Balancer Test Results!!

General Comments, Questions about all things OmnipotentS that don't go in other topics/forums
User avatar
EmanReleipS
Posts: 42
Joined: Wed Jun 09, 2021 1:11 am

Re: Balancer Test Results!!

Post by EmanReleipS »

pooty wrote: Sat Jul 24, 2021 6:17 pm I've added this page, its not quite complete
https://www.omnipotents.com/utstats/onsrankings.php
(Ignore the actual rankings..they don't mean much on this page, just look at the details)

Cores Destroyed
Nodes Destoryed (and per hour)
Nodes Constructed (and per hour)
Constructing Nodes Destroyed (and per hour)

I think I will add these to the main player page ( https://www.omnipotents.com/utstats/ind ... ts=players ) since it already has sorting built in.

I did a "composite score" summed up the per hour numbers...and it looks pretty good. (I think that page needs filtered on at least 60 mins of play), but to me its a better "rating" that anything else we have.
Ooooh, this is a really nice page! How did you build this? I think it could make a nice addition to CEONSS' stats page as well.
User avatar
pooty
Posts: 4351
Joined: Sat Apr 03, 2021 10:22 am
Location: Michigan
Server Sponsor: Yes
Server Admin: Yes

Re: Balancer Test Results!!

Post by pooty »

Where the hell am I on this list? McL)o(vin
Did you change your name on line? Or get stuck with PlayerNN? I think it updates the name if you change it, we can double check after you play a match or two.. I know you were in there...
Ooooh, this is a really nice page! How did you build this? I think it could make a nice addition to CEONSS' stats page as well.
Just copied the rankings.php and modified the code. I had to update the underlying SQL queries to pull the additional data. I can post up the source once I get it finished up. I think adding those to the main player page is a good thing too--not sure what version of StatsDB you're running, as the main page is driven of index.php which does a bunch of things.
User avatar
Enyo
Posts: 1622
Joined: Mon Apr 05, 2021 11:27 pm
Server Sponsor: Yes
Server Admin: Yes

Re: Balancer Test Results!!

Post by Enyo »

pooty wrote: Sat Jul 24, 2021 6:17 pm I've added this page, its not quite complete
https://www.omnipotents.com/utstats/onsrankings.php
(Ignore the actual rankings..they don't mean much on this page, just look at the details)
Nice work pooty! That last Hour Sum column seems to be a very good ranking of players, minus a small handful of anomalies. This whole table really exposes who goes to nodes and who doesn't. I think this is definitely heading in the right direction.

Now, how do we get the balancer to use that for shuffling teams?
“Never argue with stupid people, they will drag you down to their level and then beat you with experience.”
― Mark Twain
User avatar
pooty
Posts: 4351
Joined: Sat Apr 03, 2021 10:22 am
Location: Michigan
Server Sponsor: Yes
Server Admin: Yes

Re: Balancer Test Results!!

Post by pooty »

To get the balancer to use that, I have to write/change the EvenMatch mutator to read from either the database or a file of recent stats. Its probably going to have to be file based (current one is), so I have to figure out an automated way to generate that file from the stats database and send it to the game server. Problem is I don't have full control of the game server..maybe a php script to sftp it to the game server (which is the opposite of what the stats thing does.. it transfers a file from Gameserver to stats server. So the code is there..if most of you don't know pretty much any UT mod/mutator/vehicle can be decomplied into source code!

I think before all that work, lets get the rankings thing fixed and get consensus that its a better ranking and then I can look at exporting it. I want to make a page that sorts by that composite ranking to see if it makes sense (seems like it does so far)
User avatar
McLovin
Posts: 1165
Joined: Sat Apr 03, 2021 12:54 pm
Location: Salt Lake City, Utah
Server Sponsor: Yes
Server Admin: Yes

Re: Balancer Test Results!!

Post by McLovin »

pooty wrote: Sun Jul 25, 2021 10:18 am
Where the hell am I on this list? McL)o(vin
Did you change your name on line? Or get stuck with PlayerNN? I think it updates the name if you change it, we can double check after you play a match or two.. I know you were in there...
Ooooh, this is a really nice page! How did you build this? I think it could make a nice addition to CEONSS' stats page as well.
...
I think it thinks I'm Player 20 [24], WTF is wrong with this thing. I never changed my name. I'm probably the only one on the entire server that only aliased with 2 aliases, McLovin and McL)o(vin. Everyone else aliases almost weekly and has 10 or 20 aliases.
User avatar
pooty
Posts: 4351
Joined: Sat Apr 03, 2021 10:22 am
Location: Michigan
Server Sponsor: Yes
Server Admin: Yes

Re: Balancer Test Results!!

Post by pooty »

See if it corrects itself. If not, I can manually update the Database but its a pain.
User avatar
McLovin
Posts: 1165
Joined: Sat Apr 03, 2021 12:54 pm
Location: Salt Lake City, Utah
Server Sponsor: Yes
Server Admin: Yes

Re: Balancer Test Results!!

Post by McLovin »

pooty wrote: Sun Jul 25, 2021 10:19 pm See if it corrects itself. If not, I can manually update the Database but its a pain.
And, now I'm back. LOL
User avatar
leon
Posts: 171
Joined: Tue Jun 08, 2021 5:35 pm
Location: Clearwater, Florida

Re: Balancer Test Results!!

Post by leon »

Sometimes during the match, some good players join the same team or they just leave the match and it ruins the balance of the game. I think in this situation, the admins should force switch some players to rebalance the teams.
User avatar
pooty
Posts: 4351
Joined: Sat Apr 03, 2021 10:22 am
Location: Michigan
Server Sponsor: Yes
Server Admin: Yes

Re: Balancer Test Results!!

Post by pooty »

Yeah its not easy though. The server won't let a switch (even from admins) that makes the teams uneven it automatically forces it back to even numbers (or 1 different max).

I'd still like to find a way to say "Force Reshuffle" from an admin if the teams get too out of whack.
User avatar
DrScrotey
Posts: 70
Joined: Sun Aug 29, 2021 9:59 pm
Location: Texas

Re: Balancer Test Results!!

Post by DrScrotey »

pooty wrote: Mon Jul 05, 2021 10:35 am So it was suggested to re-run the numbers...the stats DB keeps matches with 3 or more humans. So to avoid any data dilution by Helper-Bot, I've re run the numbers with numplayer > 9 so 5v5 or larger; I did add a couple of days included with balancer on..so the raw numbers aren't relevant just the percentages.

5v5

Balancer On: Four(count,%), Three, Two, One point matches, Total
24 0.2474 35 0.3608 21 0.2165 17 0.1753 97
Balancer Off:Four, Three, Two, One point matches, Total
39 0.2746 58 0.4085 33 0.2324 11 0.0775 142

6v6 (or larger):
Balancer ON 21 0.2386 32 0.3636 19 0.2159 16 0.1818 88
Balancer OFF 35 0.2734 55 0.4297 27 0.2109 11 0.0859 128

8v8:(or larger)
Balancer ON 15 0.2174 27 0.3913 13 0.1884 14 0.2029 69
Balancer OFF 28 0.2642 47 0.4434 21 0.1981 10 0.0943 106

10v10 (or larger)
Balancer ON 7 0.1458 20 0.4167 10 0.2083 11 0.2292 48
Balancer OFF 18 0.2338 34 0.4416 16 0.2078 9 0.1169 77

So, that's a major difference, as you can see as the number of players increase, the number of blowouts (4-0) are much higher with balancer off, and 1 pt matches are higher with balancer ON. More players likely decrease the "Enyo Effect" of certain good players that can really skew the performance on small team -- point here is a really good player is more effective on a smaller team, and likewise in a 4v4 match, a really bad noob, hurts quite a bit the other way.
Here's to resurrecting this thread.
I did a quick and dirty analysis of the 5v5 and 10v10 data. I specifically wanted to look at the uncertainty of the percentages due to sample size.
UTBalancerAnalysis.jpg
UTBalancerAnalysis.jpg (142.03 KiB) Viewed 797 times
The probabilities if the balancer is better corroborates with pooty's gut that it's slightly better, albeit the difference is swallowed by the margins of error. More data would clarify this, but collection is a slog I imagine.
In reading other parts of this thread though where enyo made a histogram of the effectiveness of the red vs. blue team I thought about the "spread" of that histogram, i.e. whether the players constituting the team are all in the middle vs. there's a low performance peak and high performance peak. One way to characterize this is the standard deviation, so instead of optimizing by balancing *only* using the first moment (the sum of the players rank scores) we could also optimize by balancing the second moment (the standard deviation of the player's running rank scores). The optimization algorithm would be more complicated though. But it might mitigate what enyo has mentioned.
Post Reply