MULTIbalancer [1.1.6.0] 30-MAR-2015 + BFHL

[ImportBot]

Originally Posted by ColColonCleaner*:

 

Can you add a new metric to the scramble and balance? NetKills. Their kills - deaths. Would like to experiment with that metric on metro14. Another metric combining kills and score would be interesting too.

[ImportBot]

Originally Posted by veio*:

 

Can someone tell me what are the best settings for a 24 slot server Chain Link? Maybe with the unstack mode? please

[ImportBot]

Originally Posted by PapaCharlie9*:

 

Can you add a new metric to the scramble and balance? NetKills. Their kills - deaths. Would like to experiment with that metric on metro14. Another metric combining kills and score would be interesting too.

Uh, isn't KDR equivalent to net kills? I know the guys I play with treat it as such -- I don't know how many times one of my pro friends looked at the final scoreboard and chortled, "Well at least my KDR was positive, unlike Papa's ...". Apparently, values less than 1.00 but greater than zero are considered negative ...

 

Score is a function of kills on most modes and Score/100 should come very close to Kills in TDM.

 

TL;DR, doesn't seem worth the effort to me.

[ImportBot]

Originally Posted by ColColonCleaner*:

 

Uh, isn't KDR equivalent to net kills? I know the guys I play with treat it as such -- I don't know how many times one of my pro friends looked at the final scoreboard and chortled, "Well at least my KDR was positive, unlike Papa's ...". Apparently, values less than 1.00 but greater than zero are considered negative ...

 

Score is a function of kills on most modes and Score/100 should come very close to Kills in TDM.

 

TL;DR, doesn't seem worth the effort to me.

NetKills is a combination of both kills and KDR. It says that someone going 10:0 is the same as someone going 40:30. Might be interesting to test for TDM.

[ImportBot]

Originally Posted by Frogtooth*:

 

A TDM player who goes 4:1 through 20 minutes of play is not at all as valuable as the guy who goes 40:10 despite sharing the same KDR. Likewise the player who goes 40:60 is not the same player as the guy who goes 40:10 despite sharing the same KPM. And the player who goes 40:4 by getting revived 30 times is not as valuable as the guy who goes 40:4 with no revives. So right now there is no metric that reflects the true value of the player. Feels to me like net kills comes closer to melding those.

 

Same problem exists with other modes. No single stat really reflects the value of each player. The conquest sniper who is going for 1000 meter head shots and a SPM of 1000 is not doing the team as much good as the guys taking and defending flags but only getting a 600 SPM. But no single stat and best I can tell no combination of stats reflects that.

[ImportBot]

Originally Posted by PapaCharlie9*:

 

I take your points. I don't believe they apply for KDR in the case of scrambler, since a full round of stats will have been accumulated for most players -- even if a player joined just 5 minutes before the end of the round, a 4:1 KDR is respectable -- but I acknowledge that for balancing there are some edge cases that may result in some false positives.

 

In any case, I'm still not seeing much benefit to adding net kills. Unless you believe cases like 40:60 stats happen often in the context of KPM, the occasional false positive isn't going to make that much difference for balancing. Even if we had a perfect function for sorting the strong from the weak, the balancer only uses strong/weak judgments when there is no urgency in balancing. If balancing starts taking too long, or the gap is too large, or the speed is Fast, stats are no longer considered, and you may have a bunch of 40:60's, not to mention 1:60's, balanced to the losing team. And we haven't even talked about new players joining, for which we have no stats (if we are talking about in-round, rather than Battlelog).

[ImportBot]

Originally Posted by PapaCharlie9*:

 

Has anyone used any of the new untested features successfully? I'd like to remove the "UNTESTED" label if that stuff is working.

[ImportBot]

Originally Posted by Frogtooth*:

 

I take your points. I don't believe they apply for KDR in the case of scrambler, since a full round of stats will have been accumulated for most players -- even if a player joined just 5 minutes before the end of the round, a 4:1 KDR is respectable -- but I acknowledge that for balancing there are some edge cases that may result in some false positives.

 

In any case, I'm still not seeing much benefit to adding net kills. Unless you believe cases like 40:60 stats happen often in the context of KPM, the occasional false positive isn't going to make that much difference for balancing. Even if we had a perfect function for sorting the strong from the weak, the balancer only uses strong/weak judgments when there is no urgency in balancing. If balancing starts taking too long, or the gap is too large, or the speed is Fast, stats are no longer considered, and you may have a bunch of 40:60's, not to mention 1:60's, balanced to the losing team. And we haven't even talked about new players joining, for which we have no stats (if we are talking about in-round, rather than Battlelog).

All the same stats apply to the unstacker as well, no?

 

But I'm more struggling to get even games in conquest. You can't get stability in a feedback system if the variables you're using to control it don't actually control it. And there just isn't any metric I can think of to apply. When I find unbalanced games, usually the unstacker is stating the losing team has the best scores and if I join the game I'll see a team full of high-scorers who are ignoring the flags. They have lots of kills and lots of vehicle kills, but they aren't taking and holding flags.

 

You want the pilots balanced between the teams, you want the tank jockeys balanced, the flag runners balanced, and the snipers balanced... but without data to tell you who is doing what I don't see how to make it happen. SPM seems the only available metric that can be applied to conquest. Which usually works ok. But among the most common complaints we hear is "all the pilots are on one side". I haven't figured out how to automate unstacking the pilots.

[ImportBot]

Originally Posted by ColColonCleaner*:

 

Has anyone used any of the new untested features successfully? I'd like to remove the "UNTESTED" label if that stuff is working.

Where are they listed?

[ImportBot]

Originally Posted by ColColonCleaner*:

 

When people are moved for even dispersion, what determines which players on the list are moved?

[ImportBot]

Originally Posted by LumPenPacK*:

 

I suppose a KPM value could represent almost the same skill level like this net kills value does.

But the validity of KPM representation might depend on how the KPM is calculated.

 

As far as I know, Battlelog calculates the KPM from the actual play time which represents the time a player is not death.

That would mean a player who have 30/10 stats has a lower KPM than a player with 30/20 because the 2nd player killed more players when he/she was alive. I'm not 100% sure about this but I know that the play time will only be increased when you have spawned. But probably your play time will be increased as long as you are in a revivable state.

[ImportBot]

Originally Posted by tarreltje*:

 

That's correct, features like balancing and scrambling have minimum player population requirements. While 3 v 1 may cause players to leave, so will admin killing a player and forcing them to move to the other team sometimes. What if the other 2 players on the team of 3 are his friends? There's no guarantee either way.

29-7-2014:

 

I have monitored my server last 2 days and i have seen it go empty for several times due the mis balance. Every time it ends up into a 3 vs 1. I have also monitored with true balancer and i havent got any problem about players complaining they were divided since they were friends. I have the fast balance on, and also i reassign new players.

 

Isnt it posible you can add a feature like overwrite all rules and keep difference of 1 player till x players are in the game?

 

The problem is, that your balancer is given me so many tight games, that i need to use it, but we have a hard time filling it up.

 

30-7-2014:

 

Today the same thing, i have managed to get my server going 2 times now. Last time if i didnt moved my self again had a 3 vs 1 situation wich would have emptyed my server. Isnt it posible that i can alter some coding? so that i can balance with 1 player difference?

[ImportBot]

Originally Posted by ColColonCleaner*:

 

Seems like it's not detecting players as spectators. Does this affect anything?

 

http://i.gyazo.com/973690546defdc7f2c1c92488d90a9e2.png

[ImportBot]

Originally Posted by PapaCharlie9*:

 

Where are they listed?

In the traditional location: the plugin update post #2.

[ImportBot]

Originally Posted by PapaCharlie9*:

 

When people are moved for even dispersion, what determines which players on the list are moved?

If you mean the Disperse Evenly List, that's fully documented in post #1. All of the players that are not in their assigned teams will be moved asap.

[ImportBot]

Originally Posted by PapaCharlie9*:

 

Seems like it's not detecting players as spectators. Does this affect anything?

 

http://i.gyazo.com/973690546defdc7f2c1c92488d90a9e2.png

It's supposed to detect spectators, thus the status line about the number of spectators. If you are sure there were spectators in the game, something might have broken during a BF4 update. It's not clear if it is Procon or the plugin that is broken. All the plugin does is pick up the CPlayerInfo.Type during a listPlayers event. If the Type == 1, it's a spectator. Looks like I don't have any logging at the point where Type is assigned to PlayerModel.Role (UpdatePlayerModel), so there's no easy way to verify without changing the plugin code.

[ImportBot]

Originally Posted by Frogtooth*:

 

It's supposed to detect spectators, thus the status line about the number of spectators. If you are sure there were spectators in the game, something might have broken during a BF4 update. It's not clear if it is Procon or the plugin that is broken. All the plugin does is pick up the CPlayerInfo.Type during a listPlayers event. If the Type == 1, it's a spectator. Looks like I don't have any logging at the point where Type is assigned to PlayerModel.Role (UpdatePlayerModel), so there's no easy way to verify without changing the plugin code.

I've noticed this as well. It puts out that message constantly when someone is observing.

[ImportBot]

Originally Posted by ColColonCleaner*:

 

It's supposed to detect spectators, thus the status line about the number of spectators. If you are sure there were spectators in the game, something might have broken during a BF4 update. It's not clear if it is Procon or the plugin that is broken. All the plugin does is pick up the CPlayerInfo.Type during a listPlayers event. If the Type == 1, it's a spectator. Looks like I don't have any logging at the point where Type is assigned to PlayerModel.Role (UpdatePlayerModel), so there's no easy way to verify without changing the plugin code.

They were detected in AdKats as spectators, confirmed with the /find command, but were not in the balancer. Just informing you of what i saw. If this doesn't affect the way balance is carried out I'm not too worried about it, but if other processes in the balancer are being hindered by undetected spectators in the server that is a little worrying.

[ImportBot]

Originally Posted by ColColonCleaner*:

 

If you mean the Disperse Evenly List, that's fully documented in post #1. All of the players that are not in their assigned teams will be moved asap.

Not the particular question i needed answered. Specifically wondering how it determines which players on the list are moved first, or before others, is it just by who dies first? I noticed some were not since "moved more than others", just looking for the specifics.

[ImportBot]

Originally Posted by PapaCharlie9*:

 

Not the particular question i needed answered. Specifically wondering how it determines which players on the list are moved first, or before others, is it just by who dies first? I noticed some were not since "moved more than others", just looking for the specifics.

By who dies first. Sequencing for moves is always who dies first, unless admin kill for fast balance is enabled.

[ImportBot]

Originally Posted by PapaCharlie9*:

 

They were detected in AdKats as spectators, confirmed with the /find command, but were not in the balancer. Just informing you of what i saw. If this doesn't affect the way balance is carried out I'm not too worried about it, but if other processes in the balancer are being hindered by undetected spectators in the server that is a little worrying.

It does impact balancing. If Adkats works, its not Procon, so I'll have to debug the plugin. Trouble is, I don't have access to a test server that reliably has spectators.

[ImportBot]

Originally Posted by tarreltje*:

 

@PC9 is it posible to adjust MB, so it wont need 5 players to start balancing, but 4 players?

[ImportBot]

Originally Posted by kcuestag*:

 

It does impact balancing. If Adkats works, its not Procon, so I'll have to debug the plugin. Trouble is, I don't have access to a test server that reliably has spectators.

I can help with that as long as it doesn't require our server to have the balancer disabled or empty it out.

[ImportBot]

Originally Posted by Frogtooth*:

 

[16:41:05 85] [MULTIbalancer]:4 Update waiting for DeputyD to be assigned a team. Role= 0

[16:41:05 85] [MULTIbalancer]:4 Update waiting for Frogtooth to be assigned a team. Role= 1

 

I added Role to the debug message. I was spectating, the other guy was playing.

[ImportBot]

Originally Posted by PapaCharlie9*:

 

[16:41:05 85] [MULTIbalancer]:4 Update waiting for DeputyD to be assigned a team. Role= 0

[16:41:05 85] [MULTIbalancer]:4 Update waiting for Frogtooth to be assigned a team. Role= 1

 

I added Role to the debug message. I was spectating, the other guy was playing.

That's a good start. The role is getting to the function, but why isn't it getting to the model? Did the status log messages say there was 1 spectator? Post log messages with Debug Level set to 4.

 

EDIT: if you want to try the test again, this change would be better.

 

Find this line of code in UpdatePlayerMode:

 

Code:

m.Role = role;

And add this after it:

 

Code:

if (role != 0)
    DebugWrite("UpdatePlayerModel: player ^b" + name + "^n, role = " + m.Role, 4);

Set Debug Level to 4 and post logs.

[ImportBot]

Originally Posted by Frogtooth*:

 

Easy enough...

 

...

m.DeathsRound = deaths;

m.Rank = rank;

m.Role = role;

 

/* ***TEST*** */

if (role != 0)

DebugWrite("UpdatePlayerModel: player ^b" + name + "^n, role = " + m.Role, 4);

 

m.LastSeenTimestamp = DateTime.Now;

...

 

Leads to:

 

[18:53:06 94] [MULTIbalancer]:0 +------------------------------------------------+

[18:53:06 94] [MULTIbalancer]:0 Status: Map = Golmud Railway, mode = Conquest Large, time in round = 00:06:09, tickets = 179/767

[18:53:06 94] [MULTIbalancer]:0 Status: Ticket difference = 588, ticket ratio percentage is 428%, average RoundSPM stats ratio = 135% (1722.3/1273.5)

[18:53:06 94] [MULTIbalancer]:0 Status: Autobalance is not active, phase = Mid, population = High, speed = Adaptive, unstack when ratio >= 120%

[18:53:06 94] [MULTIbalancer]:0 Status: Team counts [57] = 28(US) vs 29(US), with 8 unassigned, 2 commanders, 0 spectators

[18:53:06 94] [MULTIbalancer]:0 Status: Team difference = 1

[18:53:09 83] [MULTIbalancer]:4 Thread(fetcher): no more requests in Battlelog request queue

[18:53:09 83] [MULTIbalancer]:4 New player: nycmedic8405, assigned to US team by game server

[18:53:12 27] [MULTIbalancer]:4 (SIMULATING) REASSIGNING new player SexyComet from US team to US team

[18:53:12 27] [MULTIbalancer]:4 Thread(fetcher): no more requests in Battlelog request queue

[18:53:12 27] [MULTIbalancer]:4 New player: SexyComet, reassigned to US team by MULTIbalancer

[18:53:12 41] [MULTIbalancer]:4 ADMIN moved player (REVERSED) SexyComet, MULTIbalancer will respect this move

[18:53:12 54] [MULTIbalancer]:4 UpdatePlayerModel: player xxxDocHolidayxxx, role = 2

[18:53:12 54] [MULTIbalancer]:4 UpdatePlayerModel: player SFCPelt, role = 2

[18:53:12 54] [MULTIbalancer]:4 Update waiting for Remington66 to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for Xx360xN0xSc0pExX to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for stonejacker to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for Cpl-LeKinG to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for AremoKel to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for Frogtooth to be assigned a team. Role= 1

[18:53:12 54] [MULTIbalancer]:0 +------------------------------------------------+

[18:53:12 54] [MULTIbalancer]:0 Status: Map = Golmud Railway, mode = Conquest Large, time in round = 00:06:15, tickets = 179/767

[18:53:12 54] [MULTIbalancer]:0 Status: Ticket difference = 588, ticket ratio percentage is 428%, average RoundSPM stats ratio = 136% (1652.1/1218.1)

[18:53:12 54] [MULTIbalancer]:0 Status: Autobalance is not active, phase = Mid, population = High, speed = Adaptive, unstack when ratio >= 120%

[18:53:12 54] [MULTIbalancer]:0 Status: Team counts [59] = 29(US) vs 30(US), with 6 unassigned, 2 commanders, 0 spectators

[18:53:12 54] [MULTIbalancer]:0 Status: Team difference = 1

[18:53:21 01] [MULTIbalancer]:4 Player left: SexyComet

[18:53:23 00] [MULTIbalancer]:4 UpdatePlayerModel: player xxxDocHolidayxxx, role = 2

[18:53:23 00] [MULTIbalancer]:4 UpdatePlayerModel: player SFCPelt, role = 2

[18:53:23 00] [MULTIbalancer]:4 Update waiting for Remington66 to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for Xx360xN0xSc0pExX to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for stonejacker to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for Cpl-LeKinG to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for AremoKel to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for Frogtooth to be assigned a team. Role= 1

[18:53:23 00] [MULTIbalancer]:0 +------------------------------------------------+

 

SFCPelt and xxxDocHolidayxxx are in the commander slots, Frogtooth is spectating.

[ImportBot]

Originally Posted by PapaCharlie9*:

 

@PC9 is it posible to adjust MB, so it wont need 5 players to start balancing, but 4 players?

You mean for Fast Balance? How about you test it out first and then let us know how that works. It seems that admin killing a player in a 3 v 1 situation is just as likely to lose players as just leaving it 3 v 1.

 

Find this line of code in FastBalance and make the change in red:

 

Code:

int floorPlayers = 4; // minimum number for which a team difference of 3 minimum is possible

You also need to make this change a few lines below the first change:

 

Code:

if (balanceSpeed != Speed.Fast || diff < 2) {

Finally, make sure you set your balance speed to Fast in Section 3 for all combinations.

[ImportBot]

Originally Posted by tarreltje*:

 

You mean for Fast Balance? How about you test it out first and then let us know how that works. It seems that admin killing a player in a 3 v 1 situation is just as likely to lose players as just leaving it 3 v 1.

 

Find this line of code in FastBalance and make the change in red:

 

Code:

int floorPlayers = 4; // minimum number for which a team difference of 3 minimum is possible

You also need to make this change a few lines below the first change:

 

Code:

if (balanceSpeed != Speed.Fast || diff < 2) {

Finally, make sure you set your balance speed to Fast in Section 3 for all combinations.

Thnx for your reply PC9

 

I am going to test this out!!

[ImportBot]

Originally Posted by ColColonCleaner*:

 

ETA on the dispersion list fixes? That is causing some issues on our servers right now.

[ImportBot]

Originally Posted by PapaCharlie9*:

 

Easy enough...

 

...

m.DeathsRound = deaths;

m.Rank = rank;

m.Role = role;

 

/* ***TEST*** */

if (role != 0)

DebugWrite("UpdatePlayerModel: player ^b" + name + "^n, role = " + m.Role, 4);

 

m.LastSeenTimestamp = DateTime.Now;

...

 

Leads to:

 

[18:53:06 94] [MULTIbalancer]:0 +------------------------------------------------+

[18:53:06 94] [MULTIbalancer]:0 Status: Map = Golmud Railway, mode = Conquest Large, time in round = 00:06:09, tickets = 179/767

[18:53:06 94] [MULTIbalancer]:0 Status: Ticket difference = 588, ticket ratio percentage is 428%, average RoundSPM stats ratio = 135% (1722.3/1273.5)

[18:53:06 94] [MULTIbalancer]:0 Status: Autobalance is not active, phase = Mid, population = High, speed = Adaptive, unstack when ratio >= 120%

[18:53:06 94] [MULTIbalancer]:0 Status: Team counts [57] = 28(US) vs 29(US), with 8 unassigned, 2 commanders, 0 spectators

[18:53:06 94] [MULTIbalancer]:0 Status: Team difference = 1

[18:53:09 83] [MULTIbalancer]:4 Thread(fetcher): no more requests in Battlelog request queue

[18:53:09 83] [MULTIbalancer]:4 New player: nycmedic8405, assigned to US team by game server

[18:53:12 27] [MULTIbalancer]:4 (SIMULATING) REASSIGNING new player SexyComet from US team to US team

[18:53:12 27] [MULTIbalancer]:4 Thread(fetcher): no more requests in Battlelog request queue

[18:53:12 27] [MULTIbalancer]:4 New player: SexyComet, reassigned to US team by MULTIbalancer

[18:53:12 41] [MULTIbalancer]:4 ADMIN moved player (REVERSED) SexyComet, MULTIbalancer will respect this move

[18:53:12 54] [MULTIbalancer]:4 UpdatePlayerModel: player xxxDocHolidayxxx, role = 2

[18:53:12 54] [MULTIbalancer]:4 UpdatePlayerModel: player SFCPelt, role = 2

[18:53:12 54] [MULTIbalancer]:4 Update waiting for Remington66 to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for Xx360xN0xSc0pExX to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for stonejacker to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for Cpl-LeKinG to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for AremoKel to be assigned a team. Role= 0

[18:53:12 54] [MULTIbalancer]:4 Update waiting for Frogtooth to be assigned a team. Role= 1

[18:53:12 54] [MULTIbalancer]:0 +------------------------------------------------+

[18:53:12 54] [MULTIbalancer]:0 Status: Map = Golmud Railway, mode = Conquest Large, time in round = 00:06:15, tickets = 179/767

[18:53:12 54] [MULTIbalancer]:0 Status: Ticket difference = 588, ticket ratio percentage is 428%, average RoundSPM stats ratio = 136% (1652.1/1218.1)

[18:53:12 54] [MULTIbalancer]:0 Status: Autobalance is not active, phase = Mid, population = High, speed = Adaptive, unstack when ratio >= 120%

[18:53:12 54] [MULTIbalancer]:0 Status: Team counts [59] = 29(US) vs 30(US), with 6 unassigned, 2 commanders, 0 spectators

[18:53:12 54] [MULTIbalancer]:0 Status: Team difference = 1

[18:53:21 01] [MULTIbalancer]:4 Player left: SexyComet

[18:53:23 00] [MULTIbalancer]:4 UpdatePlayerModel: player xxxDocHolidayxxx, role = 2

[18:53:23 00] [MULTIbalancer]:4 UpdatePlayerModel: player SFCPelt, role = 2

[18:53:23 00] [MULTIbalancer]:4 Update waiting for Remington66 to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for Xx360xN0xSc0pExX to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for stonejacker to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for Cpl-LeKinG to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for AremoKel to be assigned a team. Role= 0

[18:53:23 00] [MULTIbalancer]:4 Update waiting for Frogtooth to be assigned a team. Role= 1

[18:53:23 00] [MULTIbalancer]:0 +------------------------------------------------+

 

SFCPelt and xxxDocHolidayxxx are in the commander slots, Frogtooth is spectating.

Oh! That's the problem, silly me! The model is set only AFTER a player has joined the server, which is to say, the first time a OnPlayerTeamChange event happens. For spectators, that event never happens, so the model is never set. This is a bug, I will fix it.

 

Many thanks for helping debug this for me!