Skill based balancing instead of time based ranking (done)

[Licho]

Stats server (used by springies that have stats enabled) now calculates elo rating of players.

This is used for team balancing instead of default time based rating.

This is top50 players from the database: http://springie.licho.eu/top50.php

12 000 games of some 55 000 players. Only games longer than 5 minutes that ended correctly are in database. FFA games are not counted into elo ratings.

If you want to use this system too let me know. Only submitting stats is a bit complicated due to security measures. Querying for elo is simple:
http://springie.licho.eu/stats.php?getelo=PRO_Day

[Tribulexrenamed]

Urm... right... Prominence is on your list XD.

[Regret]

Haha skill based on win/lose ratio in teamgames.

[Neddie]

I've played an overwhelming number of games where I have deliberately chosen a weaker team either in an attempt to balance the match or to challenge myself, and I feel that flat ELO does not take this into account. Nor does it take into account the distribution of my play across a wide variety of games. Nor does it take into account the waxing and waning of skill which occurs over time, or the extremely varied circumstances from match to match. It does not consider the matches I played to test widgets, it does not differentiate between a 2v2 clan battle, a 5v5 with mixed players and an 8v8 played to see if one could overcome the inabilities of seven. I do not even know if it takes into account involuntary disconnections. Even if the assumption that my skill could be registered in a number were valid, this would not be a good way of doing it.

We already had the ladder for ELO ranking, and I took great pains to ensure it was elective and limited in scope, this is not. This is an arbitrary substitution of naive statistics for human evaluation, no better than time balancing and surely much more susceptible to wankery. I appreciate the effort you've put in Licho. However, it must be noted that no number can accurately convey skill, particularly since performance is contextually dependent. It is folly.

If you're going to do this, then you shouldn't use data gathered without knowledge of intent before the debut of the feature. It should be new, so we can all start playing like paranoid little teamstackers at the same time and start on an even playing field. Or maybe not, maybe I don't want to stop trying to stack odds against myself. This implementation remains, however, a somewhat unethical use of information.

And, to be more directly constructive, if you're working on balance mechanics for Springie, here are three more to consider...

1. Random/rank balancing once server has reached manage value/full value, followed by elective switching by people who agree to switch.
2. Random captain selection once server has reached manage value/full value, followed by turn-based draft selection by captain.
3. Random organizer selection once server has reached manage value/full value, followed by turn-based draft. Each turn all members of present team vote on next member, organizer has deciding vote in a tie.

These should be particularly useful in light of the ELO system. People can draft select the people they know, and thus ignore the direct ELO, but the ELO will inform them and draft games will stagger stacking to some extent. These should counter some of the manipulative attempts which will result from the raw rankings.

Actually, I think this is a decent development. No ranking system is good, but you need some sort to balance by automatically. I feel that this and time ranking should both be options to balance by, however. This won't prevent me from playing games with who I wish to, and this won't change how I play. It will surely promote more stacking and manipulation based on the scores, but frankly, if somebody needs that to enjoy the game, I pity them in passing, but they're no affair of mine. If somebody wants to judge me on a number, their loss.

Let us see how it goes!

[Neddie]

Also, uhm, it doesn't seem to handle renames. Neddie is ~54.8 points higher than [S44]Neddie, which is ~1.5 points lower than [SmuG]Neddie, which is ~4 points lower than [7uP]Neddie, despite all of them being the same account renamed. I did substitute smurfs in to keep the names, though, but they all had zero play time.

Google is on there twice, as number one and number seven.

Though, it does calculate your ranking correctly, showing flatline value for your two previous names.

[imbaczek]

related reading, maybe useful for somebody: ms trueskill

[Regret]

I still think that I explained the problem better than neddie.

[Tribulexrenamed]

I still think that I explained the problem better than neddie.

Um, I think I did a better job, by pointing out that Prominence is on the list.

[HectorMeyer]

This sounds pretty cool. Looking forward to play on those servers soon.

[Licho]

I have considered trueskill too (for planetwars), but it does not seem to produce more accurate results and is much more CPU intensive to calculate.

Both systems are not ideal for team games (nothing is), but imo results produced by elo seems to be pretty accurate.

I believe they reflect real state pretty well. Its problem is slow learning. Because elo is averaged for team, so are gains/loses and it takes very long time for player's elo to reach his "level".

I wish there was accurate system to track smurfs that would make it much more useful.

Regarding stacking because of elo rating - meh i doubt it. Everyone knows its mainly from team games. If you want real ladder play 1v1, this is just for balancing purpose.
Stats exists for 2 years and i never heard people trying to improve their win % by stacking. I can hide top50 if you turn out to be right.

And neddie, if you keep testing stuff while playing, your elo will go lower, as system reflects your reduced effectivity in team games. If you stop doing it, it will go higher.
I dont see a problem here.

[Licho]

Everyone playing games balanced by this should have win ratio 50% btw. (Over longer time period of course).

[TheFatController]

This should work out better than just using .ingame imo, and the top 50 looks like a pretty good representation of good players (lots of familiar names).

Looks interesting!

[Beherith]

FFA games are not counted into elo ratings.

Like the idea, Does it do separation based on mods as well?

Since it seems the top 11 players are almost exclusively CA players. Is that because CA springie hosts are played more than BA ones?

[Otherside]

need to stop speccing and grind my elo. Rank 53 or something were id be but it dusnt take into account my old acc's :/

[bibim]

I wish there was accurate system to track smurfs that would make it much more useful.

Well, at least it could track renames if unique IDs were used in lobby protocol, as I had proposed to implement in TASServer almost a year ago...

[Beherith]

I wish there was accurate system to track smurfs that would make it much more useful.

Well, at least it could track renames if unique IDs were used in lobby protocol, as I had proposed to implement in TASServer almost a year ago...

Isnt tizbac's SmurfServ of any use?

Also, can this be applied to SPADS type hosts too? Or just springie?

[bibim]

Isnt tizbac's SmurfServ of any use?

SmurfServ doesn't track renames, it only tracks known accounts of known IPs.

[Beherith]

Hmm, this is odd:

[1:08:55 PM] <Springie> [teh]Kixu - elo rating 1704 played 138 games, has won 84% and quit 7%

[1:10:17 PM] <Springie> rakhim (bukis) - elo rating 1711 played 56 games, has won 75% and quit 4%

Is it because hes low rank and played in team games where both teams had decent players but his low rank got him more CA vets?
Thereby effectively power-levelling him?

Isnt this why Élő ratings are only for 1v1 games?

[Licho]

Yes. If you smurf with 1 stripe account it effectively power levels you until you reach your elo rating level (with old lobby time based balancing).

If you have silver star in lobby, you are put in weaker teams and can have high elo even with very bad win ratio. My win ratio is for example 57% and i have higher elo than their accounts.

[knorke]

what does ended correctly mean?
it would be cool to see the whole list or at least +/- 25 from your own position.