Forum
What would make for even teams?
Created 29th August 2012 @ 23:55
Add A Reply Pages: « Previous 1 2 3 4 5 Next »
Atmo, use any aggregate function you’d like: arithmetic mean, sum, median. Of course you’d have to live with the natural interpolation, but it’s still the best (automated) approach.
Last edited by nTraum,
The sample size is too small.
Is it?
Quoted from atmo
The sample size is too small.
So what? You’re right, but the fact that interpolation sucks with few samples is no news. You’d have to deal with that with any automated approach as well.
Quoted from nTraum
[…]So what?
So an automated system is probably futile.
Quoted from nTraum
Still, this is no reason to not use ELO. It is proven to be good in chess, cs, lol & other (epsort) tournaments. It is easy to use, it’s easy to implement and easy to adapt to your needs. Inventing your own ranking system will result in nothing but worse balancing than you’d achieve with mature systems as ELO.
/thread
Edit: of course you’ll need to modify it to get it work in 6s (instead of the 1on1), but it’s as easy as any other linear mathematical optimzation. You’ll have the best balancing when the difference bewteen the ELO points of both teams (which equals the sum of the players of that team) is minimal.
What? where did I say I am not using ELO? I am. Or rather, want to…
What exactly does using “an elo system” mean?
From what I understand, it means that a player’s rank/rating changes whenever he wins/loses/draws.
Right now I am trying to divide the teams to be as close/even as possible, based on the teams average Rank, or ELO Score if you will.
If you think you could help me with dividing the teams evenly and/or how much ELO a player gets/loses based on whether his team won/lost, please talk to me on steam. I can send you the source files as well. I could really use the help, as I really am clueless at this point.
Because of the small sample Atmo is talking about, players wont all start with the same ELO score of say 1000.
They play in a certain division, we know that prem > div1 – why not give a much better “head-start”?
In terms of dividing the team I would use a weighted formula in which certain classes rating are weighted more heavily to the overall teams rating. This way if you have a prem medic playing with div 6 team mates, the overall team rating would be lower then a prem demo playing with div 6 teammates.
I’ll use an example (I believe the actual weighting should be either decided by top players or through another self-correcting algorithm but for now I’ll just put what I think off the top of my head):
I would weight a demo’s rating the most significant on the baring of the game so let’s put it at 1.5.
I would put the medic as fairly important as well (the case can be made that it’s actually not as important in mixes) so let’s say 1.2.
The soldiers are the backbone so let’s keep them at 1.0 and the scouts are sort of a wildcard so let’s say 0.8.
So if you had 12 players with a variety of ratings you could find a few combinations for which the weighted team ratings are very close.
In regards to sample size, as I mentioned in my original post, I believe this is the main drawback to the ELO rating system for our purposes – ideally you have everyone that wants to play mixes doing so regularly and after a few weeks everyone has a decently close rating. What may happen is that we get the same activity as we currently see on pixel and it’s all for nothing and just a rating given by some admin would be better.
Someone mentioned how the ELO system works with multiple classees – either you lump all the classes into one rating or you actually have 4 separate ratings (one for each class). Again, the downside to the latter option is sample size.
If the weighting system wasn’t clear feel free to add me on STEAM and we can talk about a weighted average and/or the math involved with a dynamic rating system.
As an aside, has anyone ever contacted the tf2lobby admin in regards to improving the functionality? I see no reason why an ELO or equivalent rating system, combined with automated matchmaking right on tf2lobby cannot work well for a lot of people and would help bridge the gap between tf2lobby and div 6.
Quoted from Nicolas Cage
As an aside, has anyone ever contacted the tf2lobby admin in regards to improving the functionality? I see no reason why an ELO or equivalent rating system, combined with automated matchmaking right on tf2lobby cannot work well for a lot of people and would help bridge the gap between tf2lobby and div 6.
FLOOR_MASTER has been afk for ages unfortunately
I’d start with a graduated default system depending on divs, yeah, no point starting everyone at 1000 when there’s obviously already a big difference. Time will fine tune the players from the same div.
I’m guessing you’d need to be careful about the penalties of losing etc. to ratings – because of that whole team thing, you could be playing great but without a team to back you up it can result in sweet FA. It’s difficult to assess the contributions. A roamer would probably have more deaths, or a demo who is dealing a ton of damage but not getting most of the kills yadda yadda.
You would definitely require different ratings for different classes imo.
From what I recall from the old mpuk pickups which had the individual admin given ratings, the medics weren’t actually really involved, they just got randomly assigned. A bad medic is an issue for a team, but a good medic can’t actually do anything more than the level of their team so weighting them far over an offensive class doesn’t hold much sense to me.
dont rate medics, medics are only as good as rest of the team
Quoted from Nymthae
A roamer would probably have more deaths, or a demo who is dealing a ton of damage but not getting most of the kills yadda yadda.
Rating on kills/deaths/assists will be massively biased to pocket soldiers. So then we’d need different rating for pockets and roamers or they’d be meaningless and all roamers forever div 6.
Quoted from Nymthae
I’d start with a graduated default system depending on divs, yeah, no point starting everyone at 1000 when there’s obviously already a big difference. Time will fine tune the players from the same div.
I’m guessing you’d need to be careful about the penalties of losing etc. to ratings – because of that whole team thing, you could be playing great but without a team to back you up it can result in sweet FA. It’s difficult to assess the contributions. A roamer would probably have more deaths, or a demo who is dealing a ton of damage but not getting most of the kills yadda yadda.
You would definitely require different ratings for different classes imo.
From what I recall from the old mpuk pickups which had the individual admin given ratings, the medics weren’t actually really involved, they just got randomly assigned. A bad medic is an issue for a team, but a good medic can’t actually do anything more than the level of their team so weighting them far over an offensive class doesn’t hold much sense to me.
Your wrong about the “its not fair that you lose ELO because your team lost”. You need to change your lien of thought – because there is no such a thing as “I lost because I don’t have a team to back me up” in the ELO system.
ELO takes into consideration who was expected to win, and who was expected to lose.
Obviously, the bigger the chance for Team A to win, the less ELO they will get if they win, and Team B will lose less ELO, because they were expected to lose.
Given that teams are as even as possible, if you lose, you are a part of the loss, just as if you win you are part of the win.
Anyhow, the only features that I want to add and have yet to do are (in order of importance) :
1. Add ELO system. (scramble teams according to rank, adjust players ELO when pickup finishes)
2. Make STV demos accessible, post a link to download the STV when the pickup finishes.
3. Some simple stats for pickup. (each players K/A/D as well as there class played and team, who dominated/revenged who. who killed other teams medic the most?)
But in general, the bot works great. It is better than the current Pixel bot, and it also doesn’t crash – so if Skyride adjusts his Database to the bot, (which would also require him to change the website PHP code) this thing can start.
Add A Reply Pages: « Previous 1 2 3 4 5 Next »