Statistics algorithm?

[_Nisa_]

I am interested in the stats algorithm in https://unfreeze.fpsclasico.de

It seems to favor players with a lot of playing hours (like me) over players with higher efficiency. Can you open up the algorithm? One should naively expect someone with a high kill/death ratio be much higher in the ranking than someone who cluelessly roams around day after day and never hits anything (like me).

To be honest, based on my experience and success (and the lack thereof) in the servers, I would expect myself to be in the 100th place, not in the top 30 LOL

[adminless]

I talk about that in this topic you can read here, well, yes your case is a bit extreme hehe by looking at your stats and in comparison I probably agree with you that your rank is probably over-rated and that you should rank some bellow, I'd say around 60 or something like that (not in the 100th place in my opinion btw haha), but is that by playing around twice as your mates you've had more opportunities to get points and also it's affected by the fact that you play more in mornings, early evenings where there's not so many competition so again in comparison it positively influences your rank (on this times you use to lead or be closer to leading) so bottom line yes you right it's not "perfect".

[Black.Nastia]

wow what a hottie playing with us!)))

[tar]

The statistics are not accurate as long as there is a) no "damage done" listed and b) the actual time of playing is not considered (without time being frozen and therefore unable to do anything) in order to get a median damage done/playtime and c) no precise weapon accuracies are shown (now it only shows the kills and deaths ratio per weapon but not the accuracy). Therefore I do not understand, why the overall real accuracy (not per weapon, but at least rocket jumps are filtered from it) is shown within the game scoreboard without the possibility to get the accuracy per weapon and include both in the statistics. Furthermore, I do not understand why thaws are even included in the overall points of the game scoreboard while they are clearly excluded in the statistics. In my opinion, frags and thaws should be split up within the game scoreboard and thaws could be included in the statistics but damage done per time would be the most accurate thing to measure efficiency. It should also be taken into account that the team reorganisation logic influence the statistics. E.g., it makes no sense to list won/lost matches when switching teams inbetween.

But even all of that would not be sufficient to compare skill as the ping is different, the gameplay style is different and several other things gives advantages and disadvantages:
- player side: hardware, config
- server side: respawn hell, team reorganisation, unlag code

[onlycape]

I believe that in general terms, the default ranking generated by statistics is close to reality.
As Tar says, the skill parameter does not take into account the game time and that is why statistics are favored for players with more games, which logically will have higher accumulated scores.
Anyway, we can always reorder the ranking by any of the other parameters.
Other point: "Eff%" and "K:D" are the same. One of the two could be replaced by accuracy, for example.

[adminless]

hello guys, yes tar, I know that I missed to review the stats site for a long time as I had planned, sorry about that, unfortunately as you could see workflow here lead me in other ways and by now this stats review got rescheduled more as a long term "idle" (lower priority) task. interesting thing is that this week as I was cleaning up the server code and updating the servers (which is still and on-going task) I was precisely looking at the time code and was just about to add it (as is not that much of a deal like I said, probably that, a evening) but finally decided to leave it by now and get focus in what I was doing or if not probably I wouldn't be writing here (i.e. I would have missed to answer some posts and probably I wouldn't have pulled off the new unfreeze package, the server would be lagging etc etc) so I really hope that as I finish with this (a couple of days or a week may be) that comes next. for the rest for the most part you're probably right about what you're saying so I hope I can integrate at least some of those things in one way or another, thanks for the suggestions. I believe the main purpose for count the thaws as points in-game is as a incentive for the players to help each other and as a reward for the lower ranked ones, I don't really think that's a big deal.

yes, overall, I also agree with cape here, I know that it can be improved (as pointed) but overall I believe that it should yield acceptable results already, if we look for example at the rank which spawned this thread is already "rapidly" decaying into more "accurate" values (getting closer to 40 which I believe that it should suit for the case being), in my opinion I'd say that it only need few tweaks here and there and a few additions in order to be "complete" rather than a in-depth/radical change. also note about the influence of amount of games played that is also somehow a important factor that people who may be play a bit more get rewarded for that, of course not in a "crazy" way but at least slightly (let's say a few ranks higher), plus I believe other way it would be very detrimental, don't you think? I think that it would be really bad that the rank actually punish you for playing more, it would lead to people start playing less because of that, plus I also don't think that nobody of us want some random guy that just played for a couple of days (and could be just under the influence of all the circumstances we already mentioned) to stay there at the very top for a whole rank cycle.

and yes sharp remark there cape about the repeated fields, thanks about that, I'll see if the next time I work with that I add that too.

[Krakowska]

This server is strange
80% acc rail
70% acc lg
90% acc sg???????????????does it work well?

[tar]

This server is strange
80% acc rail
70% acc lg
90% acc sg???????????????does it work well?

Actually, it is not real accuracy. Instead, it is kills per weapon/deaths per weapon. Only the accuracy on the scoreboard shows the accuracy (of all weapons together).

[Krakowska]

This server is strange
80% acc rail
70% acc lg
90% acc sg???????????????does it work well?

Actually, it is not real accuracy. Instead, it is kills per weapon/deaths per weapon. Only the accuracy on the scoreboard shows the accuracy (of all weapons together).

Ahh, okey

[adminless]

I am interested in the stats algorithm in https://unfreeze.fpsclasico.de

It seems to favor players with a lot of playing hours (like me) over players with higher efficiency. Can you open up the algorithm? One should naively expect someone with a high kill/death ratio be much higher in the ranking than someone who cluelessly roams around day after day and never hits anything (like me).

To be honest, based on my experience and success (and the lack thereof) in the servers, I would expect myself to be in the 100th place, not in the top 30 LOL

ok, finally yesterday while I was fixing a bunch of links at the stats site and noting that lately the rank was basically broken altogether (there was way too many people on top positions that shouldn't be there, people avoiding games/teams/enemies just for the rank, many people largely over/under rated, embarrassing +3000 elo ratings and probably much more) I came across this issue and dug that stats algorithm people always wondered about and tweaked it in such a way that just by looking at the rank now, I believe that now everything actually makes sense all of the sudden.

so coming to your question the original verified stats algorithm is as follows:

Code: Select all

KillerSkill += weaponFactor * victim_based_skill_no
VictimSkill -= victim_based_skill_no

where:

victim_based_skill_no =  victimSkill * skillset_fraction

with:

skillset_fraction = 1/1000.0

and weaponFactor:

GAUNTLET = 1.5
MACHINEGUN = 1.0
SHOTGUN = 1.0
GRENADE = 1.5
GRENADE_SPLASH = 1.0
ROCKET = 1.25
ROCKET_SPLASH = 1.0
LIGHTNING = 1.25
PLASMA = 1.25
PLASMA_SPLASH = 1.0
RAILGUN = 1.5

------------------------------------------------------------------------------

ex: player with skill 1400 kill player with skill 1600 with a direct rocket
killer player new skill is 1402 (1400+1600/1000*1.25)
victim player new skill is 1598.4 (1600-1600/1000)

by looking at the code it really surprised me that there was different skill factors depending on the method of death, for the most part I agree with that (it actually makes sense to reward more a direct hit with a rocket than a indirect one and it's also one of the factors that largely explains why you was so overrated as now that there's awards report it turns out you're number 1 plasma killer which has a skill factor of 1.25) except for one, the railgun factor. I believe that for regular Quake III Arena that could apply (as the railgun resources are very limited and furthermore there's no lag compensation) but for this particular mode where railgun resources are virtually unlimited (to some extend) and there's actually lag compensation is clearly way too much overrated, therefore I reduced the railgun factor to even out a bit the game.

you're going to excuse me here but I won't particularly disclose the exact modifications I made to the rank algorithm for the very same reasons I didn't provide you with the server files, this is, in order to avoid that toxic brain death people (not you) just come here and "copy" (steal) the whole thing but I can at least tell you that in addition I made the victim skill influence on the algorithm square instead of linear (in normal words I increased the influence of this factor in the algorithm) in order to yield more realistic results (just as a example Magnus Carlsen dedicated elo is in the 2800 range, have people with +3000 elos at a public skirmish like this is plain non sense) and reduce (to some extend) that people intentionally exploit the algorithm by inflating their rank playing lower skill opponents/easier rooms as it was frequently the case lately. now most of the skill points will be distributed at peak times when the most skilled players play and it will be harder to inflate it by over/exploiting play it.

so that's it, I hope that this time at least I fully answered your question (as well as many other people, this has always been frequently asked) and sorry for the delay hehe as you could see I've had plenty of other stuff to deal with here before I could come to this so now that's it. looking at it now I feel like it turned out great in my opinion now for example your rank just lowered from your initial top 30 in the 50's range which is much more meaningful and it's even around the range I had foreseen for you at my second second post here

also as mentioned on this post I patched yesterday the awards report at the rank site that was broken from the start and that I'm still working around it but well that's a story for the usual stats and rank thread not here, so see you for now and have fun

[fernandinho1337]

the ELO calculation is interesting. to be honest, i would not differentiate too much using the weaponfactor because in the end the frag matters. you can get easy and difficult kills no matter what gun you use. it also makes a difference if you kill someone in a 1on1 scenario or in a 11on11 scenario, therefore I feel railgun is overpowered (eventhough it would favor my personal ranking which i dont care too much about). all guns should be sort of the same. it is not possible to detect a "skill hit" and a "lucky hit" anyways so why make a difference

best regards