Skill based balancing instead of time based ranking (done)

[Licho]

Ended with game over. Crashed/existed games are not in stats.
Yeah i can add full lists etc or proper display of recorded games. In fact there was such list but it was lost when unknownfiles went to hell.

[knorke]

so when both players quit instead of hunting that last mex it doesnt count?
or is that ok because one player quits a moment earlier?

[Neddie]

SPADS would need to implement it themselves.

The system also penalizes me for trying to save doomed teams, a form of extreme self-handicapping, which is really the only reason to play BA with more than three people a side. It discourages efforts beyond the scope of the limited perception of the mathematical system. I'm not saying I'm a great player, I really don't grind enough time to be so, but a sub-1500 ranking is pretty damn misleading and the result primarily of behavior your system is too simple to understand. At the very least rankings should be stratified across different games and mods.

Anyway, I would like to remind the people capable of reading a sentence that at best you should take a ranking with a grain of salt, and generally you should ignore it in favor of your own judgment. I do like the opportunities afforded by this new balance option, I'm merely unwilling to ignore the flaws of it.

The statistician often laments the inability of the sociologist to understand the beauty of the model, the sociologist often laments the inability of the statistician to see weaknesses of the model.

[CarRepairer]

Okay, to follow up on what Neddie is saying, the ranking could be clarified as "likelihood to win based on skill and previous self-challenge." So one who often places himself in more challenging situations than the average user will be less likely to win, so this ranking doesn't purely reflect his skill.

[Licho]

Also if ranking is such a problem for you, i can make it private - or only displayed to you only and optionaly publishable.

[Gota]

This requires specific setings of games..
We need to encourage players to stick with one account and than make playing these "ladder" games streamlined..without the need to insert more passwords or sign in anywhere esle aside form the lobby.
So i can host a team game of SA which is either ladder or not ladder..
This way having "amateur" "fun" games and "pro" "ladder" games side by side would be easy...
Players like neddie will be able to play with their own goals and others will be able to play the ladder variants..

[Regret]

Also if ranking is such a problem for you, i can make it private - or only displayed to you only and optionaly publishable.
Also if ranking is such a problem for you, i can make it private - or only displayed to you only and optionaly publishable.

This.

[Neddie]

It isn't a problem for me, it is a problem when other people, as in most people, who don't understand the limitations of it use it. I'm trying to explain why and how rankings are flawed so nobody of sense makes the mistake of taking them seriously.

I play in the same games everybody else does, Gota, I just often play for different reasons. Why wouldn't I seek to buck a "pro" game?

Can you split it up by games? The overall ranking is by nature misleading; for instance, you're personally in the top ten purely on the virtue of CA games, which have no or little bearing on your skill in other games, which incidentally you seldom, if ever, play. Triton is scored primarily in BA, this does not reasonably present his ability in Kernel Panic. Etcetera, etcetera.

Are you going to consider my other balancing proposals for draft-based play?

[Gota]

Well all i can say is that ranking should be mod based..and hell if possible even map based as well.
It should just be available as interesting statistics..
Everyone can understand it as they wish..
It's not liek we have money prizes or something..

[==Troy==]

if it wasnt said before, account for the "points" _only_ if the game was balanced by the ranks, and dismiss the rest of the data. That way you eliminate cases where players try to stack/ antistack.

Freeallegiance uses this system btw.

[Neddie]

Does the !cbalance method take the ELO rather than the time into account at this point?

If he did that, Troy, then he would have to discard all of the original data, which may not be a bad thing.

[Gota]

I agree.
A system based on statistics needs time and a blank page.

[lurker]

No reason to throw it out at all. It's not perfect but it's a much better starting point than flat 1500.

[el_matarife]

These rankings are better than what we currently have, but I've got a few suggestions.

1. Split up rankings on a map basis: metal maps / sea maps / pure land maps or something similar based on the really different play styles that you get on those different types of maps. Someone may totally kick ass on Comet Catcher spamming Flashes but get totally owned by hovercrafts coming from the sea on Delta Siege.
2. Split out 1v1 rankings, team game rankings, and see if you can do a team comparison based on combination of players on a clan basis. For instance, certain clans may be used to playing together and thus have a higher win ratio with two or more of them on the same team. Also consider taking into effect if people are using your Teamspeak system or not. Teamspeak usage should make a team more effective, I think.
3. Consider rewarding players who win a team game "alone" more, for instance if they win after their whole team quits or "dies" if you can analyze for that in the replay. You could also analyze the replay to see if one player does more than say half or two thirds of the economy, or damage dealt, or units killed, or some combination of the three.

[Acidd_UK]

IMHO this is strictly better than the existing time-based rank based on all mods, and therefore is win GJ Licho!

[Neddie]

Often it is the person who absorbs the most damage in particular games and mods which makes the difference, not the person who deals the most. Especially in, say, BA.

As for Acidd's point, it really depends on whether you favour an unrealistic bell curve distribution to an unrealistic arbitrary distribution. Neither of them are particularly useful when people are involved, but I do agree that this is generally superior and can be even better for balancing in particular games if the ranks are game dependent.

[==Troy==]

as lurker said, use the initial data as take off point, but dismiss the new data when teams are not balanced by this system.

I have implemented this system for an FPS game, and it proved to be quite effective to some extent (I also suggested something similar on irc about 2 months ago).

as metafire suggested, but slightly altered :


at the moment the current system will do fine if the only new data is going to be the balanced cases.

As the amount of data increases, a lot of options come to mind, such as was suggested, look for combinations of players, if there is a deviation in performance of the player, per-map player statistics etc.


Even more, we dont actually need player stats here, collect _map_ statistics, how well it is balanced, tabula should give 1:1, something like high_and_low should be biased, etc. Once that data is known, allow balancing with respect to map disbalance, so more skilled players are placed in a less favoured part of the map. Which allows people to have enjoyable games even on not so well balanced maps.

This is really a very good base to collect a lot of statistics and get some real results out.

[knorke]

It should just be available as interesting statistics..
Everyone can understand it as they wish..

+1
its no use to try and calculate a superaccurate !balance that will always guarantue superfair games.
I would just like to see some stats to look at, like games played or average gametime or modstats or whatever is possible.

for some strange reasons larger games already take ages to start (no its not mapdownloading, everybody does have cc, dsd and the 3 other maps by now) so if you want to use it for !balance just keep it as simple as possible.

[Gota]

Big games are a result of community size..if we had 600 players peak youd get faster games...this will probably not happen if the whole spring experience isn't more streamlined.

[BaNa]

hey if its better than timerank (which it is) then cool. it would be cool if spads also used these and contributed stats. Imo damage dealt would be a cool thing to get within-team rank.