Spring RTS Engine

Cheesecan wrote:Not sure if elo ratings based on 10v10 games can be accurate. 1v1 for sure, 2v2 probably, 3v3 probably, 4v4 meh, but 10v10? c'mon..you can win even if you sit and tech all game and don't make a single unit. :D

To win like this you would need good players on your team to take up your slack. With a high elo you are more likely to have low elo players on your team so it balances out.

Ignore everyone.

1 chevron = 0 hours
2 chevron = 1 hours
3 chevron = 2 hours
4 chevron = 3 hours
gray = 4 hours
gold = 300 hours
vet = 1000 hours
ghost = 3000 hours

Google_Frog wrote:

Cheesecan wrote:Not sure if elo ratings based on 10v10 games can be accurate. 1v1 for sure, 2v2 probably, 3v3 probably, 4v4 meh, but 10v10? c'mon..you can win even if you sit and tech all game and don't make a single unit. :D

To win like this you would need good players on your team to take up your slack. With a high elo you are more likely to have low elo players on your team so it balances out.

Not really.

ELO rating is an algorithm used to compare Chess players 1 on 1. It has little relevancy to massive team games in Spring.

tbh every big CA/XTA game ive seen/played was horribly unbalanced.

Wombat wrote:tbh every big CA/XTA game ive seen/played was horribly unbalanced.

My bad.

np, u provided a lot of fun ^^

Cheesecan wrote:

Google_Frog wrote:

Cheesecan wrote:Not sure if elo ratings based on 10v10 games can be accurate. 1v1 for sure, 2v2 probably, 3v3 probably, 4v4 meh, but 10v10? c'mon..you can win even if you sit and tech all game and don't make a single unit. :D

To win like this you would need good players on your team to take up your slack. With a high elo you are more likely to have low elo players on your team so it balances out.

Not really.

ELO rating is an algorithm used to compare Chess players 1 on 1. It has little relevancy to massive team games in Spring.

Wikipedia wrote:The Elo system was invented as an improved chess rating system, but today it is also used in many other games. It is also used as a rating system for multiplayer competition in a number of computer games,[1] and has been adapted to team sports including association football, American college football and basketball, and Major League Baseball.

Basically, it still works.

The imbalance problems usually stem from:

1: Absolutely new players being assumed to be of average skill level. This means that new players usually will bring down their teams with them until their elo level reaches something more appropriate. There isn't any way around this because lowering the starting elo simply lowers the average.

2: The balance function doesn't always optimally balance the elo rankings between teams. There doesn't seem to be an easy way around this. In which case, manual balance for more optimal elo distribution will give you better balance results.

Manual balance given a fair balancer who is familiar with all participants will always give you the best results.

luckywaldo7 wrote:

Wikipedia wrote:The Elo system was invented as an improved chess rating system, but today it is also used in many other games. It is also used as a rating system for multiplayer competition in a number of computer games,[1] and has been adapted to team sports including association football, American college football and basketball, and Major League Baseball.

Basically, it still works.

The quote you mentioned doesn't applies.
When you apply Elo to a team you do it to the team as whole, which changes little over time as structure, eg Red Sox's Elo.
In Spring you got the Elo for each individual player, and then a team Elo is created as function of the single, which is absolutely not covered by Elo itself (since assumes binary contendants ) and instead relies on custom code/algos to compute the Elo of the teams starting from the individual, then reflected back to the single players using another custom function.
all in all, I wouldn't call it Elo at all since it operates on heavily altered data that otherwise wouldn't fit it's paradigm.

League of Legends and Heroes of Newerth use modified Elo rating systems to rank individuals in a team-based environment

Thank you BD, I picked out a more relevant quote out of that article.

I do not know exactly how the implementation works, but as I recall Licho based it off of LoL. So maybe you are right though, maybe you can't exactly call it elo. My real point being that "Springie ratings don't work because they are designed for 1v1 and not teams" is only a community-perpetuated myth, and please stop spreading it guys because I get bruises from headdesking.

How about that win/loss ratio gets countet too instead of gametime?
and making Ranks higher or lower depending on that i think it would be far more accurate

Just make a ranking system involving W/L, gametime, and amount of 1v1-3v3 ever played.



Example: Number means number of small games won, symbol means ingame time, color of number means W/L.

Top has played averagely in a lot of small games, but has a relatively low ingame - meaning he's a smurf.

Bottom has played a ton in large games, but doesn't play much small games and loses a ton. Sucky player.

Right is a decent player, wins and has played some small games.

Left is a random dude who you can't determine because of his low ingame (I made up a rank symbol), games played, and W/L.

Doesn't look that confusing, does it?

I think current rank system works well up to around silver/gold star (smurfs withstanding). Above that could perhaps rely on win/loss to some extent. I don't think balancing games is the same problem as ranks and I don't think rank display needs to be more complex than current (further stats belong on a player profile.. not that this exists atm).

Just play primarily with people you know and manually balance based on your evaluation of comparative skill. It is much more effective and mature solution than any automated system.

Neddie wrote:Just play primarily with people you know and manually balance based on your evaluation of comparative skill. It is much more effective and mature solution than any automated system.

and also one of the factors that limits spring growth.

knorke wrote:

Neddie wrote:Just play primarily with people you know and manually balance based on your evaluation of comparative skill. It is much more effective and mature solution than any automated system.

and also one of the factors that limits spring growth.

Virtually nobody manually hosts these days, so your statement is patently untrue.

yea, these days everybody is on the dsd autohost, no need for manual host
---
If everybody only "plays primarily with people he knows", nobody will want to play with a new player because nobody knows him. This limits the playerbase to the core community that has been active for past few years. It creates an "inner circle" that is impossible to get into.

I used to host NOTA and manually balance for long sessions. Simply talking to new, unfamiliar players easily solved any problems. In contrast to current BA (for example), I'd say the atmosphere was very inviting for nubs. There's no reason autohosts could'nt be run in the same way with a few decent admins (autohost fine control could be simpler to use). I moved on to running an autohost at the time but problems with new spring versions put an end to it.

Starcraft 2 has a modified ELO-system where you must play 6 placement matches before you get a ladder rank. I think it is roughly based around win/loss ratio with opponent rank factored in.

You also have to do this when you form a party(pick the people you want to team up with). This makes team ranking more accurate than if you just take the sum of the players' individual ranks.

A real world analogy would be doubles tennis, two top-rated singles players can lose to a good doubles team that has lower individual ratings, because they play less well as a team. So it's relevant to look at past team performance.

knorke wrote:yea, these days everybody is on the dsd autohost, no need for manual host
---
If everybody only "plays primarily with people he knows", nobody will want to play with a new player because nobody knows him. This limits the playerbase to the core community that has been active for past few years. It creates an "inner circle" that is impossible to get into.

If by nobody you mean most people, and by impossible you mean difficult, I agree with you to some extent. The key word was "primarily" though - there should be a strong base of players you know, but I also feel that the system requires you play with "secondary" or "tertiary" players - those you don't know well and those you don't know. They just don't make up the majority of the game. Ideally, this would allow new players to come in, train up, and integrate. Of course, engineering the desired result is difficult.

To address your other assertions, few people want to play with a new player in any of the present rating systems either. Few established players want to play with a new player in any ranked game who isn't either a friend or so laughably low skill that they cannot hope to win.