Some data on balancing  (Read 18669 times)

Flenser

  • Full Member 
  • *
  • Posts: 62
  • Country: us
Some data on balancing
« on: August 31, 2021, 21:49 »
It's my contention that blowout games that are wildly uneven are less fun than close games, where there's some back-and-forth. If you disagree, you can stop reading now.   :)

I got some spare time, and spent it writing an app to figure out if team balance is as bad as I thought, in VCTF games. I think I've crunched the numbers correctly, but I've attached the raw data if you'd like to crunch it a different way.

The app hits the miasma.rocks server twice a minute, watching for when the map changes, and records the scores. I let it run for about a week.

There were just under 700 games played. I removed 34 ONS games, and 11 VCTF4 games.

That left 613 games.

45 of them were 0 vs. 0, so I removed those.

That left 568 scoring games of VCTF. These counts are number of games, then red vs. blue scores.


61   0 vs. 5
103   5 vs. 0

2   0 vs. 4
12   4 vs. 0

47   1 vs. 5
52   5 vs. 1

4   1 vs. 4
11   4 vs. 1

That's a total of 292 games, or 51.4% of the VCTF games that were recorded.

To me, half the games played are boringly uneven. We need a team balancer.

One surprising thing that jumps out to me is that being on red is the way to go if you want to be on the winning team in a blowout game. I'm guessing this is due to the "default team" setting somehow. If anyone has a good theory on this, I'm interested in hearing it.



I also collected data for the Freon server, where I understand there is a team balancer of some sort. It gets a lot less action.

Out of 208 scoring games, only 5 games were 10 vs. 1 or 10 vs. 2. There were no 10 vs. 0 games at all.

hagis

  • 1337
  • *
  • Posts: 404
  • Country: gb
Re: Some data on balancing
« Reply #1 on: August 31, 2021, 23:02 »
freon also have competition mode set to on,. which lowers the starting health and shield for the leading team for each round (the more uneven the score the less health the leading team gets)

though the balancer does a lot to help

personally I agree,. if you know your team "can't win" it's hard to get that motivated and is less fun

Stealer

  • 1337
  • *
  • Posts: 277
  • Country: gb
    • 76561197992159424
    • StealerNinja
Re: Some data on balancing
« Reply #2 on: September 01, 2021, 22:04 »
Aye, agree with Hagis. The competition mode makes it harder and harder for a blowout game.

Shows the balancer functions.

Not sure about motivation from my perspective, I always try to join the losing team if I'm jumping in mid match, although I do let the balancer do it's thing. These days I just celebrate hitting something other than a wall :)  ??? 8)
Stealer.T32

Piglet

  • 1337
  • *
  • Posts: 3169
  • Country: gb
Re: Some data on balancing
« Reply #3 on: September 02, 2021, 09:00 »
The_Cowboy started work on a balancer as a background thing, but I don't kniow how far he got with it.

Freon works by:

1.  storing "points per round" to an external database,
2.  at the start of each game (or player join) it gets an average value from the external datbase
3.  Freon is (normally) 10 rounds of play. At  start of round 1, or at the start of a round followng (admin calls for it, or a high value player joins) the balancer tries to ensure that the two teams' average points per round is as close as it can get it.


The problem with this is that you get points in Freon for thawing and fragging. Players can be good at one or the other or both.  You can end up with a team of thawers against a team of fraggers, or an imbalance (mid-game) of one excellent player with a group of low-scoring players against a team of medium scorers.

Another problem is that some maps go on a very long time with large numbers of players getting thawed, so points per round can be very high on those maps, raising average points way above what you'd get with fewer players on shorter maps.

If you'd like to have a go at coding a better balancer for vCTF, go for it. Maybe liaise with The_Cowboy and pool resources.



THOR`

  • Sr. Member
  • *
  • Posts: 142
  • Country: gb
    • anxious_f0x
Re: Some data on balancing
« Reply #4 on: September 02, 2021, 09:37 »
I feel like one of the biggest factors that team balancers can’t really account for, especially in objective based games modes is good teamwork.

Individual points in modes like ONS and VCTF don’t necessarily translate into victories so how do you balance players that take on support roles that may not offer up the most points but can dictate the result of a match?

I honestly really dislike team balancers as they’re never going to be perfect and they seem to cause nothing but complaints.

Piglet

  • 1337
  • *
  • Posts: 3169
  • Country: gb
Re: Some data on balancing
« Reply #5 on: September 02, 2021, 09:47 »
Maybe balance based on proportion of wins to loses? You may be great and unlucky or poor and lucky with your team.....but maybe averaging out winners and losers could be a strategy.

I think that might help a little.

No balancer will ever be perfect. A perfect balancer would be awful: stalemate.

I see them as a little nudge in the right direction.
« Last Edit: September 02, 2021, 09:52 by Piglet »

hagis

  • 1337
  • *
  • Posts: 404
  • Country: gb
Re: Some data on balancing
« Reply #6 on: September 02, 2021, 10:37 »
Balancer is not perfect for many reasons,. one example, some days people play better than others,. freon balancer lets you mix the current match ppr score with that players avg over a number of matches which helps a little

does it help to have a balancer,. well that's opinion on what matches should look like,. every match ending 9-10 or 10-9 is not the goal to me but to create a playing field where at least there's a possibility for a win from either side

core freon server currently has no balancing (no one to balance and people don't typically switch and no automated ppr balance) - matches there are stupid,. mid way through a match that was actually fairly even few players join and the outcome is known the moment they join,. just as well call it then and hope the next match randomly gives a better balance

Piglet

  • 1337
  • *
  • Posts: 3169
  • Country: gb
Re: Some data on balancing
« Reply #7 on: September 02, 2021, 11:33 »
core freon server currently has no balancing (no one to balance and people don't typically switch and no automated ppr balance) - matches there are stupid,. mid way through a match that was actually fairly even few players join and the outcome is known the moment they join,. just as well call it then and hope the next match randomly gives a better balance

We definitely see that on the vCTF server too. One glance at the player list and sometimes you know it's going to be a short brutal boring game.


The_Cowboy

  • 1337
  • *
  • Posts: 259
  • Country: in
  • CodeZilla
Re: Some data on balancing
« Reply #8 on: February 01, 2022, 00:51 »
Hey Flens, how about running this puppy again and spit out the statistics, now!
Quote from: Wormbo
You learn UnrealScript mainly by reading other people's code. Removing code without an important reason (download size reduction and lack of helpfulness are not important in that sense) is extremely antisocial IMHO.

Flenser

  • Full Member 
  • *
  • Posts: 62
  • Country: us
Re: Some data on balancing
« Reply #9 on: April 16, 2022, 16:41 »
Ah, should have put this data in this thread rather than the balancing one.

Here's the data my app collected over the past couple of weeks.

Out of 1532 VCTF games that had scores, 734 were 5-0, 5-1, 4-0 or 4-1, or  48%.

This is only slightly better than before the balancer was added, when it was 51%.

I don't know if 3% is within sampling error or whatever. I am surprised that it's only 3%.

Can this be explained by perhaps not enough data-gathering by the balancer? Or because then balancer doesn't balance when new people join?

  16 0 vs. 1
  13 0 vs. 2
   8 0 vs. 3
  21 0 vs. 4
 138 0 vs. 5
  24 1 vs. 0
  23 1 vs. 2
  18 1 vs. 3
  11 1 vs. 4
 101 1 vs. 5
  12 2 vs. 0
  27 2 vs. 1
  20 2 vs. 3
  26 2 vs. 4
  84 2 vs. 5
  27 3 vs. 0
  22 3 vs. 1
  40 3 vs. 2
   1 3 vs. 3
  23 3 vs. 4
  68 3 vs. 5
  30 4 vs. 0
  24 4 vs. 1
  29 4 vs. 2
  23 4 vs. 3
   2 4 vs. 4
  66 4 vs. 5
 272 5 vs. 0
 137 5 vs. 1
  83 5 vs. 2
  81 5 vs. 3
  62 5 vs. 4


Piglet

  • 1337
  • *
  • Posts: 3169
  • Country: gb
Re: Some data on balancing
« Reply #10 on: April 16, 2022, 16:57 »
Probably because people join or change team part way through.

The_Cowboy

  • 1337
  • *
  • Posts: 259
  • Country: in
  • CodeZilla
Re: Some data on balancing
« Reply #11 on: April 17, 2022, 01:42 »
Thanks for statistics. I'd certainly count that as progress (I am no data scientist though).
As you and Piglet have observed, there are several open ended scenarios where, once teams are balanced, fluctuations in the form of team-change, team-leave/join may enter. You are welcome to look into the algorithm and suggest. I just hope that it won't be a forceful experience for players.
Cheers mateys!

Side note: I have been working on WebAdmin functionality. Not really sure how much that would facilitate balancing mid-game. That is on hold because I have resumed working on my game Engine (the reason why I have been absent on Warfare recently).
« Last Edit: April 17, 2022, 01:46 by The_Cowboy »
Quote from: Wormbo
You learn UnrealScript mainly by reading other people's code. Removing code without an important reason (download size reduction and lack of helpfulness are not important in that sense) is extremely antisocial IMHO.

Absolute_Madness

  • 1337
  • *
  • Posts: 598
  • Country: fr
Re: Some data on balancing
« Reply #12 on: April 19, 2022, 18:18 »
few past days i alway ended up vs a stacked team. This equalizer is acting weird sometime.

The_Cowboy

  • 1337
  • *
  • Posts: 259
  • Country: in
  • CodeZilla
Re: Some data on balancing
« Reply #13 on: April 20, 2022, 02:23 »
Well it is really difficult to separate out the balance-quits (similar to rage quits, but quitters quite because of, well....) and woopsie of Equalizer. So I'd be careful before making up my mind.
Quote from: Wormbo
You learn UnrealScript mainly by reading other people's code. Removing code without an important reason (download size reduction and lack of helpfulness are not important in that sense) is extremely antisocial IMHO.

The_Cowboy

  • 1337
  • *
  • Posts: 259
  • Country: in
  • CodeZilla
Re: Some data on balancing
« Reply #14 on: April 20, 2022, 02:24 »
Well it is really difficult to separate out the balance-quits (similar to rage quits, but quitters quit because of, well....) and woopsie of Equalizer. So I'd be careful before making up my mind.
Quote from: Wormbo
You learn UnrealScript mainly by reading other people's code. Removing code without an important reason (download size reduction and lack of helpfulness are not important in that sense) is extremely antisocial IMHO.