Jump to content


WG is testing Skill-Based MM in Mad/Burning Games

MM

  • Please log in to reply
17 replies to this topic

Jylpah #1 Posted 23 October 2020 - 10:33 PM

    Lance-corporal

  • Players
  • 83 battles
  • 38
  • Member since:
    05-04-2015

Hello NA forumers,

 

Main thread on EU forums: http://forum.wotblitz.eu/index.php?/topic/54798-is-wg-using-skill-based-mm-in-mad-games/

 

I have received a replay data set from Burning Games and O'boy the data looks "Interesting". I can't remember have I ever seen such a large WR / average damage difference between the teams. I gonna run t-test later, but 2% WR difference between green and red teams over 134 battles looks simply huge to me. What do you think? I have confirmed from replay time stamps that the battles are played on consecutive battles played over few days. The player requested to stay anonymous. All the battles have been played solo. 

 

For me this seems like WG is testing skill-based MM with the Burning and Mad Games.
 

Spoiler

 

This is just one data point although there are many battles in the data set. Therefore, I kindly ask if anyone is/has been playing Burning games  to post the replays to WoTinspector.com and send me a note e.g. via Discord Jylpah#4662 or email Jylpah@gmail.com. You don't need to send me the replays, just your account nick and date range of the battles. I can filter the battles played in Burning Games mode

 

UPDATE: 2020-10-24 (bit late :-)

 

The replays need to be played solo and uploaded to replays.WoTinspector.com (e.g. via Discord WoTbot) . No need to email to me or send me PM via Discord. I figured out how to filter replays by battle mode.

 

We might be onto something here! I am expecting WG to continue their MM experiment in Mad Games starting now.


Edited by Jylpah, 29 October 2020 - 02:00 PM.


_Crusader6_ #2 Posted 23 October 2020 - 11:26 PM

    Vote Slime Mold

  • Players
  • 70639 battles
  • 15,630
  • [BDGER]
  • Member since:
    12-08-2014

I did see any skill at all in Burning Games, just seemed to be an exceptional amount of windowlickers out watching for windows too lick, and oxygen to thieve...

 


 

 
Tank Hoarder: 424 tanks in Garage:  435/435 Played Tanks Aced
 
This space for rent until next terrible mechanic ;) 
 
    Wallet Warrior: Loyal Original M60 owner
 
 

 


acrisis #3 Posted 24 October 2020 - 12:45 AM

    Captain

  • Players
  • 21303 battles
  • 12,786
  • [III-C]
  • Member since:
    12-27-2014

The only thing I can say is that, during the initial days, I loved the new mode, and appeared to see a great number of balanced teams and battles, that went down gradually, could go either way ... and in the last few days I have absolutely struggled to even get a four wins mission completed in two hours;  17 games to get 4 wins last night, while observing a very high number of lopsided <50% and >50-55% teams, 60% toons on red 40% toons on green, complete and total steamrolls. Super lopsided teams, some unlucky RNG, ... 

 

Does it prove anything? 
Or just more of the MM RNG roller coaster or pendulum? 


 

Forum essentials:  > Desktop mode 
Check recent threads. Check WG staff threads. Use search.
    > WG Bugs and current update     > Feedback     > Vehicle Bay     > Guides     > Videos 

 

     BE  (+)   

 [ NEW > ]  Lag, Packet Loss, Ghost Shells, Networking and more demystified usingPingPlotter (free)  
Blitz Community Coalition | Looney Tooners | Triarii
Blitz University | Basic Training  |  Mentor  |  The Pingplotter Guy

 


THERebrth #4 Posted 24 October 2020 - 01:06 AM

    Sergeant

  • Players
  • 9773 battles
  • 392
  • [CLEAR]
  • Member since:
    12-20-2013
if wargaming is testing out skill based mm, and im practically losing 70% of my games when I know Im a decent player? somethings not right. I mean you can tell that this guy lost over 60% of his games [edited]. Don't give me skill based mm if its gona make me lose more by giving my crap teammates because my WR is high to balance the crappy players. 1v7 like seriously?

Raging Dumbo


MarBearCat #5 Posted 24 October 2020 - 02:04 AM

    Pennsylvania's Favorite Cat!

  • Players
  • 22553 battles
  • 12,355
  • [SDWA]
  • Member since:
    11-13-2015

View Postacrisis, on 23 October 2020 - 07:45 PM, said:

The only thing I can say is that, during the initial days, I loved the new mode, and appeared to see a great number of balanced teams and battles, that went down gradually, could go either way ... and in the last few days I have absolutely struggled to even get a four wins mission completed in two hours;  17 games to get 4 wins last night, while observing a very high number of lopsided <50% and >50-55% teams, 60% toons on red 40% toons on green, complete and total steamrolls. Super lopsided teams, some unlucky RNG, ... 

 

Does it prove anything? 
Or just more of the MM RNG roller coaster or pendulum? 

 

I got 1 win out of 6 and limped away tonight.

 


 

 

Heavy Is The Cat That's Carried!


Rou_Garou #6 Posted 24 October 2020 - 02:33 AM

    Carriers needed - apply inside

  • Players
  • 34355 battles
  • 1,023
  • [OP-4]
  • Member since:
    05-04-2013

This data collected would be useless.

There is nothing normal about these battles for practical use.

If true, my play is magnified as truly turrible


 

 

 


bwcamp #7 Posted 24 October 2020 - 10:27 AM

    Senior Sergeant

  • Players
  • 8234 battles
  • 838
  • [WOY1]
  • Member since:
    12-06-2016

View PostTHERebrth, on 23 October 2020 - 08:06 PM, said:

if wargaming is testing out skill based mm, and im practically losing 70% of my games when I know Im a decent player? somethings not right. I mean you can tell that this guy lost over 60% of his games [edited]. Don't give me skill based mm if its gona make me lose more by giving my crap teammates because my WR is high to balance the crappy players. 1v7 like seriously?

Not if they have the same number of crappy players

 


The difference between camping and sniping is that the sniper looks for opportunities and the camper looks at the scenery.
- EL_Din_1946

Jylpah #8 Posted 26 October 2020 - 08:00 AM

    Lance-corporal

  • Players
  • 83 battles
  • 38
  • Member since:
    05-04-2015

View PostRou_Garou, on 24 October 2020 - 04:33 AM, said:

This data collected would be useless.

There is nothing normal about these battles for practical use.

If true, my play is magnified as truly turrible


I am afraid you are misunderstanding what data is being recorded and what is being analyzed. The point is not to analyze the results of the battles, but the randomness of the player selection across the teams. 
 

The data is very convincing that WG is applying skill-based MM in Burning and Mad Games for 60%+ WR players. For me it seems they could be using career WR as the measure for skill, but to confirm that, more data is needed.
 

If anyone wants to contribute, please post your non-cherry-picked solo replays to WoTinspector.com. Either via Discord WoTBot or via email replays@wotinspector.com. That’s all. I will get the data from there for the analysis.

 

Burning Games data

 

 

Mad Games

 

Random battles (No rigging observed)

 

 


Edited by Jylpah, 26 October 2020 - 02:23 PM.


_Crusader6_ #9 Posted 26 October 2020 - 12:42 PM

    Vote Slime Mold

  • Players
  • 70639 battles
  • 15,630
  • [BDGER]
  • Member since:
    12-08-2014

Jylpah,

  I’ll be honest I hate these ‘distraction’ game modes, but how many solo games would you like players to upload of them. 
Tier Pref? And do you want a random assortment of tanks played? 
 

 

 


 

 
Tank Hoarder: 424 tanks in Garage:  435/435 Played Tanks Aced
 
This space for rent until next terrible mechanic ;) 
 
    Wallet Warrior: Loyal Original M60 owner
 
 

 


Jylpah #10 Posted 26 October 2020 - 01:23 PM

    Lance-corporal

  • Players
  • 83 battles
  • 38
  • Member since:
    05-04-2015

View Post_Crusader6_, on 26 October 2020 - 02:42 PM, said:

Jylpah,

  I’ll be honest I hate these ‘distraction’ game modes, but how many solo games would you like players to upload of them. 
Tier Pref? And do you want a random assortment of tanks played?

 

Hi,

 

Upper tiers, preferably VIII and up. Whatever tanks. Even five games per player is a good contribution, just please do not cherry-pick the battles, but upload all. And I do hate these modes too. I am trying to share the pain here LOL.

 

Cheers,

J


Edited by Jylpah, 26 October 2020 - 02:24 PM.


Jylpah #11 Posted 29 October 2020 - 12:40 PM

    Lance-corporal

  • Players
  • 83 battles
  • 38
  • Member since:
    05-04-2015

Hello,

 

Posting an update:

 

WG continues to use the same Skill-Based MM algorithm in Mad Games as in Burning Games. The SBMM algo results better players getting worse teammates than the enemies. Regular Battles seem not to utilize Skill-Based MM yet.

 

Below is an analysis of the team compositions from players who have submitted at least 5 Regular Battles / Mad Games replays to WoTinspector. The key statistics here is to compare the enemies/allies' stats (WR, average damage, battles). Those should be the same if they were randomly selected. The player itself is not counted into the allies' stats since he has not been "randomly selected".

 

Mad Games

 

Good players are given worse teams. Players with 65%+ WR have on average team mates with 2% lower WR than the enemies. That is a lot of trash to carry.

 

Analysis

 

./analyze_wotb_replays.py --mode team --extra total player_wins player_damage_dealt  --only  --filters  '{  "data.summary.battle_start_timestamp": { "$gte": 1603526400} , "data.summary.room_type": 8 }' db:
Fetching player stats |███████████████████████████     | 2800/3323 84% ETA 0 h 0 mins

TOTAL_______________   Battles    Player WR   Allies WR   Enemies WR   Player Btls   Allies Btls   Enemies Btls   Player Avg Dmg   Allies Avg Dmg   Enemies Avg Dmg
               Total :   238    :    64.85% :    50.07% :     51.82% :       47024 :       16270 :        17370 :        1276 :        1006 :         1104


Player WR___________   Battles    Player WR   Allies WR   Enemies WR   Player Btls   Allies Btls   Enemies Btls   Player Avg Dmg   Allies Avg Dmg   Enemies Avg Dmg
            45 - 50% :    9     :    47.23% :    48.83% :     49.46% :        3337 :        9359 :        10347 :         666 :         835 :          879
            50 - 55% :    8     :    53.48% :    50.89% :     51.48% :       28127 :       17465 :        20573 :        1385 :        1126 :         1147
            55 - 65%    45    :    59.02% :    50.46% :     51.61%      18130 :       19004 :        19012 :        1318 :        1058 :         1123
            65 -   % :   176    :    67.76% :    49.99% :     52.01%       57504 :       15870 :        17164 :        1291 :         995 :         1108


Player Avg Dmg______   Battles    Player WR   Allies WR   Enemies WR   Player Btls   Allies Btls   Enemies Btls   Player Avg Dmg   Allies Avg Dmg   Enemies Avg Dmg
          500 - 1000 :    9     :    47.23% :    48.83% :     49.46% :        3337 :        9359 :        10347 :         666 :         835 :          879
         1000 - 1250 :   141    :    67.14% :    49.90% :     51.89% :       61696 :       15509 :        16845 :        1182 :         976 :         1100
         1250 - 1500 :    46    :    58.50% :    50.45% :     51.63% :       18402 :       18857 :        19208 :        1344 :        1076 :         1124
         1500 - 1750 :    42    :    67.92% :    50.48% :     52.28% :       38476 :       17474 :        18625 :        1647 :        1064 :         1142

 

Regular Battles

 

It must be noted that this has not been seen in Random Regular Battles yet.

 

Analysis

./analyze_wotb_replays.py --mode team --extra total player_wins player_damage_dealt  --only  --filters  '{ "data.summary.battle_start_timestamp": { "$gte": 1603526400} , "data.summary.room_type": 1 }' db:
Fetching player stats |█████████████████████████████   | 1750/1928 90% ETA 0 h 0 mins

TOTAL_______________   Battles    Player WR   Allies WR   Enemies WR   Player Btls   Allies Btls   Enemies Btls   Player Avg Dmg   Allies Avg Dmg   Enemies Avg Dmg
               Total :   138    :    59.62% :    49.54% :     49.68% :       19025 :       12588 :        12838 :        1350 :         848 :          870


Player WR___________   Battles    Player WR   Allies WR   Enemies WR   Player Btls   Allies Btls   Enemies Btls   Player Avg Dmg   Allies Avg Dmg   Enemies Avg Dmg
            45 - 50% :    8     :    46.72% :    49.73% :     49.39% :       17277 :       15197 :        16435 :         985 :         862 :          846
            50 - 55% :    18    :    51.70% :    50.02% :     49.45% :       22365 :       11799 :         8721 :        1312 :         764 :          693
            55 - 65% :    81    :    59.05% :    49.56% :     49.77% :       22510 :       13909 :        14957 :        1255 :         872 :          896
            65 -   % :    31    :    69.05% :    49.16% :     49.63% :        8431 :        8919 :         8760 :        1713 :         830 :          911


Player Avg Dmg______   Battles    Player WR   Allies WR   Enemies WR   Player Btls   Allies Btls   Enemies Btls   Player Avg Dmg   Allies Avg Dmg   Enemies Avg Dmg
          500 - 1000 :    24    :    53.20% :    49.74% :     49.42% :       28735 :       14287 :        13901 :         994 :         844 :          849
         1000 - 1250 :    42    :    60.96% :    49.38% :     49.54% :        6359 :       12219 :        12971 :        1182 :         800 :          804
         1250 - 1500 :    36    :    56.72% :    50.18% :     49.66% :       24810 :       10812 :        10424 :        1315 :         870 :          856
         1500 - 1750 :    13    :    60.28% :    49.93% :     50.33% :       44614 :       22959 :        20807 :        1564 :        1053 :         1060
         1750 - 2000 :    18    :    69.69% :    48.03% :     49.26% :        3210 :        6643 :         8186 :        1934 :         737 :          861
         2000 - 2500 :    5     :    62.18% :    49.85% :     51.99% :       27553 :       14745 :        20014 :        2062 :         968 :         1165

 

Skill-Based MM will make the game unplayable for any decent player by consistently putting worse players to your team. We must stop WG introducing Skill-Based MM into Regular Battles.

 

Here is what you can do to help:

 

1) Please reach out your favorite CC to put pressure on WG to be more open about the "MM tweaking" they mentioned in the previous stream.

 

2) Provide more data.  Please submit your non-cherry-picked replays to replays.WoTinspector.com. Either via email replays@wotinspector.com or via Discord WoTbot (your Discord server may already have it or then join the WoTinspector Discord server at https://discord.gg/TkPjTRc). Even just submitting 10 replays will make an impact. I will find the replays from there automatically. Just remember:

 

  • Solo replays (you can post platoon games too, but for this analysis I will filter out platoon games)
  • No cherry-picking, just sent consecutive battles - win or loss

 

The replay analysis is done with blitz-tools



acrisis #12 Posted 30 October 2020 - 03:00 AM

    Captain

  • Players
  • 21303 battles
  • 12,786
  • [III-C]
  • Member since:
    12-27-2014
Was able to sit down for a bit and email upload files one at time, from the past few days. Probably not burning games, like I named them, but mad games. 

 

Forum essentials:  > Desktop mode 
Check recent threads. Check WG staff threads. Use search.
    > WG Bugs and current update     > Feedback     > Vehicle Bay     > Guides     > Videos 

 

     BE  (+)   

 [ NEW > ]  Lag, Packet Loss, Ghost Shells, Networking and more demystified usingPingPlotter (free)  
Blitz Community Coalition | Looney Tooners | Triarii
Blitz University | Basic Training  |  Mentor  |  The Pingplotter Guy

 


GLKamper #13 Posted 30 October 2020 - 04:10 PM

    Sergeant

  • Players
  • 41109 battles
  • 424
  • Member since:
    10-31-2015

I go through peaks and valleys in these modes so I’m sure many are doing the same. I use them to grind tanks so I play very few games in a tank that’s equipped competitively. You can have a skilled player but in a stock tank with a 75% crew, skill isn’t always able to come through. 

 



minitel_NA #14 Posted 30 October 2020 - 08:40 PM

    First Sergeant

  • Players
  • 454 battles
  • 6,042
  • [EUREF]
  • Member since:
    12-19-2015

Interesting contribution :

burning games meant more damage more kills per battle. So more XP per battle.

since I was inches from purchasing new tanks, when burning mode came i did buy 3 new t7-8-9 tanks.

so there I was grinding crews and modules in upper tiers.

 

i won 10% of my games for long sessions and total 20-ish over several days,

It had gotten so bad I tried brushing my teeth with one hand and playing with the other (iPad) and I a

was still in the top half of my team per damage every single game.

 

it was so consistent and you offer a very solid explanation :

my team lower winrate than the already deficient average mad game plauer, and me in a basically lower tier stock tank, and a game mode that exacerbate facial DPM over all aspects.

 

simple maths.


Minitelrose visiting from EU, occasional player/forumer in the NA

 


acrisis #15 Posted 30 October 2020 - 09:38 PM

    Captain

  • Players
  • 21303 battles
  • 12,786
  • [III-C]
  • Member since:
    12-27-2014
I had the same win only win 10% thing ... as low tier, with some rusty play, perhaps a mistake here and there, but nobody carried me ... and idiot teams who just waste their tank hp and then you sit there 1 v 3 4 5 6 as low tier. 

 

Forum essentials:  > Desktop mode 
Check recent threads. Check WG staff threads. Use search.
    > WG Bugs and current update     > Feedback     > Vehicle Bay     > Guides     > Videos 

 

     BE  (+)   

 [ NEW > ]  Lag, Packet Loss, Ghost Shells, Networking and more demystified usingPingPlotter (free)  
Blitz Community Coalition | Looney Tooners | Triarii
Blitz University | Basic Training  |  Mentor  |  The Pingplotter Guy

 


9_V0LT #16 Posted 31 October 2020 - 12:13 AM

    Cunning Linguist

  • Players
  • 10641 battles
  • 3,174
  • [OGRE]
  • Member since:
    06-26-2017

Maybe Wargaming is and maybe they aren’t. 
Either way, it wouldn’t make any difference to the NA server, when the average queue for these games is only about 30 players. 
 

NA server is just too small for any kind of MM algorithm to work properly, at least not without significant wait times. 

 

Americans don’t like to wait. 
One of the reasons we invented the microwave and internet pornography. 


Yeah, I'm back. Let's not make a big deal about this.


Jylpah #17 Posted 03 November 2020 - 08:47 AM

    Lance-corporal

  • Players
  • 83 battles
  • 38
  • Member since:
    05-04-2015

Here are the final stats for the Mad Games. Mad Games MM was clearly Rigged. 65%+ WR players getting 48% WR on Mad Games tells how Skill-Based MM "works". Thanks everyone for the painful contributions.

 

./analyze_wotb_replays.py --mode team --extra total player_wins --only  --filters  '{ "data.summary.battle_start_timestamp": { "$gte": 1603526400} , "data.summary.room_type": 8 }' db:
Fetching player stats |███████████████████████████▊    | 7725/8932 86% ETA 0 h 0 mins

 

TOTAL_______________   Battles    % Battles     WR     Player WR   Allies WR   Enemies WR
               Total  640    :   100% :  51.2% :    62.80% :    50.23% :     51.73%


Player WR___________   Battles    % Battles     WR     Player WR   Allies WR   Enemies WR
            35 - 45%   1     :     0% : 100.0% :    43.76% :    54.24% :     52.62%
            45 - 50%   28    :     4% :  78.6% :    47.42% :    50.61% :     50.42%
            50 - 55%   33    :     5% :  75.8% :    52.50% :    50.55% :     50.97%
            55 - 65%  305    :    48% :  48.9% :    60.85% :    50.32% :     51.59%
            65 -   %  273    :    43% :  48.0% :    67.86% :    50.03% :     52.11%

 

9_VOLT wrote:

 Maybe Wargaming is and maybe they aren’t. 
Either way, it wouldn’t make any difference to the NA server, when the average queue for these games is only about 30 players. 
 

NA server is just too small for any kind of MM algorithm to work properly, at least not without significant wait times.

 

Here are the stats for the NA server. The same rigging can be observed on NA server too although the dataset is smaller:

 

TOTAL_______________   Battles    % Battles     WR     Player WR   Allies WR   Enemies WR
               Total   63    :   100% :  52.4% :    60.44% :    50.07% :     51.56%
Player WR___________   Battles    % Battles     WR     Player WR   Allies WR   Enemies WR
            45 - 50%   6     :    10% :  83.3% :    47.20% :    51.21% :     50.56%
            50 - 55%   7     :    11% :  71.4% :    52.74% :    50.77% :     52.17%
            55 - 65%   50    :    79% :  46.0% :    63.11% :    49.83% :     51.60%

 

 

9_VOLT wrote:

 Americans don’t like to wait. 

 

It seems Wargaming doesn't care.

 

 



acrisis #18 Posted 03 November 2020 - 11:53 AM

    Captain

  • Players
  • 21303 battles
  • 12,786
  • [III-C]
  • Member since:
    12-27-2014

Thank you for the hard work and analysis. 

  

It sure has felt like fighting with mostly teams where you get one capable person or it is just yourself. 


 

Forum essentials:  > Desktop mode 
Check recent threads. Check WG staff threads. Use search.
    > WG Bugs and current update     > Feedback     > Vehicle Bay     > Guides     > Videos 

 

     BE  (+)   

 [ NEW > ]  Lag, Packet Loss, Ghost Shells, Networking and more demystified usingPingPlotter (free)  
Blitz Community Coalition | Looney Tooners | Triarii
Blitz University | Basic Training  |  Mentor  |  The Pingplotter Guy

 






Also tagged with MM

1 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users