Jump to content


[WN7] What is it and how does it work?

statistics stats noobmeter performance efficiency metric compare quality rating Garbad

  • Please log in to reply
2483 replies to this topic

Praetor77 #1041 Posted Jan 15 2013 - 18:39

    Captain

  • Players
  • 29723 battles
  • 1,675
  • Member since:
    05-01-2011
Nice Neatoman! The problem is I see too much score being given at higher winrates... your scurve in WN4 capped at around 450 points, which multiplied by 0.5 gave a total of around 225 points for winrate at 70% wins.

Could you give us a graph like the tier 2 one, but for tiers 5 and 8 pls?

And how are you calculating how much burtoklov and lironman would lose? According to my calculations it would be more like 300 points. Also, this wouldnt work to weed out sub 4k game sealclubbers (most of them dont have that many games).

Edited by Praetor77, Jan 15 2013 - 18:41.


NeatoMan #1042 Posted Jan 15 2013 - 20:08

    Major

  • Players
  • 28180 battles
  • 20,407
  • Member since:
    06-28-2011

View PostPraetor77, on Jan 15 2013 - 18:39, said:

And how are you calculating how much burtoklov and lironman would lose? According to my calculations it would be more like 300 points. Also, this wouldnt work to weed out sub 4k game sealclubbers (most of them dont have that many games).
I just estimated by how much of a drop there was in the WR portion of the graphs.  I forgot to factor in the 0.5 you added to my formula, so my guess was off by at least half.

tier 5 and 8 show very little change with number of games played.  You gain a few more points the faster you go up the tiers, and the higher tier you play, but the maximum you will gain is only ~20 more points over the previous WR formula.

I forgot to factor the 0.5 in these graphs too.
tier 5
tier 8

Edited by NeatoMan, Jan 15 2013 - 20:39.


Praetor77 #1043 Posted Jan 15 2013 - 23:45

    Captain

  • Players
  • 29723 battles
  • 1,675
  • Member since:
    05-01-2011
That's pretty interesting... what do the rest think? Let me check how much WN would drop with 0.45 multiplier for a 2500 games tier 1 troll... or tier 2 troll...

NeatoMan #1044 Posted Jan 16 2013 - 02:02

    Major

  • Players
  • 28180 battles
  • 20,407
  • Member since:
    06-28-2011
Here is an alternative way to treat seal clubbers using the WR formula.  Rather than messing with the c constant, i modified the constant b in the same manner

WReff = [(a / ( b + exp((WR - c) * d))) - e] * 0.45

a = 185
b = 0.17
c = 35
d = -0.134
e = 500

b =  0.163 + (games played / 1000) / [40 * (average tier - 0.9)^2 ]   with a cap of 5000 games

The mid and upper tiers are unaffected as before, and the clubbing tiers have a lower maximum at the high end of win rates

Tier 2b  (new revision)
Tier 2 (previous) -  for comparison


I can combine the two to get the effect throughout the win rate range, rather than just the higher end, but that would probably be too much.  I think this revision accomplishes the task adequately, and better than before.

Edited by NeatoMan, Jan 16 2013 - 05:50.


Makaze2048 #1045 Posted Jan 16 2013 - 02:14

    Captain

  • Players
  • 27401 battles
  • 1,017
  • Member since:
    01-21-2012
New revision is definitely better as it maintains the curve shape, just pulls it down. The original caused too much distortion.

So the question in my mind is then which is better, this or pulling down kill counts in low tiers like WN5 is doing? Not sure both are needed (though maybe they are, each at half strength may be better than either alone...)

Guerdon #1046 Posted Jan 16 2013 - 04:33

    Corporal

  • Players
  • 2079 battles
  • 76
  • Member since:
    09-02-2012
I don't mean to sound dismissive but IMO you've got too hung up on baby seal clubbers to the point that this statistic is becoming ridiculous despite perhaps retaining some overall correlation with Win Rate (i.e. your proxy for skill).

Let's look at each part of the WN5 calculation in isolation:

Kills - One kill is worth the following number of WN5 points depending upon the tier:
= (1) 200, (2) 312, (3) 371, (4) 411, (5) 439, (6) 460, (7) 475, (8) 483, (9) 484, (10) 477
This follows a logarithmic path, sharply rising, starting to flatten out after tier 5, almost flat around tier 8 and actually falling after tier 9.  Regardless, this basically implies that one automatically gets more skilful the higher the tier one plays which we all know is incorrect, especially as the actual average number of kills broadly goes up (i.e. overall trend is up) as you go up the tiers.  WN5 therefore favours those playing in the higher tiers (especially tier 5 or more) from a kills point of view.

Damage - One damage point is worth the following number of WN5 points depending upon the tier:
= (1) 1.37, (2) 1.17, (3) 0.98, (4) 0.82, (5) 0.67, (6) 0.55, (7) 0.45, (8) 0.36, (9) 0.29, (10) 0.23
Unsurprisingly, due to the formula, this takes an exponential path.  Initially this would appear sensible (i.e. one gets less WN5 points per damage point as one moves up the tiers because one deals more damage with each shot).  If we ignore the 500 coefficient multiplier for damage (i.e. higher than the WN4 coefficient of 460) one can see the effective damage targets per tier:
= (1) 364, (2) 427, (3) 508, (4) 611, (5) 741, (6) 907, (7) 1117, (8) 1385, (9) 1725, (10) 2158
It can therefore be seen that the target damage in the lower tiers is still relatively much higher in the low tiers than it is in the high tiers compared to the actual average damage dealt.  It is also useful to compare damage with kills.

Kills / Damage - One kill gets the same number of WN5 points as the following damage points depending upon the tier:
= (1) 146, (2) 266, (3) 377, (4) 502, (5) 651, (6) 835, (7) 1061, (8) 1,338, (9) 1671, (10) 2059
This too takes an exponential path.  At first glance, this looks reasonable whereby the number of damage points to equate to a kill rises as one goes up the tiers.  However, it is no where near representative of the average hit points per tier (neither absolutely nor relatively).  At the lower tiers (i.e. 1-6 & especially 2-5) it is well above but in the higher tiers of 7-10 it is broadly equivalent to the hit points of a medium tank.  In short this means that like kills, WN5 significantly favours those playing in the higher tiers from a damage point of view.

Spots - One spot gets the same number of WN5 points regardless of tier.
Again this looks reasonable at first glance as it would appear that there's no reason why a spot at one tier should be valued any different to another.  However, this is not the case because you're adjusting the other metrics for tier level.  Consequently, one must compare to the other metrics to gain relative meaning.  So, one spot is the equivalent of the following number of kills depending upon the tier:
= (1) 62.5%, (2) 40.1%, (3) 33.7%, (4) 30.4%, (5) 28.5%, (6) 27.2%, (7) 26.3%, (8) 25.9%, (9) 25.8%, (10) 26.2%
So, whilst in tiers 7-10 one needs to spot nearly 4 tanks to get the same WN5 points as one gets from a kill, in tier 1 spotting 8 tanks is the equivalent of killing 5, which will simply encourage suicide scouters.

Defence - This is maxed at 2.2 and there's no reason why any player should have too much trouble in reaching this target if they wanted to, which makes it kind of pointless anyway.  Regardless, we'll assume that the maximum has been reached and compare that to kills (i.e. for the same reasons and in a similar way to spots).  The maximum defence points is the equivalent of the following number of kills depending upon the tier:
= (1) 1.10, (2) 0.71, (3) 0.59, (4) 0.54, (5) 0.50, (6) 0.48, (7) 0.46, (8) 0.46, (9) 0.45, (10) 0.46
Obviously, we see the same pattern as for spots whereby one needs 2.38 times the number of kills in tier 1 compared to tier 10 to get the same number of WN5 points that one gets from maximising the defence points (& for each spot).

Win Rate - No WN5 points are scored for a 47.01% win rate, with negative scores below this and positive scores above it.  This appears too low as I would guess (although I admit I don't know for sure) that the average is probably 48%.  Theoretically, it should be around 49% (i.e. 49% wins, 49% losses & 2% draws) but I suspect there are many well above average players which will push the average down 1% (but 2% seems a bit of a stretch; however I could be wrong).  It's fairly linear from 40% to 60% (especially 42% to 52%) and so should help differentiate the majority of players.  It starts to offer diminishing returns above 59% and once you get above this win rate, one gets less than a 10 point increase in WN5 for every 1% of extra win rate.

Guerdon #1047 Posted Jan 16 2013 - 04:35

    Corporal

  • Players
  • 2079 battles
  • 76
  • Member since:
    09-02-2012
If someone wants to get a high WN5 score, they should kill in the middle to high tiers (ah, look at all those mid tier derpers), damage in the highest tiers possible (again, no real surprise there as WN5 favours high tiers overall), spot in the low tiers (especially tier 1, ah look at all those suicide scouters) and get defence points in the low tiers (ah, look at all those low tier base campers looking to pick off the scraps from everyone else's hard work).

So, the path to Super Unicum WN5 status with balanced overall statistics:

Low tiers - Fly around with the T1 Cunningham getting lots of initial spots before hiding near to the base (assuming you're still alive) to wait for the enemy to start capping before rushing in and resetting with your machine gun to get defence points too.  Similar tactics can be repeated with the M2 LT, M3 Stuart, BT-2, BT-7, A-20 & T-50 etc.  People talk about baby seal clubbing with the Leichttraktor but kills & damage are now relatively irrelevant at this level compared to spots (& maximising defence points is worth more than 1 kill too). 8 spots & 2 defence points = 6 kills!!!  It is 1,200 WN5 points which will put you in the top 25% without having to fire a round (if you got your 2 defence points from ramming).  Add a few damage & kill points and you're well on the way.

Mid tiers - Derp with guns likely to 1 shot kill, especially with gold rounds (i.e. M4 Sherman, KV-1, KV-2, Pz IV, VK 3601 etc.).  Can be combined with base camping to ensure 1 shot kills on weakened cappers.  Can include some fast & / or hard hitting lights to keep up the spots & / or defence points (i.e. T-20-2, ELC AMX & VK2801).

High tiers - Take the best tanks for dealing maximum damage (i.e. most TDs plus heavies or mediums depending upon your preferred play style).  Being on the front line will get you spots and base camping / arty defence will get you some defence points.  Don't forget to ensure above average damage along the way to the top tiers with plenty of TDs and especially some arty (i.e. Russian particularly SU-26 and French so end up with BatChat 155).

Overall - Get up the tiers as quickly as possible as WN5 massively favours the high tiers, especially tier 5 or more, particularly for kills and damage which each account for around 1/3 of the overall WN5 score with the other third being split relatively evenly between spots, defence points and win rate.

Edited by Guerdon, Jan 16 2013 - 04:37.


Guerdon #1048 Posted Jan 16 2013 - 04:36

    Corporal

  • Players
  • 2079 battles
  • 76
  • Member since:
    09-02-2012
Ah, well what a surprise, the above is not so different from what we see happening (except perhaps for the low tiers which is what these recent manipulations were meant to protect against but instead just introduce a different problem) and so WN5 is basically a statistic which reflects how people would play to maximise their statistics anyway.

Unfortunately, I think you've got a case whereby you're trying to make the overall number fit the target, which is fine but by adjusting each independent variable to overcome its weakness such that whilst the whole might look OK, it's made up of erroneous parts which just happen to cancel each out to a large extent.

Basically, the better players will generally have played many games and thus will generally have a high average tier.  This is primarily because they maximise their XP in the highest tiers and maximise their credits in Tiers 5-8, which funds their Tier 10 clan wars tanks, which gives them the gold they need to run premium accounts, have complex equipment, premium consumables and 100% gold retrained, highly skilled, crews etc.  All of this helps make them to be a better player.  In other words, you've just got a self fulfilling prophecy.

So, you're simply saying that to get the highest rankings most players must move up to the higher tiers but the best players can still stat pad in the low to medium tiers if they want to.  Whilst, on average, the skill level may be higher in the higher tiers, simply playing in the higher tiers does not make you a better player (even if it does provide you with the things that might make you a better player).

My conclusion is that WN5's overall bias towards the higher tiers is badly flawed and it only partially negates the effect of low to medium tier stat padders.

Makaze2048 #1049 Posted Jan 16 2013 - 05:38

    Captain

  • Players
  • 27401 battles
  • 1,017
  • Member since:
    01-21-2012

View PostGuerdon, on Jan 16 2013 - 04:33, said:

I don't mean to sound dismissive but IMO you've got too hung up on baby seal clubbers to the point that this statistic is becoming ridiculous despite perhaps retaining some overall correlation with Win Rate (i.e. your proxy for skill).
WR is a flawed proxy for skill and so adjustments must be made when the produced results do not match the desired results (ie. UMBs having higher rankings than Gs). That is admittedly a subjective criteria but seems to be the consensus on what the community wants.

View PostGuerdon, on Jan 16 2013 - 04:33, said:

Kills - One kill is worth the following number of WN5 points depending upon the tier:
= (1) 200, (2) 312, (3) 371, (4) 411, (5) 439, (6) 460, (7) 475, (8) 483, (9) 484, (10) 477
This is due to the anit-seal clubbing stuff in WN5. I do think that perhaps the reduction curve could be slightly sharper so that the plateau happens at tier 5 or 6 rather than tier 8 and that the dip at 10 is odd and should be eliminated. While I'm personally not OK with the metric explicitly favoring high tiers, I am perfectly OK with it explicitly not favoring low tiers. So I agree that the curve could use some tweaks perhaps.

In a way this is also an argument for NeatoMan's games played method of pulling out seal clubbers.

View PostGuerdon, on Jan 16 2013 - 04:33, said:

Damage - One damage point is worth the following number of WN5 points depending upon the tier:
= (1) 1.37, (2) 1.17, (3) 0.98, (4) 0.82, (5) 0.67, (6) 0.55, (7) 0.45, (8) 0.36, (9) 0.29, (10) 0.23
= (1) 364, (2) 427, (3) 508, (4) 611, (5) 741, (6) 907, (7) 1117, (8) 1385, (9) 1725, (10) 2158
That might actually be an issue, the curve does appear biased towards high tiers seeming to start fitting properly around tier 7 and continuing through 10. Not overly concerned about tire 1-3 but it would seem that tier 4-6 may not be getting a fair shake.

View PostGuerdon, on Jan 16 2013 - 04:33, said:

Kills / Damage - One kill gets the same number of WN5 points as the following damage points depending upon the tier:
= (1) 146, (2) 266, (3) 377, (4) 502, (5) 651, (6) 835, (7) 1061, (8) 1,338, (9) 1671, (10) 2059
This actually seems lined up rather properly. If one looks at the heavies in each tier they are just above each of those numbers.

View PostGuerdon, on Jan 16 2013 - 04:33, said:

Spots - One spot gets the same number of WN5 points regardless of tier.
= (1) 62.5%, (2) 40.1%, (3) 33.7%, (4) 30.4%, (5) 28.5%, (6) 27.2%, (7) 26.3%, (8) 25.9%, (9) 25.8%, (10) 26.2%
The argument here basically boils down to applying the seal clubbing pull down across the finished product rather than just kills so as not to encourage a specific playstyle that is not effected by it. Agreed.

View PostGuerdon, on Jan 16 2013 - 04:33, said:

Defence - This is maxed at 2.2 and there's no reason why any player should have too much trouble in reaching this target if they wanted to, which makes it kind of pointless anyway.
And yet somehow people don't, which is why it's useful.

View PostGuerdon, on Jan 16 2013 - 04:33, said:

Win Rate - No WN5 points are scored for a 47.01% win rate, with negative scores below this and positive scores above it.  This appears too low as I would guess (although I admit I don't know for sure) that the average is probably 48%.  Theoretically, it should be around 49% (i.e. 49% wins, 49% losses & 2% draws) but I suspect there are many well above average players which will push the average down 1% (but 2% seems a bit of a stretch; however I could be wrong).  It's fairly linear from 40% to 60% (especially 42% to 52%) and so should help differentiate the majority of players.  It starts to offer diminishing returns above 59% and once you get above this win rate, one gets less than a 10 point increase in WN5 for every 1% of extra win rate.
You're assuming that the idea was to make the WR component both contribute and subtract in equal measurement across the player base. But there's not reason it couldn't be all contribution, all that changes is your scale and thus your overall coefficient values to keep things normalized.

View PostGuerdon, on Jan 16 2013 - 04:36, said:

Ah, well what a surprise, the above is not so different from what we see happening (except perhaps for the low tiers which is what these recent manipulations were meant to protect against but instead just introduce a different problem) and so WN5 is basically a statistic which reflects how people would play to maximise their statistics anyway.
I would rephrase that as people playing to maximize their chances of winning, and that being the case that's exactly what we're after here.

View PostGuerdon, on Jan 16 2013 - 04:36, said:

Basically, the better players will generally have played many games and thus will generally have a high average tier.  This is primarily because they maximise their XP in the highest tiers and maximise their credits in Tiers 5-8, which funds their Tier 10 clan wars tanks, which gives them the gold they need to run premium accounts, have complex equipment, premium consumables and 100% gold retrained, highly skilled, crews etc.  All of this helps make them to be a better player.  In other words, you've just got a self fulfilling prophecy.
That is a tiny fraction of the player base you're describing. But ignoring that it's a self fulfilling prophesy for multiple reasons not least of which is selection bias. In other words the people who are already good at tanks are the ones that stick around long enough to get all that stuff that makes them even better. I don't see that as something that should matter in this context as we're attempting to measure performance and unfortunately there's not way to filter for equipment or consumables and so it must be considered as part of their performance. Not to mention rest assured if you play them then they will have all that stuff and thus perform up to their rating.

View PostGuerdon, on Jan 16 2013 - 04:36, said:

My conclusion is that WN5's overall bias towards the higher tiers is badly flawed and it only partially negates the effect of low to medium tier stat padders.
My conclusion is that it's less flawed than WN4 but could potentially use some additional tuning of the curves Which means progress is being made and that's what matters. That NeatoMan's alternate anit-sealclub method is worth a look at in greater depth. And that the pull down for low average tier should perhaps be applied to all contributors, not just kills.

Praetor77 #1050 Posted Jan 16 2013 - 12:58

    Captain

  • Players
  • 29723 battles
  • 1,675
  • Member since:
    05-01-2011
I agree with Makaze 100% on his responses to Guerdon.

Also, Guerdon, we are focusing on seal clubbers because that´s what epople want and it´s what WN4 was actually failing at. Nevertheless, I wanted to point out that 97% of the players in the random sample had changes in WN5 that were less than 10% of their final score. Of course the 3% that had larger changes were players with more than 2000 batltles and avg tier played 1-3.

I think Neatoman´s curve will be a very good thing to implement, so I can tone down the damage and kill bias (though kill bias is actually based on observed data.take a look at vbaddict´s top tanks for kills/game:

1.   MS-1  2.56
2.   Leichttraktor  2.47
3.   T18  2.36
4.   PzKpfw II Ausf. J  2.29
5.   Panzerjaeger I  2.28
6.   T2 Medium Tank  2.28
7.   T1 Cunningham  2.24
8.   PzKpfw 38H735 (f)  2.19
9.   AT-1  2.13
10.   SU-26  2.09
11.   PzKpfw B2 740 (f)  2.02
12.   Vickers Medium Mk. I  2.00
13.   Hetzer  1.95
14.   Tetrarch  1.95
15.   T57  1.94

All low tier tanks with tier 1s being NASTILY over the top)


And I also purpusefully extended the drop in scores through tier 3 and 4  (though much less than tiers 1 and 2). 5 is lightly penalized, and 6 is barely touched.

Back to Neatoman´s curve, I see a problem. We will keep getting stinky 100 game seal bashers on the top of low tier rankings, instead of the good 7000 game seal bashers... :Smile_sceptic:

Edited by Praetor77, Jan 16 2013 - 13:09.


NeatoMan #1051 Posted Jan 16 2013 - 13:32

    Major

  • Players
  • 28180 battles
  • 20,407
  • Member since:
    06-28-2011

View PostPraetor77, on Jan 16 2013 - 12:58, said:

We will keep getting stinky 100 game seal bashers on the top of low tier rankings, instead of the good 7000 game seal bashers... :Smile_sceptic:
i purposely kept the # of games a little on the higher end before seeing a drop in scores. I can play with it some more, and see if I can get it to start kicking in earlier, without being unreasonable for the average player.

Valan #1052 Posted Jan 16 2013 - 13:58

    Major

  • Players
  • 34661 battles
  • 2,059
  • Member since:
    07-13-2012

View Postmaokai, on Jan 15 2013 - 02:44, said:

A quick eyeball test should tell you that the WN5 score is drastically affected by the average tier played.  Is that because they are much better players, or because they spend so much time playing top tier tanks?

One other byproduct of purchasing Tier 8 Premium Tanks like the T34, JT88, Type 59 and using Gold have 100% trained crews and buying a year long Premium Account is maximize Crew Skill & Perk training.  

Why do this? It eliminates experience point loses, incurred when a non-paying player otherwise has to move up a normal tiered tree seven times in a row using the same Crew. Each time a new tank is bought for silver credits, that Crew has to redo its majority training to some degree. One player lose experience points, while another player has those same experience points applied towards an winning edge: Crew secondary training.


What some individuals see as "Baby Seal Clubbing" behavior is to many other non-paying players, an efficiency drive to reduce frictional costs and while maximizing possible rewards given a set limitations created by Wargaming incorporated.  By playing fewer tanks than paying players, these so-called "Baby Seal Clubbers" are seeking to maximize their attempts to earn a 50% experience bonus while at the same time seeking reduce the known move-up frictions. Whereas paying players seem to lack a move-up disincentive, because these same frictions are removed and instead additional bonuses for credit & experience are received regardless of a players ability.  It simply takes a skilled "Baby Seal Clubber" much more time to receive their Crew's first Skill or Perk than it does for a "Wallet Warrior" Crew to receive their Crew's first Skill or Perk.  An equally skilled "Wallet Warrior" possesses multiple built in advantages, rigged by Wargaming Inc regardless of talent or skill--all of which skew player publicly released play statistics and known ranking systems.


If a World of Tank efficiency system truly seeks to measure player talent or skill accurately, it should have to attempt to tease out the relative impacts of gold on outcomes. Even more interestingly if an efficiency system could do this, it would in fact show players to what degree Crew secondary Skill & Perks effect outcomes.  Is a "Wallet Warrior" inefficiently spending their gold? Is a "Baby Seal Clubber" wise to emphasize slowly rising up the tiers, to maximize outcomes?



Finally to Maokai above question, it's likely because the top WN5 are generally Warrior Wallets and seek to maximize their experience points gained on a per match basis.

Edited by Valan, Jan 16 2013 - 14:11.


NeatoMan #1053 Posted Jan 16 2013 - 22:18

    Major

  • Players
  • 28180 battles
  • 20,407
  • Member since:
    06-28-2011
I almost have something worked out, I just have some questions to fine tune it

- What average tier would you start factoring in seal clubbing corrections?  only below a tier 3 average?  higher?

- after how many games would you expect it to take for someone progressing slowly (non-premium acct) to be beyond the average tier you answered above?  200 games?  500? 1000?

- approximately how much of an overall correction would you expect for an extreme clubber?  e.g. minus 200 eff?  400?  1000?

Makaze2048 #1054 Posted Jan 16 2013 - 23:40

    Captain

  • Players
  • 27401 battles
  • 1,017
  • Member since:
    01-21-2012

View PostNeatoMan, on Jan 16 2013 - 22:18, said:

- What average tier would you start factoring in seal clubbing corrections?  only below a tier 3 average?  higher?

- approximately how much of an overall correction would you expect for an extreme clubber?  e.g. minus 200 eff?  400?  1000?
I'm not sure. I've been thinking about various examples and average tier is starting to feel a bit... blunt to me. For example, a player could play 4000 tier 3 games and then only 1000 tier 8 games in their prem. That person would then have an average tier of 4, out of what most would consider seal clubbing range and yet they're almost certainly a seal clubber. The presence of tier 8 prems and their ability to skew average tier and the inability to tell if a player has an even distribution of tiers within that average is making it difficult for me to pick a good tier point. Once again dossier files as opposed to the profile data would make this so much easier (though technically you can get tier played counts off of the profile data, but it's definitely more complicated both in terms of data collection and usage).

Does it need to be a flat subtraction or should it lower the results by a percentage?

NeatoMan #1055 Posted Jan 17 2013 - 01:58

    Major

  • Players
  • 28180 battles
  • 20,407
  • Member since:
    06-28-2011
Here's is a graph of the correction factor vs average tier, at a different intervals of games played (colored lines)

http://home.comcast....mages/WRfix.jpg
with diagram here

The formula adjusts on it's own as you play more and more games.  Each colored curve is an interval of games played from 100 to 2000 games.  They all top out near 100% of the win rate contribution to WN#, which is around +260 eff pts .  When corrected for seal clubbing we drive that maximum down to a level according to where you fall with your avg tier on the games played curve.  I have the most severe corrections set to 30%, which translates into a maximum of -80 eff points for those tiers most affected.

The more games you play, the higher the avg tier that gets affected by the correction. At 100 games played (first black curve) a tier 1 drops to 60%, while tier 3 nears 100%.  At 2000 games played (last brown curve) tier 2 and below bottom out at 30%, and you near 100% around tier 4.5.

The link below shows how the correction factor affects the win rate S curves.  It shows how the correction factor "squashes" the S curves for the affected tiers.  Each line represents the S curves you get for tiers 1 through 8 by following along the brown 1000 games curve on the above graph.

e.g. tier 2 on the 1000 game curve on the above graph leads to ~39% correction factor.  Applying that fix to the tier 2 S curve squashes it down to ~ 39% of the original size (blue line on graph below), leading to negative win rate contributions to the WN# formula

http://home.comcast....ges/WRfix1k.jpg


- we can change where the curve bottoms out anywhere from 0 to 100% (or -220 to +250 eff pts); 0% would squash the S-curve totally flat on the bottom around -220 eff pts.   The 50% correction yields a maximum of 0 pts at high win rates.

- by selecting where to cap the number of games we can limit at what tier and how much the curve drops off.

- games played can also be adjusted to makes the curves tail off sooner (or later), e.g. make 500 games look like the 1000 curve, etc

Edited by NeatoMan, Jan 18 2013 - 03:57.


Boom_Box #1056 Posted Jan 17 2013 - 02:13

    First lieutenant

  • Beta Testers
  • 11752 battles
  • 905
  • [REL-V] REL-V
  • Member since:
    11-17-2010

View PostMakaze2048, on Jan 16 2013 - 23:40, said:

The presence of tier 8 prems and their ability to skew average tier and the inability to tell if a player has an even distribution of tiers within that average is making it difficult for me to pick a good tier point. Once again dossier files as opposed to the profile data would make this so much easier (though technically you can get tier played counts off of the profile data, but it's definitely more complicated both in terms of data collection and usage).

While collection and usage is certainly more complicated, it is doable from the profile and API.

What part of my (scripted) profile looks like:

Posted Image


yes, there is a bit of seal clubbing going on there...

Praetor77 #1057 Posted Jan 17 2013 - 16:26

    Captain

  • Players
  • 29723 battles
  • 1,675
  • Member since:
    05-01-2011
Non-trolls avg tier is above 4 at 500+ games.

I wouldnt make the penalty bigger with more games, I would just actually apply a penalty to WR portion of formula to all avg tier 1-4 players who have more than 1000 games played, or something like that. IMHO it doesnt make much sense to punish more the more games they play...

Edited by Praetor77, Jan 17 2013 - 16:43.


NeatoMan #1058 Posted Jan 17 2013 - 16:58

    Major

  • Players
  • 28180 battles
  • 20,407
  • Member since:
    06-28-2011

View PostPraetor77, on Jan 17 2013 - 16:26, said:

Non-trolls avg tier is above 4 at 500+ games.

I wouldnt make the penalty bigger with more games, I would just actually apply a penalty to WR portion of formula to all avg tier 1-4 players who have more than 1000 games played, or something like that. It doesnt make much sense to punish more the more games they play...
The way the formula works is that the punishment is minimal with few games played and gradually grows as you play more.  This reflects the fact that you should be moving up the tiers, but aren't.  The way you limit the punishment would be to cap games played at 1000, for example.  Anything higher than that would continue to use the 1000 game curve.

  No sense in applying the full correction right away as this hurts non-clubbers, or applying it only as you cross the 1000 game threshold, which would give a large drop suddenly at 1000 games.  Unless you don't care about those with fewer than 1000 games, then we can simply apply the full fix from the start.


edit:  maybe it's unclear what I can do with this formula.
http://home.comcast....RfixDiagram.jpg

I have set it up so that by altering any number of constants I can
- change the magnitude of correction
- change when the corrections kick in in terms of games played
- change how quickly the corrections increase with games played
- change at what tier the corrections start to kick in
- change how much correction per tier
- change the maximum tier affected by a correction
- distinguish between clubber and non-clubber by # of games played

Right now i'm only guessing with the parameters, presenting graphs and you guys are saying "too much here, not enough there".  How about you tell me what you want, and I'll set the parameters to match.

- how much of an efficiency penalty should a 70% win rate tier 1 clubber have (e.g. those UMB players)?  Tier 2 clubber? Tier 3 clubber?
-The full range of the win rate portion currently accommodates ~500 eff pts, how much of that would you subtract for each of those clubbers?
- after how many games is someone considered a clubber?  is a player with avg tier 2 and 100 games a clubber? no.  at 1000 games?  yes.  Do you even care about someone with fewer than 1000 games?
- at what point do you not consider clubbing?  you mentioned tier 4 and 500 games.
- Do you even care about all this and only want a fixed correction to be applied at a specific point?  How much and what tiers?

Edited by NeatoMan, Jan 18 2013 - 04:02.


Chu_Man_Phat #1059 Posted Jan 17 2013 - 22:53

    First lieutenant

  • Players
  • 23003 battles
  • 658
  • Member since:
    09-07-2011
Omg, strolling through all these posts make one thing very apparent to me.

You guys have absolutely no life whatsoever.

Guerdon #1060 Posted Jan 18 2013 - 00:31

    Corporal

  • Players
  • 2079 battles
  • 76
  • Member since:
    09-02-2012

View PostMakaze2048, on Jan 16 2013 - 05:38, said:

The argument here basically boils down to applying the seal clubbing pull down across the finished product rather than just kills so as not to encourage a specific playstyle that is not effected by it. Agreed.

Indeed!  The anti baby seal clubbing was first applied to Damage and now has been applied to Kills too but is still absent from Spots & Defence points.  You're quite correct that it needs to be applied across the finished product and is for what I've been asking since the very start.

Once again, IMO the only way to have a sensible output is to get an underlying measure of actual skill demonstrated without any adjustment.  If required (which it appears it is) one then applies a factor to this underlying rating, which is likely to come from average tier played (and possibly total number of games) to come up with an overall ranking.

On a few of other points you raised:

- I think the consensus is purely from the high average tier players, with many games, who want to see themselves at the top of the rankings and don't appreciate those with a lower average tier appearing above them.

- Kills - I realise the reason but think it's being applied incorrectly and far too onerously for the lower tiers in general.  There are better ways to deal with the baby seal clubbing issue, as mentioned above.

- Damage - It's great that you recognise the bias towards the higher tiers here too.

- Kills / Damage - You say that the damage targets per kill is about right for heavies.  As mentioned, I looked at Mediums because there are no heavies in the lower tiers and this is the approximate percentage of target damage / actual average HP per tier - 1 (i.e. to show the additional percentage required:
= (1) 12%, (2) 64%, (3) 75%, (4) 54%, (5) 63%, (6) 17%, (7) 4%, (8) -1%, (9) 1%, (10) 4%
This is just another example of WN5 significantly favouring the highest tiers.  Similarly, you've already agreed that tiers 4-6 are not getting a fair shake from a damage perspective and probably agree with tiers 1-3 too but you are not overly concerned.

View PostPraetor77, on Jan 16 2013 - 12:58, said:

I agree with Makaze 100% on his responses to Guerdon.

Then please move away from 'adjusting' each independent variable to try to deal with the baby seal clubbing issue and go back to a simple measure of skill actually demonstrated to gain an underlying score.  Then adjust that score using average tier played (and perhaps total number of games & / or average games per tank) etc. to come up with your adjusted WNx score which will fully eliminate the baby seal clubbing issue without penalising everyone in the lower tiers playing 'normally'.





Also tagged with statistics, stats, noobmeter, performance, efficiency, metric, compare, quality, rating, Garbad

2 user(s) are reading this topic

0 members, 2 guests, 0 anonymous users