Weapon balance testing

[divVerent]

The "Over The Lazy Dog" and "PlanetNexuiz" servers run with statistics recording for weapon balance. Please everyone play on them so we can get a release with nicely balanced weapons out.

Everyone who did NOT take this chance in helping to balance out the weapons has no right to complain later.

The weapon balance data works like this: for every frag, the weapons of the killer and his victim are recorded. From this, a kill matrix "Users of this weapon killed users of that weapon so many times" is generated, from which the weapon balance scores are generated using a least squares estimate.

The resulting numbers will have the property that a weapon holder of a weapon with score X will win against one with weapon with score Y at a probability of about (X-Y) * 0.25 + 0.5. E.g. one with score +1 always wins against one with score -1. The average score is normalized to 0.

Weapon balance captured with ballistic bullets enabled, but with the stronger Uzi:

Code: Select all

  Nex                             1.01787 |   .    +0.111 +0.186 +0.262 +0.517 +0.617 +0.366 +0.515 +0.547 +0.806 +0.696 +0.799 +1.000   .   
  Uzi                             0.77656 | -0.111   .    +0.179 +0.118 +0.214 +0.500 +0.043 +0.625 +0.520 +0.660 +0.600 +0.742 +1.000   .   
  Rocket Launcher                 0.43152 | -0.186 -0.179   .    -0.209 +0.256 +0.099 +0.227 +0.160 +0.307 +0.485 +0.563 +0.691 +1.000   .   
  Mortar                          0.42326 | -0.262 -0.118 +0.209   .    -0.071 +0.053 +0.286 +0.200 +0.405 +0.483 +0.478 +0.521 +1.000   .   
  Hagar                           0.18761 | -0.517 -0.214 -0.256 +0.071   .    +0.429 -0.143  0.000 +0.333 +0.438 +0.267 +0.613 +1.000   .   
  Heavy Laser Assault Cannon      0.15445 | -0.617 -0.500 -0.099 -0.053 -0.429   .     0.000 +0.182 +0.257 +0.500 +0.471 +0.714 +1.000 +1.000
  Camping Rifle                   0.10461 | -0.366 -0.043 -0.227 -0.286 +0.143  0.000   .    +0.167 -0.143 +0.302 +0.538 +0.548   .      .   
  Crylink                        -0.03240 | -0.515 -0.625 -0.160 -0.200  0.000 -0.182 -0.167   .    +0.043 +0.277 +0.630 +0.750 +0.333   .   
  Electro                        -0.30860 | -0.547 -0.520 -0.307 -0.405 -0.333 -0.257 +0.143 -0.043   .    +0.302 +0.143 +0.243   .      .   
  Shotgun                        -0.69036 | -0.806 -0.660 -0.485 -0.483 -0.438 -0.500 -0.302 -0.277 -0.302   .    +0.111 +0.089 +0.333   .   
  T.A.G. Seeker                  -0.77381 | -0.696 -0.600 -0.563 -0.478 -0.267 -0.471 -0.538 -0.630 -0.143 -0.111   .    +0.081 +1.000   .   
  Laser                          -1.04216 | -0.799 -0.742 -0.691 -0.521 -0.613 -0.714 -0.548 -0.750 -0.243 -0.089 -0.081   .    -1.000   .   
  Grappling Hook                 -1.35614 | -1.000 -1.000 -1.000 -1.000 -1.000 -1.000   .    -0.333   .    -0.333 -1.000 +1.000   .      .   
  Port-O-Launch                  -2.00000 |   .      .      .      .      .    -1.000   .      .      .      .      .      .      .      .   


Obviously, the Uzi was too strong then. The currently running test is old settings, with antilagged hitscan bullets.

[FraNcoTirAdoR]

Wow, very cool statistic system divVerent! The high peak of the nex is really easy to explain, since the statistics run on CTF server, and the most liked and preferred maps are wide and open maps, it comes naturally that people prefer long range accurate weapons. This is the explanation why camping rifle is performing well also, but I still think it's a bit weak

About the machinegun: I agree that with the raised refire rate it got very strong, but also 2 factors raised its value in the weapon matrix: first the same as with nex: the wide open maps, that made mg secondary an optimal long-range weapon especially with the reduced spread on it. Second: people noticed, and I also poked many guys attention on the mg change and the massive usage of it since its test period, so during the testing hours it was much higher than normal.

Please keep in mind that the weapon balance cant be optimizad by examining only 1 gametype! In DM the numbern can/will/should be totally different.
Its really really shame that currently there is no SVN Deathmatch server, I would like to see the statistics from there too. Its highly possible that the rocketlauncher and electro would have a quite huge peak, while accurate weapons like nex and machinegun would end in the middle field (even this overpowered mg too, I mean just try to aim accurate while mortars pushing you up and down, rockets blowing into your face, and electroballs blinding your eyes talking about at least 6-8 player standard nexuiz maps, like agressor, warfare etc)

Also by examining those gametype, where players have choice and possibility to pick up and use whatever weapon they want (CTF really IS that one because of the size of the map lower the intense of action providing enough pickup time) its quite unlike that the most preferred nex-mortar-rl-mg quartett will be broken in short term.

As a conclusion I think that the 2.4.2 machinegun setting was somehow just nicely balanced (well maybe except the long range usage, but it can be lowered by raising the spread a little, or making range-dependent damage), and the particle based version in svn pulled this down quite much, but the new antilag can bring back the original usability.)

Please update this statistic from time to time, its really cool

[divVerent]

Actually, I'll be running a SVN DM server if that's what is needed.

[FraNcoTirAdoR]

Actually, I'll be running a SVN DM server if that's what is needed.

Sounds great Please post the name if its up, cant wait to see it, personally I have been missing it for really long.

[esteel]

Well for that matter, i have serveral servers running on planetnexuiz.de which all gather stats:

public:
port 1338: svn 2on2 tdm server
27000: the well known 6on6 svn ctf server

private:
port 35000: svn server with 2on2 tdm / 3on3 ctf via vote
7331: svn 1on1
29876: svn ctf 5on5

So there should be something for everyone.

[divVerent]

I am running:

emptyset.endoftheinternet.org
port 26000: CTF
port 26001: DM
port 26002: Race (okay, this usually generates no kill stats, unless in mode of the actual race)

All frags on the servers mentioned in this thread are gathered into that statistics program. If anyone hosts a SVN server with these current settings too, he can contact me to get the command to report weapon stats.

[halogene]

very cool approach! I will test the MG with both ctf and dm and try to provide helpful feedback, but unfortunately not before next week as I won't have time to play this week

[divVerent]

Code: Select all

  Nex                             0.52372 |   .    +0.296 +0.132 +0.436 -0.043 +0.667 +0.484 +0.289 +0.655 +0.349 +0.514 +0.844 +1.000   .   
  Camping Rifle                   0.35726 | -0.296   .    -0.400 +0.600 +0.556 -0.333 +0.351 +0.172 +0.500 +0.500 +0.308 +0.795 +1.000   .   
  Mortar                          0.30328 | -0.132 +0.400   .    +0.250 -0.385 +0.200 +0.082 -0.067 -0.083 +0.500 +0.492 +0.643   .      .   
  Heavy Laser Assault Cannon      0.27705 | -0.436 -0.600 -0.250   .    +0.714 -0.091 +0.231 +0.455 +0.333 +0.636 +0.429 +0.273   .      .   
  Hagar                           0.26620 | +0.043 -0.556 +0.385 -0.714   .     0.000 +0.077 +0.222 -0.333 +0.125 +0.727 +0.778   .      .   
  Crylink                         0.15984 | -0.667 +0.333 -0.200 +0.091  0.000   .    -0.059  0.000 +0.500 +0.333 +0.143 +0.250   .      .   
  Rocket Launcher                 0.15556 | -0.484 -0.351 -0.082 -0.231 -0.077 +0.059   .     0.000 +0.143 +0.250 +0.347 +0.564   .      .   
  Uzi                             0.12997 | -0.289 -0.172 +0.067 -0.455 -0.222  0.000  0.000   .    -0.333 +0.250 +0.200 +0.545   .      .   
  Electro                         0.06814 | -0.655 -0.500 +0.083 -0.333 +0.333 -0.500 -0.143 +0.333   .    -0.077 +0.583 +0.120   .      .   
  T.A.G. Seeker                  -0.05483 | -0.349 -0.500 -0.500 -0.636 -0.125 -0.333 -0.250 -0.250 +0.077   .    +0.172 +0.200   .      .   
  Shotgun                        -0.16083 | -0.514 -0.308 -0.492 -0.429 -0.727 -0.143 -0.347 -0.200 -0.583 -0.172   .    +0.222   .    +1.000
  Laser                          -0.30501 | -0.844 -0.795 -0.643 -0.273 -0.778 -0.250 -0.564 -0.545 -0.120 -0.200 -0.222   .      .      .   
  Port-O-Launch                  -0.55951 | -1.000 -1.000   .      .      .      .      .      .      .      .      .      .      .      .   
  Grappling Hook                 -1.16083 |   .      .      .      .      .      .      .      .      .      .    -1.000   .      .      .   


RL appears kinda weak, but as it hasn't changed, it'll sure rise up again. Actually, in the per-server stats, the RL is much weaker on pn.de's CTF server than on my DM server. Maybe the maplist...

[GreEn`mArine]

Thanks for providing this opportunity. Played a bit CTF already. Weapon balance seemed OK so far. I will test more of course.

[FraNcoTirAdoR]

Same here, after 4 hours of testing the mg and shotgun with this new antilag and current settings are very nice and usable. Also nice to see the list updates, seems like the balance spread got much narrower in long term Nothing surprising about that, that on the first 2 place there are the 2 sniper rifles because of the openness of many maps, and I also think that even with antilag the camping rifle isnt way too strong, maybe just a little. The reload, the reload-on-switch, and limited magazine, and high ammo usage is decreasing the amount someone can spam with it. Maybe if the ammo usage or the refire time was raised, that would make it place not so high, (and would also bring it closer to nexes behaviour), but the current strength of this weapon appears to be like a sniper-rifle should be: damages quite lot (but not vaporizes immediately like nex), accurate, and stealthed.

There is also something cool that can be seen from the statistics: so, the balance of the weapons seems to be okey, at least there isnt too huge spread between them, BUT there is a huge gap between electro and TAG, and it should be normal with laser and shotgun since they are starting weapons, so spawned players will be always fragged with them, but erm... TAG just shouldnt be between those

It's true what you wrote about rocketlauncher, I also noticed that the players (not the new ones actually) start not to prefer it too much, they rather use mortar instead. The constant blow-back to face can cause it