no

after the hitbox change i realized one thing to be true i already suspected when they were introduced in quake4:
a good portion of my lg skills came from heavily abusing the '45° hitbox bug' which is now obsolete. in a way good aimers get robbed of a skill the newbies never knew existed.

the update also made me switch from tankjr to keel and eventhough i haven't played much in the last 7 days my rail accuracy didn't drop - my lg and mg stats did.

no idea if that makes any sense but in my case the game became even more centered around rg/rl/sg than before :/

With weapon crippling effect gone, there should be no bonus whatsoever for good aim other then damage dealt and excellent medals.

I have the same accuracies now then before, only shaft acc is ever so slightly weaker, but deals more damage so it's basically the same.

I suppose aiming should be some sort of gaussian distribution with bad aimers having a much broader distribution.

I was thinking the same. Now that you pointed it out I gave it a try in matlab using a normal distribution.

>> s = 3;
>> f = @(x)1./sqrt(2*pi*s^2).*exp(-(x).^2./(2*s^2));
>> cuboid = quad(f,-1.27, 1.27)
cuboid = 0.3279
>> cylinder = quad(f,-1, 1)
cylinder = 0.2611

distribtion s = 4
24,9% drops to 19,7%

distribution s = 3
32,8% drops to 26,1%

distribution s = 2:
47,5% drops to 38,3%

distribution s = 1:
79,6% drops to 68,3%

so, seen relatively the hitbox change does hurt bad aimers a little bit more than good aimers.

As it should in my opinion.

Edit: God damn my avatar I made today at work is nice! :D

yep, but in ironbyte's example it's not that much of a difference.
But of course it's only an approximation, so many factors like the ability to predict movement etc play a role.




guess who approved it ;)

I asked you to approve it, though you're still the man! :D

Look, it's shaking its booty =)

discu stu

Yes, but only very little (according to this model): -21% for s = 4, -20% for s = 3 and -19% for s = 2.

sure, but we all got 80% aim, don't we ;P

Only LoCK.

Truest worst ever spoken.

No. There is no conclusion as there was no real data to begin with.

The new hitboxes might "hurt" when missing more, but "please" when getting missed more.

but the missing part only depends on the opponent.

if he aims better after this update the advantage becomes a little bit bigger.

Let's assume noone aims better when the hitbox is getting smaller.

Dodging is reinforced the same as precise aiming by the change. You cannot look at aiming isolated and predict which kind of player will benefit the most.

I think less accuracy will enable more depth, while not making the game any harder to get into. win-win

you are taking the argumentative approach, which is fine.

But we are kinda discussing the math here ;)
I am aware that it is a rough approximation, but at least it is something!

Normal models are used for events that happen independently of each other. Define i to be the ith LG slug you fire and X_i a random variable that is 1 if it hits and 0 if it doesn't hit. The random variable X = (\sum_{i=0}^n X_i)/n will be precisely your accuracy. However, since the X_i aren't pairwise independent, it will be nothing close to a normal distribution.

If you want justification for the X_i being non-independent consider the situation where ith slug is a hit then it's highly probable that i+1th slug is a hit.

One interesting observation is, however, that P(X_i | X_{i-1}) should be close to constant. That is if you consider {X_i} as a stochastic process it should be very close to a martingale or in the worst scenario a supermartingale.

Summary: the normal distribution isn't what you're looking for. Neither will the gamma or exponential distribution work. I believe the thing that would model the phenomenon best is a diffusion process of the type dX_i = \sigma dW_i + \gamma dV_i where W_i, V_i are independent Wiener processes and \sigma controls how good "track aim" you have and \gamma how good "flick aim" you have.

good point.

:[

LOL Wiener processes

It's true they are not pair-wise independent.

You are tracking your opponent who suddenly changes direction. With a reaction time of 200 ms you will be off target for atleast 8 server frames.

So given X_{i-1}, we can make a qualified prediction about X_i.

Reaction time, however, can be fitted to the sum of an exponential and a normal distribution, according to wikipedia. Does this mean that reaction times (somewhat related to the time you are off target) can be considered independent stochastic variables?

Just that I believe no one shafts purely based on reaction.

Precisely.

Not very random.. :d

Huh? How isn't it random?

I made some graphs of this model.

"I mean you fall within a bellcurve."

It's probably more like a gamma distribution.

elaborate, please.

If "aim" were a random variable, it would probably not have a normal distribution. There are a handful of atrociously bad aimers, a lot of slightly below average aimers, and it slowly tapers down through average and above average, until you get to the ridiculously good aimers which are few and far between.

This is pretty much how a gamma distribution looks, provided the right 'k'.

Edit: To better explain, gamma distributions have the most frequent value come before the average. This seems more accurate than normal, which has the average equal to the most frequent.

ah ok.

you are talking about the distribution of aiming skill from several persons, right?
My proposing a gaussian distribution was about the aim of a single person. How much he deviates from the target.

More graphs :D. Gamma distribution.
I've defined a hit as X<1. The lower k and theta, the better aim.

but why?


I mean according to wiki gamma distribution is commonly used for waiting lines and insurances. I am unable to see how those compare to probabilities of hit or miss on a hitbox.

Don't laugh at me now, but I recorded a demo of a shaft duel and analyzed it frame by frame, measuring the width of the enemy and the distance I was off. Now, I know demos aren't accurate, this is only one battle, and the data set is very limited.

Here's the data I got:



x = (distance from center / half-width of model)
So x<1 would be a hit.

ok, can you maybe do the same but this time negative numbers for deviations to the left and positive to the right?

k = shape factor, theta = scale factor
X would be the absolute distance from the "center line".

I'm not very good at statistics but this seems very reasonable.

see #41 I think he isn't talking about one person but a group and their respective skill.

He's certainly talking about a population.

Different weapons would have different distributions. You don't aim the same way with the rail as with the lg for instance.

I also had the most stupid idea ever. Playback a demo, record a movie of it and analyze it frame by frame using some program. I don't think it would even be worth the try though :D. Visually doing hit detection from screenshots.. *sigh*

yeah, I agree. I was mainly talking about lg, though. But I think all hitscan weapons should behave more less like that.

I also had the most stupid idea ever. Playback a demo, record a movie of it and analyze it frame by frame using some program. I don't think it would even be worth the try though :D. Visually doing hit detection from screenshots.. *sigh*

compared to that my coding a program idea is genius :P

The more I think of it, the more I think it's feasible.

Find a lg battle, cl_avidemo. Supply the screenshots containing nothing but the lg battle to a program. The model's position can be found based on a distinct enemycolor.

One problem is "the crosshair" seems to trail in demos.

yes, there are probabably measuring errors through the demos, and in the end you still need to program something, even with a picture recognition or something like that.

Sure, if you do it correctly you will get a more reliable result, but I'm not convinced that increase is worth the effort.

We would be able to analyze far more data from different players. I think it's worth a try :)

go for it, then ;)

You're gonna beat me to it? ;)

nope.

Today is my last day this year and I'm leaving for vacation on tuesday.
Also, I never did something with recognizing pictures, so it would take me lots of time to do it.


But you have fun ;)

It's working :)
Written in matlab.

\o/

The effects of unlagged impacts this alot. Watching it frame by frame makes it clear shots fired way off visually still registers as hits. At certain points the distance to the strafing enemy is constant and every shot is still a hit (according to crosshairhit). :(

how about you left someone play with fraps?

l-l-lag

get a decent pc, np

I don't think the demos are accurate to crosshair position or to the position the model appeared at while playing.

No, they are not. Do you have any more (specific) information regarding this?

Fraid not :(

Would it not make it a bit pointless doing this off of demos though?

not if it's something like: model always lags 50ms behind the crosshair.

But I really doubt it's something as easy as that.

I'll have to think about it. The crosshair is definitely lagged if demos are recorded server side (which was stated somewhere..). The problem, I guess, is that with unlagged the client decides if it was a hit or not.

[Speculation]
As demos are recorded server side, I think the view is delayed by (1000/sv_fps + 1000/cl_maxpackets + ping)/2. Atleast I notice less delay in demos recorded at sv_fps 125.
[/Speculation]

The delay will cause a wider distribution than it really is, but it's the same for every player. It's really difficult to do anything about it from screenshots. I did try a few approaches by calculating ahead, but it didn't work out very well.

activate sparks, adjust sparktime to less than lg reload time and count the sparks

ok just read again, this cannot help with deviation...

ql > q3 after last update already !

burnedd > guard after few months already !

f*ck , now I'm gonna kill my self! :(((

Play me, and u are going to own him in 2 weeks.

better I will kill my self ^^

I had a thought last night.

If we want to find out what distribution fits to "aiming" someone should write a little flash program or so.
Just have a green rectangle move left and right randomly on a screen at a speed that pretty much matches quake's strafespeed.
Then let people try to keep their mouse fixed on it and record the mouse positions relative to the rectangle's (vertical) center line.

Have several people with different skill levels do it and you should get a good impression.

.. and change the area of the rectangle.

So, you're gonna code it? :)

I am certainly not :)

Never coded anything in flash and am not starting for this.

You can do it in asm, np

show me


;)

flash is shit, i may be able to do it in javascript but the results would be easy to fake

don't care about being able to fake the results, it's more for testing purposes.

Would be awesome if you could have it oubput a table or list of the deviations :]

what speed would it need to move at? lets assume a fixed 800x600 px area, what would the size and speed be?

320 ups is the standard speed, dunno the exact hitbox size and quick google fails, too :/

When you change direction (left/right strafe) you slow down and speed up again.

guess we should just ask some modders...


(I did find some comment that the hitbox size was 30, but I don't know if it can be trusted ...)

You could at least read the q3 source code and set it up for q3 settings, then switch/add ql ones if they are ever found/released/whatever.

or ask someone who knows the q3 values or can find them fast.

Yeah

Q3 values were 56 units high by 30 units wide.

thanks, any numbers on the speed changes ini mentioned?

I mean it doesn't have very exact just a general direction is probably enough.

For QL? No idea.

ok, so it's 320 ups with a 30 units model.

so, just make the model a certain size and let it travel a distance of 10,7 times it size per second.
Also do some acceleration function for directional changes, maybe like a square function that reaches top speed in 0,2 seconds or so.

Hope that helps :)

LG percentages for everyone has dropped, a LOT. It's no myth.

good

another graph wich clearly points out rail has to do 100dmg

look!

Oh, now you missed it. :(


it was a ufo with a sign that says: "rail needs 100 dmg"

it clearly doesn't

why would you want to slow the game down even more?

it does

I think a rail hit should be equal to a direct rocket. So make rl 80dmg and I am fine with it but atm it feels unbalanced to me. It wont slow down the game. Ofc it will be much more important to hit the rails or to not get hit by a rail but atm its like a rail doesnt matter much. In a rail vs rail situation it doesnt matter who hits first because both players will usally survive and so on. Its not about fast reactions anymore just about who can hit more rails...

yea because hitting a direct rocket is just as easy as hitting a rail.

you cant hit a direct rocket at low-mid range?
Seriously its up to the enemy dodging skills regarding direct hits. So I think its equal hard or easy to hit

It's a LOT harder to hit a direct rocket than it is to hit a rail.

Why do you think rail should be as powerful as the rocketlauncher? Railgun obviously takes less skill to use, and because of that it could be abused a lot easier.

I think there are a lot of reasons why rail should be as powerful as the rocketlauncher... but I am too stoned atm to sum them up.
Railgun doesnt take less skill! Its up to the situation and range. Win a rail vs rocket situation at close range. Thats skill.

Still a rocket to the face is so much more skilled than to hit a rail... weird how you seem to think otherwise

Win a rail vs rocket situation at close range. Thats skill.

Depends on the situation really:P

Anyway, get sober, then give us your arguments.

"Win a rail vs rocket situation at close range. Thats skill. "

You aren't supposed to win that situation.

rail is still the weapon of choice for long range and people use itfor that all the time.

If it becomes possible to jump into close range and win with rail vs. rl or lg that's when the balancing is seriously fucked up.

Just as it would be bad if long range battles against rail with sg or rl or even plasma are often won.

It wont slow down the game. Ofc it will be much more important to hit the rails or to not get hit by a rail

interesting

I do like the new hitboxes. However, I think accuracy always depends on the game mode - tdm, ctf (defense) are very important when it comes to good aiming. 1v1 on the other hand is different; you can put off your opponent with good timing, desicion making and where you are in relation to him (time and space).

I do think that rail is still a bit overpowered but that's just the nature of the gun. If you miss a railshot in close range you are usually dead. The weapon is designed to make a lot of dmg from the distance, whereas the rocket launcher does the complete opposite.

I think it is fairly balanced. I mean, what else can you do? They already decreased the damage. Do you really want to decrease the rocket launcher dmg as well? I don't think that this is a wise desicion. I think shotgun is still as useful as it was before the patch - maybe a tad bit harder to aim.

Smaller hitboxes favour the people that have very precise aiming. I see people, than had accuracies of around 42 in average, dropping to low 30's because they got used to the fucked up hitbox. I think very precise aiming should should be rewareded - and indeed it feels good to hit again. Unlike before.

Maybe I drifted off a bit. I don't know what I wanted to say in the first place. xD Lovely graphs though :P

Btw: My plasma actually has increased and so did RL and rail. MG, and shaft are still off a few percentage. sv_fps 40 is awesome.

What I have found is my game was largely based on hitting a high enough level of acc to be aggressive. Now playing CA i get butchered by rocket only players i was dominating just a week before the update. My whole gameplay will have to adapt to a primarily defensive rocket based game.

I feel it has had more of a negative impact on the high aim players at the moment. Sure everybody's acc is lower its just the shit players that couldn't aim before are used to hitting nothing so know exactly how to play that way (i.e. I'm going to have to learn how to play like a scrub again).

Agree on all points ; (

meh

i can't hit shit yet. feels like osp. i preffered the old hitboxes or cpma

The question is: Does it have to be easy to hit in a game in order to be fun? I'm not sure. It defininetly helps but it's not all! I can cope with the new hitboxes - I like the rewarding feel when you hit. But I said that already. :D

ok ok we got it it would feel more rewarding when u hit but what if u cant hit shit?! cant believe it i mean i had no idea i was benefiting from the old hitboxes that much..but my shaft dropped from an average of %35 to %20!!!

Well, there is probably a vast different between us two because my shaft dropped from an average of 41 to probably 35 or something I mean, that's quite hard to compare...

It's the old story again that "bad" aimers are hit worse than the better ones.

so from reading the discussion, the game gets less noob-friendly ;)

true

i probably learned more about math in this thread than this whole semester

*statistics ;)

that aint no science, just fitting reality to your models like them econometrists do

true, but all the models ironbyte and lejoon used are part of the subject "statistics".

Wanna buy lessons? :D

tehe