!vote to trade players for team balancing

[MajBoredom]

Here is a spads / autohost / lobby server feature request that would greatly improve the quality of big team games and is not difficult to code.

The premise is that TrueSkill, for all its academic merits, will never be quite as good as player judgment when it comes to balancing the teams.

So, why not let people !vote to trade two players? Often an auto-balanced team lineup only needs a single player switch in order to arrive at a better balance. The players themselves are better at spotting smurfs and compensating for their skill. Since autobalance would still give an initial lineup for either team, this would be an interesting way of combining man & machine to balance the teams.

Some syntax like: !vote switch [DSD]Fan [COM]b0mbr

Presumably there is already code that lets you identify a player using short strings like just "Fan" or "COM" in the above example.

+1 or bump if you agree!

[gajop]

I don't play much, but I really don't think players should be doing the balancing: 1) it's a conflict of interest, 2) they overestimate their ability to judge their own skill as well as that of other players.

[Silentwings]

Yeah, I agree both things gajop said.

The algorithm spads uses is pretty well accepted as the 'best' out there, but if you want to try and write better - code it up, get some data on a few hundred games from the replay site and compare accuracy of predictions.

[MajBoredom]

How does it meet the definition of a conflict of interest? Do you mean: "an individual player is interested in a balanced game but may also just want to win." ... ? Voting tends to mitigate these individual interests since the voters can come from any of the teams.

If the players have to vote on the trade (perhaps with a high percentage quorum) then no individual player's interest is acting upon the balance -- it's rather in the quorum's interest.

Notice that the same "conflict of interests" (a fair balanced game vs "I want to win") occurs with map selection, starting boxes, and unit restrictions -- and all of these are handled effectively by player voting.

[Silentwings]

map selection, starting boxes, and unit restrictions -- and all of these are handled effectively by player voting.

Unit restrictions and startboxes are not normally handled by voting. Maps can be changed by votes (with stiff restrictions) but since maps are not really related to balancing I don't see what it has to do with this.

How does it meet the definition of a conflict of interest? Do you mean: "an individual player is interested in a balanced game but may also just want to win."

I assume that what gajop meant.

Voting tends to mitigate these individual interests since the voters can come from any of the teams.

Voting would do a bad job of reflecting the interest of the (in my estimation, large majority of) people who are disinterested and wish they could just get on and start the game...

As far as I can see the only question here is "is there anything major wrong with the teams the current algorithm generates" and, unles you can offer some statistics to say that there is, I'm sticking with no.

[MajBoredom]

As far as I can see the only question here is "is there anything major wrong with the teams the current algorithm generates" and, unles you can offer some statistics to say that there is, I'm sticking with no.

It's because of the ability to create smurf accounts which foil the algorithm. (For clarity's sake: this is when a strong player [e.g. 40 TS] creates a new account which is given 25 TS by the server.) I'm sure the algorithm compares favorably to other algorithms. But there are also times when the game history itself is not informative enough. Here, no algorithm based only on this information would produce a good balance.

This problem is worst in large team games (6v6 and up) because a single player's strength can be masked in several consecutive games. In statistical language, the "sample complexity" of an account's TS estimate is very high for large team games -- or, a larger number of large team games need to be considered before the algorithm can "tease out" an accurate TS value for an individual player. His skill alone plays a smaller role in determining the outcome than it would in, say, 1v1 or 2v2.

In the mean time, a 40 skill player can masquerade as a 25 skill player for quite a long time, foiling the algorithm. He may even still produce win/loss outcomes that are comparable to a 25 skill player, but only because he plays high-risk strategies which cost his team, or because he singles out another player (i.e.: "trolling"). In these situations, the 40 skill player may statistically appear like a 25 or 30 skill player.

Some would argue that this "statistically justifies" the 25 TS estimate. Also, they may say that it would justify the team balance that it creates. However, the smurf player will often play poorer strategies simply because he believes that the team is already stacked in his favour due to his undervalued TS score. If the game is less challenging for him, he may correspondingly play a weaker game. In this case, the undervalued TS score is "self-sustaining" even though it is inaccurate.

I think this produces poor games: strong smurf players play wild or errant strategies because they can buy themselves the insurance of a stacked team using an anonymous account.

In this set of cases, the majority players and spectators can often identify the smurf and also form a quorum against him in a vote, but at the moment they can do nothing and the smurf can abuse the algorithm (and any other algorithm based on game outcome) quite easily.

I'm sure that if a !vote switch option were available, most players would still rely on auto-balancing and the TS system for the vast majority of all games; it involves less effort. However, there are also times when a !vote switch option would be employed unanimously.

[Silentwings]

when a strong player [e.g. 40 TS] creates a new account which is given 25 TS by the server.)

You should speak to a host/lobby admin if you want to complain about a player smurfing. I don't think what you're suggesting would help there.

[dansan]

when a strong player [e.g. 40 TS] creates a new account which is given 25 TS by the server.

The system is really good at detecting smurfs.

[Jools]

At present you can achieve the same thing with two commands:

!force [DSD]Fan team i
!force [COM]b0mbr team j


You want to merge these two into one command I think. Why not. Can it be done with a plugin to spads? I think so.

It's always good to have more options, even if the TS balancing algorithm is better than NSA:s terrorism detection one :)

The problem with TS isn't smurfs, it's the fact that it's not possible to assess a player's performance entirely by quantitative means. People play differently depending on who they play with, and who they play against. In other words: the skill of a team is more than the sum of the skill of its individual players.

But the TS is a very good model. But a model is not the same as reality.

[MajBoredom]

Well, I didn't intend to turn this into a "TS-bash" thread -- I just think that the players should have an override option that works on top of any auto-balancing system. They wouldn't invoke it most of the time, but there are times when they would definitely want it for reasons that the auto-balancer can't predict by quantitative means.

I know that you could achieve the same result with several bot commands, but keep in mind that for non-admin users this means issuing several consecutive votes. Also, the game would be temporarily in a poor state (e.g.: 5v3 instead of 4v4) and possibly with the auto-balancer working against them as they conduct a sequence of votes. These are arguably the reasons why they never pursue this option even when they'd like to.