Jump to content


Simulation of the 6.2 MM

simulation 6.2 matlab

  • Please log in to reply
7 replies to this topic

jylpah #1 Posted 06 September 2019 - 07:35 PM

    First Sergeant

  • Player
  • 31611 battles
  • 6,140
  • [AFK]
  • Member since:
    08-05-2014

 

I have felt tanking been too boring so I did a series of simulations of how the 6.2 would impact on the players' WR. As always, the simulations reflect the reality only as much as the underlying assumptions match. The simulation is run on Octave (Matlab). 

 

The boring background

 

This playerSkill vector provides very accurate results (i.e. simulated WR matches the target WR for players. It is bit hard to see, but the targetWR is exactly in the center of the simulated WR:

 

 

 

You can see that in the extremes there is some variations from the targetWR, but otherwise the simulation provides very accurately the WR distribution we wanted to get.

 

Next I started to change the MM by taking players of different WR levels out from the main queue. I tested cut-offs of 30%, 35%, 40% and 45%. Here is how the removal of low WR players from  the MM queue impacted to WR after I rerun the simulations with a subset of the total player population (delta to target WR).

 

30% WR cut-off did not impact the WR much at all mostly since there are so few players below that WR.

 

35% WR cut-off had only a modest impact. Please do not read too much of the "largest" impact being on the high WR players because that is not caused by the removal of low WR players, but by a larger difference between simulated WR and target WR among high WR players (see the graph above).

 

 

 

But 40% cut-off started to have some impact. Basically lower WR players lost 1% of their nominal WR since the battles became harder for them. Remember, WR is a relative term and removing low WR players will increase the skill floor thus making all the players relatively worse than before.

 

 

 

45% WR is not considered that high in the game, and moving all the sub 45% players to their own MM queue had even larger impact. 45-50% WR players lost 3% of their WR, 65%+ players 1.6%.

 

 

I was expecting larger impact, but at modest WR cut-offs (30-35%) the impact was rather small.

 

The next simulation topic will be to estimate the impact of platooning at different WR levels. I have attached the Octave/Matlab files to this post in case someone is curious to play around those. There is lot of crap there since I have developed the scrips over longer time (initially to test CanCracker's MM proposal as documented in the MM thread in 2016). It might be really hard to decipher what has been the intention.

 

usage (and relevant files)

loadPkgs;

initBattleSigma;

plotStats(playerSkill, targetWR, simWR)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


gallabru #2 Posted 07 September 2019 - 03:32 PM

    First Sergeant

  • Player
  • 18237 battles
  • 1,600
  • [LAGER]
  • Member since:
    09-01-2016
Nice.
Not sure about what you mean by 'removing players from the queue'. These are numerical simulations ? How do you estimate the outcome of the 'battles' ?

Edited by gallabru, 07 September 2019 - 03:33 PM.

My team stats in real time  : http://forum.wotblit...ats-repository/

Carlsberg clan family is recruiting. Contact us.


jylpah #3 Posted 07 September 2019 - 07:35 PM

    First Sergeant

  • Player
  • 31611 battles
  • 6,140
  • [AFK]
  • Member since:
    08-05-2014

View Postgallabru, on 07 September 2019 - 05:32 PM, said:

Nice.
1) Not sure about what you mean by 'removing players from the queue'.
 
2) These are numerical simulations ?
 
3) How do you estimate the outcome of the 'battles' ?

 

1) It means that players with lower WR than the threshold will be moved away from the main MM queue. The simulation considers only the main MM queue.

 

2) The simulations are Monte-Carlo / numerical

 

3) Battle outcomes are estimated as it follows:

 

- Each player has skill-level: player i skill is playerSkill(i)

- Team strength is average of the seven players' skills

- First Draw battles are simulated by calculating a (uniform) random number betweem [0,1]. If the random number is smaller than equal of DrawRate (2.2249%), the battle is a draw.

- Otherwise, a cumulative normal distribution function is calculated for Team A strength - Team B strength. The value is between [0,1]

- A (uniform) random number is calculated between [0,1]

- If the random number is smaller than value of the cumulative normal distribution function, Team A wins, otherwise Team B wins

 

Here is the matlab code for it. The function calculates multiple battles in a run to speed up the simulation. All the variables are vectors below:

 

battlesim.m

 function [result, p_winA] = battlesim(playerSkill, A, B)
    # runs multiple battles at once
  global drawRate;
    global sigmaBattle;
    ##global playerSkill;
    
    teamSkill_A = mean(playerSkill(A),2);
    teamSkill_B = mean(playerSkill(B),2);
    
    P_winA = normcdf(teamSkill_A-teamSkill_B,0,sigmaBattle);
        
    % Random battle
    s = size(P_winA);
    rnd = rand(s);
    
  result = zeros(s);
  result(rnd <= P_winA) = 1;
  result(rnd > P_winA) = -1;
 
  draw = rand(s);
  result(rnd <= drawRate) = 0;
      
endfunction

 

 


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


Mjr_Leicht #4 Posted 08 September 2019 - 09:38 AM

    Senior Sergeant

  • Player
  • 15249 battles
  • 895
  • [BT_SV]
  • Member since:
    01-21-2019

Interesting stuff...

One thing, I thought WG had said they were basing this new boot camp MM on something other than WR...


560273467.png

Jukkis74 #5 Posted 08 September 2019 - 10:01 AM

    First Sergeant

  • Player
  • 40268 battles
  • 1,097
  • [AFK]
  • Member since:
    06-29-2014

View PostMjr_Leicht, on 08 September 2019 - 11:38 AM, said:

Interesting stuff...

One thing, I thought WG had said they were basing this new boot camp MM on something other than WR...


Yes they said so, but in practice it (selected players to other MM) will be pretty much the low end of MM distribution. Some will just jump back to normal MM every now and then.



jylpah #6 Posted 08 September 2019 - 12:07 PM

    First Sergeant

  • Player
  • 31611 battles
  • 6,140
  • [AFK]
  • Member since:
    08-05-2014

View PostMjr_Leicht, on 08 September 2019 - 11:38 AM, said:

Interesting stuff...

One thing, I thought WG had said they were basing this new boot camp MM on something other than WR...


Yes and WG hasn’t disclosed how exactly they pick the players. The purpose of this simulation was to assess the impact on game performance of different skill-level players. Therefore the model is simplified and static. The players who got into the bootcamp/noob queue never get out of there. In the model each player has their skill-level and they never improve or get worse. 


BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  


Titus_Scato #7 Posted 09 September 2019 - 02:17 AM

    First Sergeant

  • Player
  • 42260 battles
  • 8,632
  • Member since:
    10-18-2014

View Postjylpah, on 08 September 2019 - 12:07 PM, said:


Yes and WG hasn’t disclosed how exactly they pick the players. The purpose of this simulation was to assess the impact on game performance of different skill-level players. Therefore the model is simplified and static. The players who got into the bootcamp/noob queue never get out of there. In the model each player has their skill-level and they never improve or get worse. 


I was thinking of raising a support ticket, and asking to be made 'an honorary noob'.

 

Don't think I'll bother now, after reading this.

 

;)

 

 



jylpah #8 Posted 09 September 2019 - 08:40 PM

    First Sergeant

  • Player
  • 31611 battles
  • 6,140
  • [AFK]
  • Member since:
    08-05-2014

View PostTitus_Scato, on 09 September 2019 - 04:17 AM, said:


I was thinking of raising a support ticket, and asking to be made 'an honorary noob'.

 

Don't think I'll bother now, after reading this.

 

;)

 

 

 

That was _exactly_ the experience that motivated me to run the simulation. I felt I have been CanCrackered sooo hard after the 6.2. Only now recovering a little. Just checked, last 14 battles with T-54: avg dmg 2256, WR 50%. WN8 3155 deep purple.

 

Yes. small sample.

:izmena:

 


Edited by jylpah, 09 September 2019 - 08:43 PM.

BlitzAnalysiz.com  In-depth analysis of WoT Blitz game statistics

Jylpah @EU  Jylpah @Asia  Blitz-tools  






Also tagged with simulation, 6.2, matlab

1 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users