independent 'clan' or 'skill' ts count for better balance

independent 'clan' or 'skill' ts count for better balance

Requests for features in the spring code.

Moderator: Moderators

Post Reply
User avatar
MelKor
Posts: 38
Joined: 15 Oct 2010, 16:05

independent 'clan' or 'skill' ts count for better balance

Post by MelKor » 20 Sep 2017, 15:22

First I'll explain the problem:

Some experienced players usually play in teams to get an unfair advantage from standalone ones, with the objective to increase personal TS and troll everyone, many times with smurfs collaborators. Is a source of imbalance in mid/long term and keep away many players when they detect the stack and discourage new players...

I don't want to avoid mixed games only as teamed or alone players, but a kind of control must be implemented because this is a problem.

I propose to add a parallel TS for team players, I mean, keep 2 TS list for all players, one as always, an 'alone TS' updated when the server is configured to play alone(skill mode for example), and other 'team TS', when a is configured to play in a team ('clan' mode). The server selects and updates the 'alone TS' or 'team TS' given a configuration on the server, being the actual option 'clan' or 'skill' able to be used for this.

A server configured as a team ('clan'):
.- As is now, the server allow any player but give prority to clan then skills forbalance using their 'team TS', and the server always updates 'team TS' if they play in teams (which is the fact, they are playing as a team) avoiding the clan stack to increase alone TS.
.- If there are players playing alone, use the 'alone TS' for balancing and update the 'team TS' at the end.

A server configured as 'no team' (skill):
.- by default this configuration only use 'alone TS' forbalance and update 'alone TS' at the end of the game. In this way, we allow a good balance for a 'skill' mode game, but not allowing to use of 'team TS', avoiding the clan games modify the skill mode balance.

In resume, I just propose a clear separation between clan and skill modes using a parallel TS list, for a fear balance given the kind of game to play.

In this way, we solve problems like:

1.- Abuse of experience troll players as or with smurfs.
2.- Decrease the imbalance in games in mid/long terms (games be more balanced in time).
3.- Discourage using teams games (usually cheaters too) to increase personal TS.
4.- Discourage cheat.
5.- Allowing newbs to play more fair games.
6.- In this way player without a team can still playing but knowing where the TS will go.

As in any game, teams are a good stuff, people like to play with some players more than others, but play in this way gives an advantage which should be managed for sake of a good and fear balance.

MK
Last edited by MelKor on 20 Sep 2017, 15:45, edited 3 times in total.
0 x

User avatar
triton
Lobby Moderator
Posts: 329
Joined: 18 Nov 2009, 14:27

Re: independent 'clan' or 'skill' ts count for better balance

Post by triton » 20 Sep 2017, 15:30

I think I saw more games where I can't play with my teamspeak friends, than contrary.
From what I know when you use Cbalance command, players can't be with their team/clan if the balance deviation is too high, am I wrong?
0 x

User avatar
MelKor
Posts: 38
Joined: 15 Oct 2010, 16:05

Re: independent 'clan' or 'skill' ts count for better balance

Post by MelKor » 20 Sep 2017, 15:34

triton wrote:I think I saw more games where I can't play with my teamspeak friends, than contrary.
From what I know when you use Cbalance command, players can't be with their team/clan if the balance deviation is too high, am I wrong?
I changed the proposal, read again, please.
0 x

User avatar
Silentwings
Moderator
Posts: 3578
Joined: 25 Oct 2008, 00:23

Re: independent 'clan' or 'skill' ts count for better balance

Post by Silentwings » 20 Sep 2017, 16:32

Some experienced players usually play in teams ... Is a source of imbalance
I am not sure if you're correct, but there is an easy way to check if this effect really exists: compare the win % of players with over, say, 30TS to the win % of clans containing exclusively players with over 30TS. (I am not sure how easy it is to pull the necessary data from the replay site, though.)

Regardless, I am sure that games containing one big clan on a single team have a different character to "normal" games, and I can understand why the un-clanned players would not to want them there.
a kind of control must be implemented
If it was going to be done, I think a better solution is a compromise balance mode, call it "ts + clan preference", which, on each re-balance, would respect clans but
(i) limit the max clan size to be = the current team size divided by (say) 3 and rounded up
(ii) disallow any clan whos average TS was outside of (say) 1 std dev from the mean TS of all present players
Or suchlike. Any clans failing to satisfy (i) would be split up into equal sized pieces on separate teams until they did (or, failing that, disbanded entirely), and any clans failing to satisfy (ii) would be disbanded entirely.

[I think your suggestion of having separate TS systems for TS-clan and plain TS balance modes is not a good idea. It lowers the learning rate heavily, due to less games being fed into each system. Also, it would sometimes result in people who had never clanned having different scores in each system (especially new players) and this would be ridiculed by people who aren't familiar with Bayesian learning, perhaps even abused by people trying to avoid whichever system has given them a higher score.]
when you use Cbalance command, players can't be with their team/clan if the balance deviation is too high, am I wrong?
You're right, but I think it only kicks in for extreme cases. It also wouldn't pick up the issue that's being claimed by the OP - because the OP claims that good players are artificially stronger, as a consequence of being together in a clan, than their individual TS scores would suggest, and its the individual TS scores that are used to calculate the balance deviation.
0 x

User avatar
MelKor
Posts: 38
Joined: 15 Oct 2010, 16:05

Re: independent 'clan' or 'skill' ts count for better balance

Post by MelKor » 20 Sep 2017, 18:35

Silentwings wrote:
Some experienced players usually play in teams ... Is a source of imbalance
I am not sure if you're correct, but there is an easy way to check if this effect really exists: compare the win % of players with over, say, 30TS to the win % of clans containing exclusively players with over 30TS. (I am not sure how easy it is to pull the necessary data from the replay site, though.)

Regardless, I am sure that games containing one big clan on a single team have a different character to "normal" games, and I can understand why the un-clanned players would not to want them there.
The full context is needed to understand the request, is not only an experienced player in a team, is an experienced player, sometimes with smurfs accounts in a team, cheating and trolling. As results, many players just don't play in this condition and many times, when the game has just few players people just don't play with real newbs in game because this results too unbalanced.

Even if there is no cheating or smurfing, is a common sense that the team and alone games must be treated different, given the advantage, and this proposal does that, probably modifying some details, but keeping the TS splited.
Silentwings wrote:
a kind of control must be implemented
If it was going to be done, I think a better solution is a compromise balance mode, call it "ts + clan preference", which, on each re-balance, would respect clans but
(i) limit the max clan size to be = the current team size divided by (say) 3 and rounded up
(ii) disallow any clan whos average TS was outside of (say) 1 std dev from the mean TS of all present players
Or suchlike. Any clans failing to satisfy (i) would be split up into equal sized pieces on separate teams until they did (or, failing that, disbanded entirely), and any clans failing to satisfy (ii) would be disbanded entirely.
Statistics to work require a reference, which we don't have, and many hours of work and people with time to test a sollution.
Silentwings wrote: [I think your suggestion of having separate TS systems for TS-clan and plain TS balance modes is not a good idea. It lowers the learning rate heavily, due to less games being fed into each system. Also, it would sometimes result in people who had never clanned having different scores in each system (especially new players) and this would be ridiculed by people who aren't familiar with Bayesian learning, perhaps even abused by people trying to avoid whichever system has given them a higher score.]
when you use Cbalance command, players can't be with their team/clan if the balance deviation is too high, am I wrong?
You're right, but I think it only kicks in for extreme cases. It also wouldn't pick up the issue that's being claimed by the OP - because the OP claims that good players are artificially stronger, as a consequence of being together in a clan, than their individual TS scores would suggest, and its the individual TS scores that are used to calculate the balance deviation.
Split the TS in the way as I propose solve the problem from the root, eliminating the perverse incentive which is to mix the team and personal TS, given the advantage of the team game, the details of the implementation can be modified from the original proposal, but the core of the idea is good as I understand how other games solve this problem.

Plus, in a second thought, some games have 'team' and 'standalone' modes complete separated, and probably a good reason is to avoid this kind of problems. I'm not suggesting to do the same, spring have too few players to do that.

MK
0 x

User avatar
Silentwings
Moderator
Posts: 3578
Joined: 25 Oct 2008, 00:23

Re: independent 'clan' or 'skill' ts count for better balance

Post by Silentwings » 20 Sep 2017, 18:44

As far as I can tell you've said nothing that reduces (or even mentions) the two concerns I had.
Statistics to work require a reference, which we don't have, and many hours of work and people with time to test a sollution.
It doesn't, good thresholds for my (i) and (ii) can easily be set by an experienced person, with no calculation required.

As I said, I rate your own solution as poor from a statistical point of view, because it (approximately) halves the rate of learning. A "proper" statistical solution would involve modifying the belief propagation used in TS to handle hierarchically structured teams - that would require hours of work and is not worth the effort.
0 x

User avatar
MelKor
Posts: 38
Joined: 15 Oct 2010, 16:05

Re: independent 'clan' or 'skill' ts count for better balance

Post by MelKor » 20 Sep 2017, 19:29

Silentwings wrote:As far as I can tell you've said nothing that reduces (or even mentions) the two concerns I had.
Statistics to work require a reference, which we don't have, and many hours of work and people with time to test a sollution.
It doesn't, good thresholds for my (i) and (ii) can easily be set by an experienced person, with no calculation required.

As I said, I rate your own solution as poor from a statistical point of view, because it halves the rate of learning. A "proper" statistical solution would involve modifying the belief propagation used in TS to handle hierarchically structured teams - that would require hours of work and is not worth the effort.
I don't see the concerns so clear, so please clarify.

Please help me to understand how the rate of learning will be delayed, the game already is painful for newbies as it, indeed can't be worst (for BA at least), and the idea is to help new people to play more fair games.

Respect to the statistics, what about to keep the algorithms and only pick one list of the other (alone or team TS) at the moment to balance the games?, because as I understand the algorithm balance the TS from a map list linking the players with the TS list, you can replace that TS list from two options given the kind of game, 'clan' or 'skill', simple as that.

As I understand, involve complex statistics will be only a patch for a moment in the TS, and probably must be reviewed in time, so must be avoided. If I'm wrong, let me know where to get the information to talk properly.

MK
0 x

User avatar
ThinkSome
Posts: 276
Joined: 14 Jun 2015, 13:36

Re: independent 'clan' or 'skill' ts count for better balance

Post by ThinkSome » 20 Sep 2017, 19:42

Here is an even better idea: Why not hide trueskill values entirely? Have autohosts submit player lists to SLDB for balancing. Then noone would know what their trueskill is and there would be noone trying to inflate their personal trueskill.
0 x

User avatar
FabriceFABS
Posts: 345
Joined: 28 Jul 2010, 16:20

Re: independent 'clan' or 'skill' ts count for better balance

Post by FabriceFABS » 20 Sep 2017, 20:31

Why not finally.

I repeat again and again that TS was NOT created in regard to know who is the best, but created to offer a SO-better balancing system in removal of the oldy one that used to balance teams regards to players' time of spec/play hours.

Most of players here play only for make his own TS growing, before anything else.
Most of the players gives so much TS importance, too much IMO, really too much.
0 x

User avatar
Silentwings
Moderator
Posts: 3578
Joined: 25 Oct 2008, 00:23

Re: independent 'clan' or 'skill' ts count for better balance

Post by Silentwings » 20 Sep 2017, 22:52

Please help me to understand how the rate of learning will be delayed,
You are storing two TS scores for each player, but only updating one of these scores in every game that is played. This means that each of the TS scores, individually, will be updated less often, meaning that the TS algorithm will take longer to learn (a good estimator of) a players true ability.
0 x

User avatar
MelKor
Posts: 38
Joined: 15 Oct 2010, 16:05

Re: independent 'clan' or 'skill' ts count for better balance

Post by MelKor » 20 Sep 2017, 23:38

Silentwings wrote:
Please help me to understand how the rate of learning will be delayed,
You are storing two TS scores for each player, but only updating one of these scores in every game that is played. This means that each of the TS scores, individually, will be updated less often, meaning that the TS algorithm will take longer to learn (a good estimator of) a player's true ability.
OK, I understand now.

I'll explain in a different way to show my point clearly, hope you guys get it:

Players in a team game, play differently as alone. By induction, if you put the results of the games in different TS lists (clanTS and skillTS for example), using same algorithm (TS in this case) will show different values. Then we can say that a player has two different statistic context, otherwise playing alone or stacked will give same results, which is not the reality.

All we'll agree until here, no problem to understand, right ?

Now, you suggest to mix the results of clan and skill modes to the same algorithm in the same list and pretend to get the same results. That means to mix two different statistic context and pretend to get same results. The TS algorithm just can't get same results, except if the player has the same behavior alone or in a team, and that is not the case as I wrote before. So mix the data from different modes is a statistical mistake.

In a second thought, probably that's why people complain so much about balance in games, which suggest the TS have a problem or the problem is the data source, in this case, the data come from different statistic contexts.

As you say, the algorithm will converge faster (not sure how fast indeed, in a matter of weeks will be no difference, I bet), but the data comes from different contexts, biasing the statistics, I suggest to split the statistics given the contexts, in this way the TS algorithm will work much better.. slower, but with good data.

Oh god, this post took me a lot of time to write, I did my best, hope you guys get my point.

MK
0 x

klapmongool
Posts: 843
Joined: 13 Aug 2007, 13:19

Re: independent 'clan' or 'skill' ts count for better balance

Post by klapmongool » 21 Sep 2017, 08:33

ThinkSome wrote:Here is an even better idea: Why not hide trueskill values entirely? Have autohosts submit player lists to SLDB for balancing. Then noone would know what their trueskill is and there would be noone trying to inflate their personal trueskill.
Because of smurfs. When TS is hidden you can't identify those anymore. Also sometimes you want to get a game started with an uneven number of players (cos you are done waiting for more). Players in that scenario want to have an idea what they are up against, before agreeing to it.
0 x

User avatar
ThinkSome
Posts: 276
Joined: 14 Jun 2015, 13:36

Re: independent 'clan' or 'skill' ts count for better balance

Post by ThinkSome » 21 Sep 2017, 14:10

klapmongool wrote: Because of smurfs. When TS is hidden you can't identify those anymore. Also sometimes you want to get a game started with an uneven number of players (cos you are done waiting for more). Players in that scenario want to have an idea what they are up against, before agreeing to it.
But you will know what you are up against. You will know that the teams are as balanced as they can be wrt. to TrueSkill.
0 x

klapmongool
Posts: 843
Joined: 13 Aug 2007, 13:19

Re: independent 'clan' or 'skill' ts count for better balance

Post by klapmongool » 22 Sep 2017, 09:57

ThinkSome wrote:
klapmongool wrote: Because of smurfs. When TS is hidden you can't identify those anymore. Also sometimes you want to get a game started with an uneven number of players (cos you are done waiting for more). Players in that scenario want to have an idea what they are up against, before agreeing to it.
But you will know what you are up against. You will know that the teams are as balanced as they can be wrt. to TrueSkill.
No;

- smurfs are players that show up with low (or starting) TS while they are actually high TS (because they use proxies or other means to evade being matched to their high TS account). Thus the system can't take their actual skill into account when balancing.

- small games with an uneven number of players don't balance out very well at all. There is a massive difference between one 40TS vs two 35TS players or one 40TS vs two 20TS players. Hide the TS and people won't know if they end up in a game that is worth it.
0 x

User avatar
ThinkSome
Posts: 276
Joined: 14 Jun 2015, 13:36

Re: independent 'clan' or 'skill' ts count for better balance

Post by ThinkSome » 22 Sep 2017, 12:52

And how would TS help you in determining that these are smurfs? Though if you don't want smurfs then you can just impose minimum rank to join.
0 x

sprunk
Posts: 83
Joined: 29 Jun 2015, 07:36

Re: independent 'clan' or 'skill' ts count for better balance

Post by sprunk » 22 Sep 2017, 19:46

You can have the system output rough balance quality without revealing precise TS values.
0 x

Post Reply

Return to “Feature Requests”