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

Guerdon #1061 Posted Jan 18 2013 - 00:37

    Corporal

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

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

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

Only if they're actively targeting the highest tiers instead of just playing the game.

As I've already mentioned, those who want to "get on" run premium accounts and tier 8 premium tanks.  Likewise, they progress to Tiers 5-8 as quickly as possible to maximise their credits and the higher tiers generally to maximise free XP, as well as to get into Clan Wars for free gold.

Conversely, there are those who simply play the game, playing each tank before moving up.  Without premium accounts and free XPing your way up you get a big penalty for re-training across classes.  It therefore makes some sense to play say all the lights of a nation before moving on to the mediums & heavies.  For example, with the Russian lights one may well play the BT-2, BT-7 & A-20, as well as the T-26, T-46, T-50 & T-50-2.  If running these two lines, the A-20 crew will open the mediums with the Tier 5 T-34 and the T-46 would open them up with the Tier 4 T-28 which then opens the heavies with the Tier 5 KV-1.  Consequently, it's not unreasonable to have one light crew go all the way to the T-50-2 with the other being retrained to the mediums before getting a third to start the heavies.  All this simply means that one will have many games in the low tiers compared to someone trying to "get on" to maximise credits, free XP and get to the top tiers and say clan wars for free gold etc.

By way of example, I know of two players who have each adopted one of these paths:
Tier 1 = 5 tanks, 61 battles, average 12.2 // 5 tanks, 111 battles, average 22.20
Tier 2 = 12 tanks, 198 battles, average 16.5 // 7 tanks, 237 battles, average 33.86
Tier 3 = 13 tanks, 663 battles, average 48.69 // 8 tanks, 213 battles, average 26.63
Tier 4 = 8 tanks, 449 battles, average 56.13 // 5 tanks, 201 battles, average 40.20

So, according to you, the "troll" is the one who has actually has many less games and a lower average in tiers 1 & 2 compared to the premium account player who was in a rush to "get on" and get to the highest levels quickly because that's where the money's made.

Similarly, in tiers 3 & 4, whilst the first player has more battles it's largely because he's played many more tanks.  Furthermore, if you allow for the second player having his premium account and thus getting 1.5 times XP and so can progress much quicker, the first player has effectively averaged slightly less in tier 4 and only slightly more in tier 3 and yet the first player wouldn't have been using 100% gold retrained crews, complex equipment, premium consumables, gold ammo or free XP (especially all that from the Tier 8 premium tanks) etc. either.

Basically, the first player has progressed through tiers 1 & 2 quicker and after allowing for not having a premium account etc. has effectively progressed through tiers 3 & 4 quicker too.

In short, the first player has progressed as fast as he can but has taken a different path and yet you see such a player as a troll.  It's the latter that's actually the relative troll having played nearly twice as many tier 1 & 2 battles, many of which will have been with 100% trained crews with additional skills & perks, as well as complex equipment & premium consumables etc.

This is another example of why you shouldn't simply assume that people with a low average tier are baby seal clubbing trolls.  Most of those are the ones who have already progressed and go back with a relatively unfair advantage!

NeatoMan #1062 Posted Jan 18 2013 - 01:38

    Major

  • Players
  • 28180 battles
  • 20,423
  • Member since:
    06-28-2011

View Postdavec79, on Jan 17 2013 - 22:53, said:

Omg, strolling through all these posts make one thing very apparent to me.

You guys have absolutely no life whatsoever.
and yet you are sitting here reading it....

DessieDoo #1063 Posted Jan 18 2013 - 09:17

    Sergeant

  • Players
  • 9562 battles
  • 206
  • Member since:
    07-15-2012
Looks like a good formula, but as a light tank, I always worry I'll be labeled a scrub because I don't often do much damage.

Boom_Box #1064 Posted Jan 18 2013 - 09:39

    First lieutenant

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

View PostDessieDoo, on Jan 18 2013 - 09:17, said:

Looks like a good formula, but as a light tank, I always worry I'll be labeled a scrub because I don't often do much damage.

You are in a tough job which is poorly rated. Eff gives you more credit for spotting and cap. NM factors something in to put you on par with Eff, and WN pretty much fails you as a real metric.

It is apparent by looking at your profile that your most played tanks serve you well and your win rate is very likely a result of your contribution as a scout. A lot of people would see that if they looked past a single number metric. Since scouting is a fundamental function of the game it would be nice if WG gave us the detailed data that would correct this deficiency.

NeatoMan #1065 Posted Jan 18 2013 - 16:12

    Major

  • Players
  • 28180 battles
  • 20,423
  • Member since:
    06-28-2011
More seal clubbing work

I estimated a conservative average tier progression based on my own early advancement without premium status.
50 games - 1.5 avg tier
100 games - 2.0
500 games - 3.0
1000 games - 4.0
2000 games - 5.0

From that I derived the following seal club correction curves, capped at 2000 games
correction factor diagram

I have it set up so that if you maintain or are above the average tier progression listed above you will have minimal to no correction to your efficiency scores.  If you are below that tier you will face some penalty depending on how far below those target values you are.  

Applying the correction factors from the above graph to the win rate S-curves yields the following:
1000 games
1500 games
2000 games


edit - another idea, along the lines that Guerdon suggested, is rather than applying the correction factor to the s-curves, we could apply it directly to their overall efficiency scores instead.  We'd probably have to make the correction less drastic by having it bottom out at 40-50% rather than 10% as I did above.  The more i think about it, I think this last suggestion might work a lot better.

Edited by NeatoMan, Jan 18 2013 - 20:33.


Solinvictus #1066 Posted Jan 18 2013 - 22:44

    Major

  • Players
  • 29863 battles
  • 3,351
  • Member since:
    03-20-2012
Any way to use this in xvm instead of the normal eff?

IceNein #1067 Posted Jan 18 2013 - 22:57

    Sergeant

  • Players
  • 11170 battles
  • 113
  • Member since:
    04-03-2011
Every time I see this thread I have to laugh.  "Actually statistically sound efficiency formula."  Hm.  I know!  If we go through a ton of effort to make your efficiency approach your win rate that'll be real useful.  If only there were a stat that was indicative of how often you win, and how well you contribute to your team.  I guess we'll have to get the real men of genius out of retirement to solve this one.

Praetor77 #1068 Posted Jan 20 2013 - 00:25

    Captain

  • Players
  • 29723 battles
  • 1,675
  • Member since:
    05-01-2011
Not yet Solinvictus, we are waiting to hear back from XVM developers...

Vandelay #1069 Posted Jan 20 2013 - 00:34

    First lieutenant

  • Players
  • 11324 battles
  • 756
  • Member since:
    05-08-2011

View PostIceNein, on Jan 18 2013 - 22:57, said:

Every time I see this thread I have to laugh.  "Actually statistically sound efficiency formula."  Hm.  I know!  If we go through a ton of effort to make your efficiency approach your win rate that'll be real useful.  If only there were a stat that was indicative of how often you win, and how well you contribute to your team.  I guess we'll have to get the real men of genius out of retirement to solve this one.

If you had bothered to read through the thread, you'd know why correlating to win rate is a good idea.

Guerdon #1070 Posted Jan 20 2013 - 03:11

    Corporal

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

View PostFolterknecht, on Jan 18 2013 - 05:42, said:

@Guerdon: It usually makes no sense to drag your crews along from t1-5.

IMO, it makes no difference whether you move your crew or get a new one in tiers 1-4 because you'll rarely reach sufficient training to mean that re-training with silver would get you over 75% (i.e. you need to get your crew to over 83%).  That said, if you were free re-training, you'd get a benefit in almost every case in tier two or above (i.e the crew only needs to be 56% or more).

Anyway, It's largely irrelevant for the point that I was making, which was that one player may choose to play say all the lights of a nation (or even a variety of low tier lines) before moving on to the mediums and heavies (or higher tiers in general) rather than just rushing to the higher tiers for more damage, XP and mainly credits etc.

There is nothing fundamentally wrong with this approach and is not baby seal clubbing because of still having relatively little experience, as well as mainly running stock tanks, with relatively untrained crews and without complex equipment, premium consumables or premium ammo etc.

Such play should not be penalised for baby seal clubbing just because it has a relatively low average tier (for now, which will increase over time as more tanks are 'completed', each of which will generally be at a higher tier).

View PostNeatoMan, on Jan 18 2013 - 16:12, said:

More seal clubbing work - I estimated a conservative average tier progression based on my own early advancement without premium status.
50 games - 1.5 avg tier
100 games - 2.0
500 games - 3.0
1000 games - 4.0
2000 games - 5.0

Due to the above, I see these as too low.  I'd guess you should increase by at least 50% and probably 100%.  Yes, it can be done at those rates but it doesn't have to be.  Baby seal clubbers will have literally hundreds of games in the low tiers and for those high tier stat padders an extra 50 tier one or 100 tier two games etc. would make practically no difference to their overall rating.  Another way would be by average battles (i.e. games / tanks).

Edited by Guerdon, Jan 20 2013 - 03:13.


NeatoMan #1071 Posted Jan 20 2013 - 03:32

    Major

  • Players
  • 28180 battles
  • 20,423
  • Member since:
    06-28-2011

View PostGuerdon, on Jan 20 2013 - 03:11, said:

Another way would be by average battles (i.e. games / tanks).
That would be good, though i don't have access to that info

IceNein #1072 Posted Jan 20 2013 - 03:54

    Sergeant

  • Players
  • 11170 battles
  • 113
  • Member since:
    04-03-2011

View PostVandelay, on Jan 20 2013 - 00:34, said:

If you had bothered to read through the thread, you'd know why correlating to win rate is a good idea.

I have read through this thread.  There is a statistic that always correlates to win rate.  That statistic is win rate.  The only other thing you need is the common sense that when you look at somebody's stats and they have 2k games played with a T18, their stats might be inflated.

There is an actual statistically sound efficiency formula.  I refer to it as win rate.

Vandelay #1073 Posted Jan 20 2013 - 04:25

    First lieutenant

  • Players
  • 11324 battles
  • 756
  • Member since:
    05-08-2011

View PostIceNein, on Jan 20 2013 - 03:54, said:

I have read through this thread.  There is a statistic that always correlates to win rate.  That statistic is win rate.  The only other thing you need is the common sense that when you look at somebody's stats and they have 2k games played with a T18, their stats might be inflated.

There is an actual statistically sound efficiency formula.  I refer to it as win rate.

Sigh.

The whole point of this is that you can pad win rate, but you can't pad other stats like damage and kills (unless you play really low tiers, which is what we're working on right now).

If you correlate to win rate, you'll have most of the population correct, but because the damage and kills of win rate padders won't correspond to their win rate, their WN will be lower than what it should be. This shows that they are a padder.

Vulcans_Hammer #1074 Posted Jan 20 2013 - 05:39

    First lieutenant

  • Players
  • 26964 battles
  • 995
  • Member since:
    07-03-2011
I suppose I'm a baby seal clubber,or a stat padder? with almost half my 10,000 games played in the T-50

maokai #1075 Posted Jan 20 2013 - 07:52

    Staff sergeant

  • Players
  • 23943 battles
  • 352
  • [ESC] ESC
  • Member since:
    06-15-2011

View Post99VulcansHammers, on Jan 20 2013 - 05:39, said:

I suppose I'm a baby seal clubber,or a stat padder? with almost half my 10,000 games played in the T-50

Since Wargaming doesn't report damage upon spotting, it makes it really difficult to evaluate tankers who primarily drive scout tanks.

Otter_von_Bismarck #1076 Posted Jan 21 2013 - 09:47

    Captain

  • Players
  • 21391 battles
  • 1,455
  • [OTTER] OTTER
  • Member since:
    02-18-2011
Whats the status on WN5? Has the script for the WoT Player Profile Page been updated to include WN5 ratings? Will Wotlabs include it soon?

Boom_Box #1077 Posted Jan 21 2013 - 10:01

    First lieutenant

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

View Post_Von_, on Jan 21 2013 - 09:47, said:

Whats the status on WN5? Has the script for the WoT Player Profile Page been updated to include WN5 ratings? Will Wotlabs include it soon?

Yes, my greasemonkey script currently shows WN5 as defined by Praetor77, where:
KILLS *(1240-1040/(TIER^0.164)-(TIER^4.7)*0.001)+DAMAGE*500/(184*EXP(0.24*TIER)+130)+SPOT*125+MIN(DEF,2.2)*100
+((185/(0.17+EXP((WINRATE-35)*-0.134)))-500)*0.45

I'm tempted to turn on all of the stats so everyone can see the scoring breakdown of each category. At the moment only a few people, most of whom are contributors in this thread, can see the full script output.

Neverwish #1078 Posted Jan 21 2013 - 12:11

    Captain

  • Players
  • 8631 battles
  • 1,980
  • [LABS] LABS
  • Member since:
    06-10-2012
Praetor, is WN5 ready to go? Can I go ahead and implement it in WoTLabs?

Praetor77 #1079 Posted Jan 21 2013 - 15:34

    Captain

  • Players
  • 29723 battles
  • 1,675
  • Member since:
    05-01-2011
Neverwish, let me have one more day to test something.

I think the drop in score for non-sealclubbers in WN5 score is small enough. I have been thinking about the winrate solution proposed by Neatoman, and though it could be improved, like someone said above, if we take out the damage and kill disadvantage for tiers 1-3 and only use the winrate bashing, a player with 1500 WN and avg tier 8 can go into tier 2, play a couple hundred games and bump his WN enormously.

Also, like I said before, I would rather have a fixed winrate portion penalty for seal clubbers according to avg tier played, and not factor in games played. Maybe something can be worked out for WN6. For winrate 70%, I would like something like -200 points for avg. tier 1, up to 0 points for avg tier 5, and 200 points for tier 6+. I will see what I can do.


Boom_box, your latest script (0.7.5.54) is showing the normal profile page without any additions, cant see WN or anything.... weird.... This was on Chrome. On Firefox with Greasemonkey, it works perfectly.

Edited by Praetor77, Jan 21 2013 - 16:53.


NeatoMan #1080 Posted Jan 21 2013 - 16:49

    Major

  • Players
  • 28180 battles
  • 20,423
  • Member since:
    06-28-2011

View PostPraetor77, on Jan 21 2013 - 15:34, said:

if we take out the damage and kill disadvantage for tiers 1-3 and only use the winrate bashing, a player with 1500 WN and avg tier 8 can go into tier 2, play a couple hundred games and bump his WN enormously.
I'm not so sure it will be boosted as much as you think.  Maybe i'll set up some scenarios to check just how much.

Here is what I have as far as the win rate fix.  I think if you simply want to adjust for tiers 1-3 and not factor in games played then what i got is probably way too complicated.

Using WN4 data, I fabricated a 2000 pt WN4 player, and then checked how his efficiency would drop with average tier and games played.
http://home.comcast....es/WN4fixXP.jpg
black line = player with a 2000 unadjusted WN4
purple = same 2000 player who ends up at average tier 6 after 3000 games, using the win rate adjustment
blue = average tier 5 after 3k games
green = average tier 4
lt green = average tier 3
yellow = average tier 2
pink = average tier 1

They all start at around 80% of WR after 100 games.  Since that's such a small number of games I think lowering WR contribution for everybody at low games is not a bad idea.  By the time you reach 1000 games you are near full win rate contribution, which then gets corrected depending on what average tier you attained.

Here is how it affected various seal clubbers in the large WN5 data set, based on how they would look if they had the same stats at 500, 1000, and 3000 games
http://home.comcast....WN4fixStats.jpg

since all the players are above 2000 games, i'm not so sure how concerned you are about how much gets fixed below 2k games, which is where this formula adjusts most dynamically.  After 2k or 3k games a simpler formula can be designed that ignores the # of games played.

now the fun part.

the original WR portion:   WReff = [(185 / ( 0.17 + exp((WR - 35) * -0.134))) - 500] * 0.45

replace 185 with the following: 185*( 0.2 / (0.2 + EXP{[ tier + (1/tier^2) - (games/(100*tier^2)) ] * -1.2} ))
formula is capped at 3000 games





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

1 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users