target (if it doesn't move). 90% of shots will be less than 5 tiles off, 95% of shots less than 11 tiles (yes, this is a pretty bad shot!) 85 % of shots will be 1.2 tiles off.

Code: Select all

Accuracy 80%, Range 86:

Quantile   80.0  82.5  85.0  87.5  90.0  92.5  95.0  97.5  100
Error         0  0.30  1.20  2.76  4.94  7.70  11.1  15.1  19.8

After some upgrades (you can go up above 100%, which would make every shot hit!)

Accuracy 90%, Range 86:

Quantile   80.0  82.5  85.0  87.5  90.0  92.5  95.0  97.5  100
Error         0  0     0     0     0     0.35  1.50  3.38  6.05

Don't lower accuracy too far. If you set accuracy 0, the relative error will be 480%. So a unit that has a weapon range of 10 tiles with accuracy 0 may happen to hit a unit that is 58 tiles away. For ripple rockets, the standard deviation at accuracy 0 would be 103 tiles - so half of the shots would just come down anywhere on the map. Could be fun though - an EMP variation that reduces accuracy...

Let me emphasize again: this patch requires rebalancing. it will make all weapons more accurate on average, and even more so for low range use. A regular flamer has a chance to hit of 40%. In 60% of cases, it would be about 1 tile off -- even at shortest distance (the famous 45 degreees off!) With this patch, at a distance of 1 tile, flamers would be 70% within 0.45 tiles, 85% within 1 tile. This patch makes all short ranged units hit much more likely by chance, and at the same time makes long range units occassionally shoot really bad (which I actually like).

I have been considering to use the existing "hit chance" as the likelihood to be within 1 tile at distance 10 on average. But I havn't worked out the math for that yet. We may need another fixed point, such as 0.5 hit chance = perfectly on target. But that would keep the balance somewhat as is, I guess...

Note that when upgrades makes accuracy go to 100%, we get the "starcraft" gameplay anyway.

[Per]

We absolutely need inaccuracy to decrease the importance of fire concentration. It has already gone "a bit too high" with the invention of the new netcode (that is, now it actually works on both computers). With inaccurate fire the middle unit only receives part of the enemy army's fire.

Didn't fire concentration got overpowered from overkill detection code...? Normally fire concentration is balanced by the overkill of directing all units to fire on the same unit, but our units may be too smart for that. Also, the very long range of weapons makes fire concentration trivially easy to use (when you remember it; no need to position units carefully).

[Iluvalar]

Seems like you devs want to waste time on that epic fail idea of yours.

Which idea is that?

Realistic angular deviation accuracy. I litteraly spent hours of my time, trying to explain why it would be a fail backed with graphs and maths. You didn't read it. Here, you are still asking what I'm talking about and launch a poll with all options worst than the previous. You didn't even took my option or even Reg312's simpler option into consideration. NoQ said we should go on a private talks with people that proved they know something about balance. To reward him you launched a public poll instead.

Now, I stopped arguing... It's too complicated. Go for It ! Make those changes (remove a if in my patch to make pure gaussian distribution) and waste your time trying to balance a whole new game that can't be.

[Originway]

how would a pure physics based approach not work?
we don't have wind or friction to deal with so it is easier

[raycast]

This version of the patch is calibrated slighly differently.

Shots are less accurate - only up to accuracy/2 it will be a perfect shot. (So at accuracy 200%, every shot is perfect!)
However, the precision starts degrading after that, so that at the given accuracy, at a distance of 5 (sorry, I did not fix the comments. I initially used 10), the standard deviation will be 1.

I.e. for accuracy 100%, 50% of shots will be perfect, and 50% will be within 1 tile.
For accuracy 60%, 30% will be perfect, additional 30% within 1 tile, the remaining 40% will be worse.

Accuracy 50%: 25% perfect, 50% within .25 tiles, 75% within 1 tile, 90% within 1.7 tiles, and 100% within 2.25 tiles.

Accuracy 40%: 25% within 0.01 tiles, 50% within 0.36 tiles, 75% within 1.21, 90% within 1.96, 100% within 2.56 tiles.

At different range, scale these numbers by range/5. So at range 1, accuracy 40% will always be within 0.5 tiles. At range 86, an upgraded ripple rocket with 112% accuracy will occasionally be 12 tiles off, but in 56% shots perfectly on target, and 68% within 1 tile. At range 8, the maximum deviation will be 1.24 tiles, and 95% within one tile.

Maybe we should have accuracy degrade with squared distance. One might argue that the area grows with radius^2.

In the end, all theory is moot. Either it is fun to play with this kind of behaviour - good and bad shots - or it is not. I actually like how artillery seems to "home in" on your units when you get closer. The first shots are usually far off (with the occasional lucky shot) but once you get close, those hellstorms become a pain.

[Iluvalar]

Let me emphasize again: this patch requires rebalancing.

Nice Job.
I already said that, but since you are one of the reasonnable guy around I'll tell you. The worst problem occur when you cross the line where one of the 2 weapons have 100% accuracy. The comparative strength of the weapons jump from 60%/40% (150%) to 100%/100% (100%) in a very tiny gap. That will cause the mean accuracy to never occur. The optimal way to balance it is to cut in half. There will ALWAYS be in the game a player that is 25% stronger than the other because his units are on the good tile (say 5 tile distance vs 6 tile distance).

That will have the same effect of giving one player assault guns in a HMG fight. That would be hugly ! And if he can maintain that during as little as 1 unit death on 1 vs 1 (about 1 minute). Their comparative strength will bump to 152% in his favor. Rendering any counter impossible.

Investing more time than the opponent in positioning would guarantee a victory, forcing both side to invest more and more time into it until most of the effort in the game (after basic base building) will go into manoeuvring. Unless you inject another huge mechanic change in the game. Hope you understand that.


Still, your suggestion is right what I voted for. good job.

[Iluvalar]

how would a pure physics based approach not work?
we don't have wind or friction to deal with so it is easier

I made you a drawing and all the explanations in the other topic. On top of those repeated for raycast right now. When you don't understand the answers, change your questions or give up.