Loading...
  OR  Zero-K Name:    Password:   

Suggestion to change the team balancer

30 posts, 731 views
Post comment
Filter:    Player:  
Page of 2 (30 records)
sort
I am told that the team balancer matches total elo on both sides.

I think this is a bad way of doing it.

I think it should distribute the lowest ranked players evenly across sides and then put the rest, split as evenly possible at the top.


It is 100% imposible to win a match with one or two high ranked players and a bunch of low ranks vs all golds.

Outside of extraordinary circumstances, no one player in a rts can win vs 2 players of only slightly lesser skills. Being in a team with a bunch of low ranks is just... instant resign.
+9 / -0

3 months ago
I second this, balancing on avg elo is horrible. It's what makes me avoid large teams bc of how unbalance they can be.
+0 / -0


3 months ago
quote:
I think it should distribute the lowest ranked players evenly across sides and then put the rest, split as evenly possible at the top.


Can you be more specific and fully describe your algorithm? e.g. how are you defining "lowest ranked players"? The bottom 2? Bottom 10% rounded up? Those more than 300 elo bellow the median player for that game?
+2 / -0
3 months ago
While I agree that improvements might be possible the hard part is deciding on what you want to improve, making a metric for that and testing it is in fact better than the current status.

My favorite idea is (which I am too lazy to test as I said above): sort players in order of WHR, put best player in team A, second best in team B and so on. This would solve/improve complaints like "they have 5 blues and we have only 1 blues" as the difference will always be of 1 for any rank.

I feel balancing is bad not that much because of actual team composition but on how people decide to play that game (there are both low level and high level trolls that one game play great and another game they play horrible), and doubt there is any real solution to that.
+2 / -0
I don't have deep dive knowledge of how it works, but my understanding is that it tries to make 2 teams that have the same sum of elons.

What I am suggesting is more along the lines of take lowest ranked, put them in 1. Them 2nd lowest ranked, put them in 2. Then take third lowest rank, put them in 1... and so forth until all players are in either 1 or 2.

Then, if the total is very different, shift players from the top until the sums are close to equal.



What we often get, especially in small teams, is like 2 blue 1 silver 2 brown vs 4 gold and one silver. The difference here would be that all the lowest ranked players are not on the same team.
+2 / -0
3 months ago
Even if I agree your algorithm sounds good, I don't think it should be implemented live unless there is more analysis done (like a clear objective and measure of what it improves).

Not to mention the other problematic cases (like odd number of players) that must be thought of. For an example of analysis for such a corner case you can see http://zero-k.info/Forum/Thread/36555 . And still, as far as I am aware, this mathematically demonstrated solution was still not implemented.

But if you feel inspired you can use the code at that page to run your own analysis.
+1 / -0

3 months ago
At least under some conditions the balancer does try to match the standard deviation in the ratings of both teams, as well as the average.

The balancer is also supposed to put party members / clan members on the same team if it can. We could argue about whether this is desirable, it has been discussed in the past.

I am inclined to agree that the best way to approach this problem is to start by defining an objective rather than an algorithm.
+10 / -0

3 months ago
I would agree that the balancer seems to not know what it's doing, except that there are many times when what seems to me like bad balance turns out to be the opposite of what I expect.

So far no replays have been posted, but here's a few examples of times when I started the game thinking exactly along the lines of what CArankGalamesh is saying; something like "what is this balance? We are down one player, our highest ranked player is silver, and the enemy's highest ranked player is blue!" And yet, we still won. Here's that battle specifically, which is one that I just now found looking through my last 30 or so games. Not only that, one of the games that I played on the same day had nearly the same arrangement of teams and the opposite team won, which makes a lot of sense when what FRrankmalric is saying is true:
quote:
I feel balancing is bad not that much because of actual team composition but on how people decide to play that game (there are both low level and high level trolls that one game play great and another game they play horrible), and doubt there is any real solution to that.

I suspect this is true not only for specific "trolls" but on a smaller scale for all players. We all play very differently from game to game and the balancer has no way to predict that.
Here's two more examples of battles that show balance that seems unfair but the results are unexpected:
https://zero-k.info/Battles/Detail/1721021
https://zero-k.info/Battles/Detail/1739432
+3 / -0

3 months ago
It's very hard to reason about balance after battles have happened, because the given ratings and probabilities use information from the future. I can't imagine anyone thinks this is the right way to do it, it's just convenient.

I agree that you'd need to define the objective of the balancer. I don't think it should be for teams to be as equal as possible. For example, it should help people win a game if they've had a run of losses (I think it does this).
+1 / -0

3 months ago
Even if subjective experience is deceiving and the wonky-looking games the balancer sometimes produces are actually balanced from the perspective of winrate, it is still worth examining whether modifications could produce more fun matches.
+2 / -0
3 months ago
I think the greatest flaw of the balancer is actually how it handles uneven team counts, especially in small games. Having an extra com and fac under your control doesn't help much next to the level of attention-splitting you need to do in a 1v2 or 2v3. It's less bad once you get to 5v6 or higher, and in the largest games it's pretty unnoticeable. But those small games? Ouch.
+0 / -0

3 months ago
quote:

I feel balancing is bad not that much because of actual team composition but on how people decide to play that game (there are both low level and high level trolls that one game play great and another game they play horrible), and doubt there is any real solution to that.


Agree on some things but not others. I don't think the balance is bad, it just feels bad sometimes because
- players are inconsistent
- game's natural rock-paper-scissors may make people win or lose hard depending on what they run into
(namely cloak stuff and air which can show up somewhere fast or with no warning, relatively early)
- clueless new players VS "new" players that actually have 5+ years rts experience (some many years in similar games)

+2 / -0

3 months ago
"all welcome" sounds nice, but perhaps there could be a rule that prevented really new players from joining the lobsterpot, requiring "red" rank or above.

+1 / -0
3 months ago
quote:
"all welcome" sounds nice, but perhaps there could be a rule that prevented really new players from joining the lobsterpot, requiring "red" rank or above.
Would go the other "end". Make a "stricter" Palladium with limited number of players IF there is a "Teams All Welcome" with a certain number of players. Like "when there are more than 32 players waiting or playing teams all welcome you can have a 4v4 with a minimum ELO of 2xxx.".

That would not reduce as fast the number of players, while still allowing high ranked players to have team games.

But probably is too hard to implement for small benefit. Many high ranked players seem to like lobpot as well.
+0 / -0
At the moment (as far as I know), anybody at any time can create a 4v4 room with a minimum elo of whatever you like. So it could be done manually.
+0 / -0

3 months ago
Another suggestion would be to do something like league of legends

People would select desired role on their client : economy builder, air, front line.

The balancer/match maker would then take it into account
+2 / -1
3 months ago
[Quote]I am told that the team balancer matches total elo on both sides.

I think this is a bad way of doing it.

I think it should distribute the lowest ranked players evenly across sides and then put the rest, split as evenly possible at the top.


It is 100% imposible to win a match with one or two high ranked players and a bunch of low ranks vs all golds.

Outside of extraordinary circumstances, no one player in a rts can win vs 2 players of only slightly lesser skills. Being in a team with a bunch of low ranks is just... instant resign.[Quote]

I agree 100%. There are some games that not a single person on earth could win, because the low elo players make it impossible. Its like having your own teamates work against you.
+1 / -0

3 months ago
quote:
People would select desired role on their client : economy builder, air, front line.

The balancer/match maker would then take it into account

I guess that could be automated. Have multiple WHR systems run in parallel, but only one is based on winning the match. Others could quietly award pass/fail based on other performance metrics (in the upper 50% for destroying stuff, for example). The final balancer would then have multiple dimensions to consider.

Who knows. I'm not too upset by the current balancer, it provides a lot of variety.
+0 / -0

3 months ago
Please do not legitimize plopping in the back line.
+3 / -0

2 months ago
A request for people posting replays to prove that the balancer did a bad job: Please load the replay and report their numerical ratings, rather than just saying "look 3 silver and bronze vs 4 yellow!" It's very common for a Yellow ranked player to have a higher rating than a Silver player, so looking at rank color alone isn't a good way to make your point. Next, for each of these games please also post your suggested team balance; I suspect the number of games where you can do a better job than the balancer manually will be diminishingly small.

Finally, instead of simply being told how the balancer works, it might be a good idea to look at what it's actually doing instead of speculating (see Code Here).
+2 / -0
Page of 2 (30 records)