Jump to content


[WN8] Feedback Required

WN8 stats

This topic has been archived. This means that you cannot reply to this topic.
164 replies to this topic

Poll: Vote (129 members have cast votes)

Which system is more useful?

  1. 0% of average stats = 0 (80 votes [62.02%] - View)

    Percentage of vote: 62.02%

  2. 30% of average stats = 0 (49 votes [37.98%] - View)

    Percentage of vote: 37.98%

Vote Hide poll

NeatoMan #101 Posted Sep 23 2013 - 23:33

    Major

  • Players
  • 17998 battles
  • 7,957
  • [5PANZ] 5PANZ
  • Member since:
    06-28-2011

View PostCrabEatOff, on Sep 23 2013 - 22:51, said:

Lets try another rephrasing-  
.
What should 0 score on WN8 mean?
.
Option A - this person does zero or near zero average damage per game, with similar numbers of kills, spot etc
.
Option B - this person does dpg below the threshold needed to differentiate their win-rate from an actual non-moving not-shooting bot - not zero dpg, but usually less than 1 average shot per game, similar for spots, etc
the problem is how you ascertained what a non-moving, zero damage bot win rate is.  It seems to be arbitrarily assigned and allegedly "backed" up with sound math.   It is not the appropriate way to find the zero point.
.
First of all you'd need an afk study to see what the zero win rate actually is.  You would also need to have a different  "zero contribution" win rate for different tiers.  High tier/premium tanks that end up in top spots more often are going to have lower AFK rates.  Secondly the negative platoon/TC effect is going to lower win rates for any given level of damage.
.
Take my E100, which is heavily weighted by quite a few unsuccessful clan war battles.  139 battles, 2263 avg damage, 42.4% win rate.  
.
This is going to give the appearance of higher than zero damage for your arbitrary "zero contribution" 40% win rate.  I'm sure zero contribution in an E100 is going to be much lower than a 40% win rate.

ConchitaWurst #102 Posted Sep 23 2013 - 23:37

    Major

  • Players
  • 11165 battles
  • 3,624
  • [EBOLA] EBOLA
  • Member since:
    04-09-2013

View PostCrabEatOff, on Sep 23 2013 - 23:21, said:

Something like this you mean? http://forum.worldof...-wn6-read-here/
Its something I usually do before we drop the rating (at least for WN6 and 7), but you make a point about doing it to keep people caught up when they drop into the thread. Partly its because WN* is a volunteer project, so we work in fits and starts. Two weeks ago I couldn't have answered a single question about WN8 because I hadn't read anything in months, and it took me a good 30-40 mins to catch up. I don't really consider those 30-40 mins to be an investment, but then again no one else on the internet reads.
Good idea, might not have time to implement before WN8 release, but should we see a WN9 project, I'll try to maintain an FAQ during development.  I'll add the questions you asked here into bjshnog's first post on WoTLabs, and if you come up with any more, I can post glib answers for them, without all the background, links, etc.
In the meantime, I am already working on the WN8 release post, to be translated into Russian and German, and hopefully posted on FTR as well.
Something even simpler, just the obvious questions somebody not involved with the process will ask, that tend to be one line questions and one line answers. People who just want to know how it will affect them and are not into the technical aspect of it. If you think it's worth compiling it, I can send some simple questions a random person will likely ask if you guys can't think of any.

RodneyDangerfield #103 Posted Sep 23 2013 - 23:40

    Major

  • Players
  • 51131 battles
  • 5,529
  • [RELIC] RELIC
  • Member since:
    04-29-2011
Although I like bots gettin an big 0 I would prefer if the final numbers stayed close to the same they are now.   When it really comes down to it if their rating is 0 or 200 it don't matter.... They are bad.

Crysiiz #104 Posted Sep 23 2013 - 23:42

    Staff sergeant

  • WGLNA Gold League Player
  • 11697 battles
  • 464
  • [VILIN] VILIN
  • Member since:
    04-25-2011
The problem with comparing server avg stats is that certain tanks, IE IS4 were moved around in tiering receiving certain buffs and nerfs, skewing the stats of the tanks in favor of one way or another, that being said, 0% player stats is getting my vote.

Pahech #105 Posted Sep 23 2013 - 23:52

    Major

  • Beta Testers
  • 15370 battles
  • 2,905
  • [OTTER] OTTER
  • Member since:
    09-28-2010
I too believe in contribution, however, minor, should warrant more than 0 stats.  It's not the job of WN rating to get rid of bots, it's WG's.

KilgorSoS #106 Posted Sep 23 2013 - 23:53

    Major

  • Players
  • 26602 battles
  • 11,792
  • Member since:
    07-23-2011

View PostCrysiiz, on Sep 23 2013 - 23:42, said:

The problem with comparing server avg stats is that certain tanks, IE IS4 were moved around in tiering receiving certain buffs and nerfs, skewing the stats of the tanks in favor of one way or another, that being said, 0% player stats is getting my vote.

Also, the T110E5 may have hundreds of T30 HT battles blended into the stats, I know mine does. My IS8 stats are 99% IS4 stats, etc. Tier4 Priest stats are now Tier3 priest stats, etc.

And the Biggest one of all...85% of my Bat stats are from the Bat when it was Tier9 with the 100mm gun! I would hope that these variables are being taken into account, and that is the kiss of death when trying to assign parameters to certain tanks.

RodneyDangerfield #107 Posted Sep 23 2013 - 23:58

    Major

  • Players
  • 51131 battles
  • 5,529
  • [RELIC] RELIC
  • Member since:
    04-29-2011

View PostKilgorSoS, on Sep 23 2013 - 23:53, said:

Also, the T110E5 may have hundreds of T30 HT battles blended into the stats, I know mine does. My IS8 stats are 99% IS4 stats, etc. Tier4 Priest stats are now Tier3 priest stats, etc.And the Biggest one of all...85% of my Bat stats are from the Bat when it was Tier9 with the 100mm gun! I would hope that these variables are being taken into account, and that is the kiss of death when trying to assign parameters to certain tanks.

If you really think about it, the same could be said for the first players to get tier 10s on the server.  With the old mm and no one to contest them.

Beerstein #108 Posted Sep 24 2013 - 00:11

    Major

  • Players
  • 15964 battles
  • 10,181
  • [PBKAC] PBKAC
  • Member since:
    08-18-2011

View PostPraetor77, on Sep 23 2013 - 22:30, said:

Welcome to my ignore list. You can spew all the lies you want, I am quite confident people with half a brain will understand you are spreading fallacies.
BTW, Beer, the baseline does not consider the bottom 30% to be worthless. In fact, less than 0.5% of all players in the game will have 0 WN8. And once again, WN8 is made for maximum utility for use in XVM. 1 shot every 40 games and 1 shot every 10 games are just as useless in determining the outcome of the game. Also, as I said, applying a baseline and turning WN8 into a ratio scale improves correlation to adjusted winrate, and makes a MUCH wider and closer to normal distribution of WN8, as compared to WN7.
At the moment, the number of players below 500 WN7 is so minuscule, that the value of that score, 500 WN7, becomes meaningless. With the ratio scale, all that happens is that the 500-2500 WN7 scale which encompasses 99.9% of WOT players, becomes a 0-4000 WN8 scale which encompasses 99.9% of WOT players. The top and bottom values are simply stretched, which ensures a more normal distribution of scores, and more accurately depicts player skill.
This will be my last post here, it is 100% clear there is no intelligent discussions to be had here. I wish to invite any intelligent people in this thread to join the official discussion if you are willing to take the time to read what is written, make sense of the data and the formulas posted, and only then voice your opinions and/or ask questions. Thank you all for reading.
Makes a lot of sense, there's no reason to expand the system a great deal outside the scope of reasonable use the number anywhere beyond a point is meaningless as far as overall scale goes as long as it's not too small like 1-10 since after a point everyones rating will vary to a degree and trying to pinpoint with higher numbers is silly.

Folterknecht #109 Posted Sep 24 2013 - 00:46

    First lieutenant

  • Beta Testers
  • 207 battles
  • 683
  • Member since:
    12-29-2010
remove_IDIOTS_from_forums

ConchitaWurst #110 Posted Sep 24 2013 - 00:49

    Major

  • Players
  • 11165 battles
  • 3,624
  • [EBOLA] EBOLA
  • Member since:
    04-09-2013

View PostFolterknecht, on Sep 24 2013 - 00:46, said:

remove_IDIOTS_from_forums

follow your own advice

Volacious #111 Posted Sep 24 2013 - 00:54

    Staff sergeant

  • Players
  • 20097 battles
  • 265
  • Member since:
    04-01-2012

View PostBigbxmach23, on Sep 22 2013 - 04:13, said:

The bigger the scale, the easier it is to create a more accurate picture of player skill.
To raise an extreme example, imagine a scale of 1-10 vs a scale of 1-1000, etc. It's like watching a movie at 240p vs 1080p, you can see ALOT more with the latter.
If the bigger scale isn't any more accurate, you don't gain anything.  You might even add confusion.
          
(like watching standard def content on a high-def tv if you want to stick with that analogy)
            
I actually like the 0-100 ratings you see in some of the garage and scoreboard UIs now - simple and efficient.
    
Also, I like what someone else mentioned above - a 2k games version should be part of the standard.  (60 day being somewhat unreliable due to variation in the rate of games played per day).  Making 2k the default would essentially eliminate the effect of re-rolled accounts.

Edited by Volacious, Sep 24 2013 - 00:58.


bjshnog #112 Posted Sep 24 2013 - 01:01

    Major

  • Players
  • 16719 battles
  • 2,055
  • [-IVY-] -IVY-
  • Member since:
    07-13-2011

View PostXMdead, on Sep 23 2013 - 13:48, said:

[/size]
A)  In 1 you did not understand.  {{xwn}}simply displays the value of WN (I think WN6 by default at this time).  And yes, there is the ability to change the value used to say what ranges are for what color.  But all of those are static ranges, even if configurable.  As players learn how to pad the stat there is an inevitable creep of players so that more move up into blue/purple... devaluing the meaning of those colors.  What I'm suggesting is to prevent the color devaluation altogether.  Define purple to be "top 5% of population", blue 6-15%, etc (or other %s) and let the borders between colors float so that the % of players in each category remains fixed.  As players get better they move up... and push the "no longer so good" players down.  XVM and stat sites can recalculate those borders to match the % at their convenience (whether that is daily, weekly or once a month).  But the key is for WN to publish what those %s are and not the XVM or stat site owners.

B) For WN8 (not WN8D) 2 and 4 seem to contradict each other and even the poll.  You say you can't calculate WN8 PER TIER because the API does not give the stats per tank, but the poll question implies that you do (since it talks of damage and WR per tank).  Can you explain the apparent inconsistency?  Or is the poll only for WN8D?  

C) Finally, if WN8D requires the dossier and dossiers aren't available to stats sites... what is its value?

A) xwn is a percentile value, the only difference is it's not dynamic. You could propose that idea to WoTLabs or Noobmeter or whatever, but it's not something that could be incorporated into the rating itself.

B/C) CptSkyhawk did a good job of clearing that up.

Praetor77 #113 Posted Sep 24 2013 - 02:30

    Captain

  • Players
  • 29723 battles
  • 1,695
  • Member since:
    05-01-2011
Tank rebalancing, metagame changes, etc. are the biggest noise factor for WN8. However, the sample is large enough so taht we should have a mix of playing dates for the tanks sampled, so this should partly be accounted for by the data itself (for example, for determining expected stats for Batchat, a percentage of those players will have played it when tier 9, and a percentage when only tier 10). Sadly, there is not much more we can do about it. Also, the noise is still much, MUCH lower than the noise in WN7 generated by considering every tank in the same tier to be 100% equal statswise.
Anyways, remember it is NOT your specific stats on the batchat which are compared to the "expected" stats.
Regarding the scale, the bigger scale IS more accurate. At least, it correlates better with adjusted winrates, which indicates WN8 with the ratio scale is better at measuring player skill than WN8 with a linear scale. Otherwise I never would have brought it up. I tried it and it worked wonders, so I proposed it for WN8.
To whoever said we are hunting bots, that is just wrong. I couldn´t care less. The idea is to establish what stats can be expected from a player who contributes nothing to his team. I used an estimated amount at first, 42%, I am now using 40% WR. What I did was use an evolutionary algorithm to select the winrate baseline at which the correlation of WN8(ratio scale) to rWINRATEc was the highest. It settled at an rWINRATE value which roughly corresponds to 40% winrate.
Telia, I am glad you abandoned the warpath and came to senses. Nobody on the WN crew minds people dissenting, as long as it is expressed politely, backing up your opinion with data, or at least actually taking a few hours to read through and understand what is going on before going on an insult/discredit rampage.

ConchitaWurst #114 Posted Sep 24 2013 - 02:37

    Major

  • Players
  • 11165 battles
  • 3,624
  • [EBOLA] EBOLA
  • Member since:
    04-09-2013

View PostPraetor77, on Sep 24 2013 - 02:30, said:

Tank rebalancing, metagame changes, etc. are the biggest noise factor for WN8. However, the sample is large enough so taht we should have a mix of playing dates for the tanks sampled, so this should partly be accounted for by the data itself (for example, for determining expected stats for Batchat, a percentage of those players will have played it when tier 9, and a percentage when only tier 10).

Anyways, remember it is NOT your specific stats on the batchat which are compared to the "expected" stats.


Lastly, the bigger scale IS more accurate. At least, it correlates better with adjusted winrates, which indicates WN8 with the ratio scale is better at measuring player skill than WN8 with a linear scale. Otherwise I never would have brought it up. I tried it and it worked wonders, so I proposed it for WN8.


Telia, I am glad you abandoned the warpath and came to senses. Nobody on the WN crew minds people dissenting, as long as it is expressed politely, backing up your opinion with data, or at least actually taking a few hours to read through and understand what is going on before going on an insult/discredit rampage.

PS: Neato, I am using 40% WR due to having reported around 15 bots myself, and there is a thread on wotlabs with known bots, and the average winrate for them was between 38 and 44%. Also, what I did was use an evolutionary algorithm to select the winrate baseline at which the correlation of WN8(ratio scale) to rWINRATEc was the highest. It settled at an rWINRATE value which roughly corresponds to 40% winrate.

It is Monday, I try not to put anything to heart especially in the mornings due to lingering presence of toxins from the weekend. Hopefully others don't either :)

CptSkyhawk #115 Posted Sep 24 2013 - 03:11

    Captain

  • Beta Testers
  • 10228 battles
  • 1,835
  • [TRUTH] TRUTH
  • Member since:
    01-06-2011

View Postremove_TELIA_from_game, on Sep 23 2013 - 23:13, said:

Apparently this is not the first time this has happened but you have to admit it is rather easy to read the thread and come to that conclusion. Why don't you make a sticky, like WN8 FAQ thread? And just list all the common questions in it, so people will have a resource from which to gather answers to questions that are known?

Will WN8 lower or raise my stats
- answer

Will WN8 lower score for certain tanks (penalize certain tiers and/or tanks)
- answer

Will WN8 count spotting damage
- Answer

Will wn8 account for landslide stacked teams (somebody just asked this a while back in this thread)
- answer

etc etc. And as information becomes ready, it can be added to the faq.

This way  when people read 18 pages+, half of which are people posting which tanks are OP, we can just ignore that and not keep going back to it splurging into the development thread further diluting the reliable information from the theories/ideas and propositions.

Drop me a pm or hit me on ts or something if you want, for any reason.

Will WN8 lower or raise my stats?
I'm going to go out on a limb and say your WN8 will actually be higher than your current WN7.  With the old formula from when they FIRST started working on comparing the expected numbers broken down by the amount of battles in each tank, you would be just under 1600.  With the new formula, well, I don't have it, but I am going to go out on a limb that 1600 will be a little higher.

Will WN8 lower score for certain tanks (penalize certain tiers and/or tanks)?
No, WN8 compares each tank to each other so if you are really good in a Hellcat, you will be able to play it and have a good rating.  If you are one of the top M3 Lee drivers, you can play it and have a good rating.

Will WN8 count spotting damage?
No.  WG does not release that data, and thus the developers of any third party rating system are at the mercy of WG.

Will WN8 account for landslide stacked teams (somebody just asked this a while back in this thread)?
No.  WG does not give any information on a per match basis.  Developers can only use numbers that are released by WG.

Praetor77 #116 Posted Sep 24 2013 - 03:31

    Captain

  • Players
  • 29723 battles
  • 1,695
  • Member since:
    05-01-2011

View PostCptSkyhawk, on Sep 24 2013 - 03:11, said:

Will WN8 lower or raise my stats?
I'm going to go out on a limb and say your WN8 will actually be higher than your current WN7.  With the old formula from when they FIRST started working on comparing the expected numbers broken down by the amount of battles in each tank, you would be just under 1600.  With the new formula, well, I don't have it, but I am going to go out on a limb that 1600 will be a little higher.

Will WN8 lower score for certain tanks (penalize certain tiers and/or tanks)?
No, WN8 compares each tank to each other so if you are really good in a Hellcat, you will be able to play it and have a good rating.  If you are one of the top M3 Lee drivers, you can play it and have a good rating.

Will WN8 count spotting damage?
No.  WG does not release that data, and thus the developers of any third party rating system are at the mercy of WG.

Will WN8 account for landslide stacked teams (somebody just asked this a while back in this thread)?
No.  WG does not give any information on a per match basis.  Developers can only use numbers that are released by WG.

Someone has been paying attention... :D  :great:

CptSkyhawk #117 Posted Sep 24 2013 - 03:43

    Captain

  • Beta Testers
  • 10228 battles
  • 1,835
  • [TRUTH] TRUTH
  • Member since:
    01-06-2011

View PostPraetor77, on Sep 24 2013 - 03:31, said:

Someone has been paying attention... :D  :great:

1690*(0.24*rFRAGS+0.71*rDAMAGE+0.16*(min(rSPOT,1.2))+0.08*(min(rDEFENSE,0.7)) +0.08*(min(rCAP,0.7)))


That is the formula I have from when the stuff first switched from the "WN8+Normalized" to the "WN8+Whateveritisnow"

Can you update me on the current formula, though?

Praetor77 #118 Posted Sep 24 2013 - 04:05

    Captain

  • Players
  • 29723 battles
  • 1,695
  • Member since:
    05-01-2011
We are still open for criticism and such. The idea is to create a table of comparable stats for each tank, using vbaddict´s dosier database. This table was built using the top 50% of players who play each tank (using a ratio to measure this, WN8ontank/WN8account), and using all that data to do a linear regression to a player of about 1200 WN8. That should make the stats for each tank nicely comparable amongst each other.
Once this table was finalized, I recalculated player WN8, then recalculated tank stats, etc. three times.
The resulting table is posted here:
http://forum.wotlabs...e-13#entry58083

Once that table is done, what we do to calculate a player´s WN8 is first calculate his "expected" stats. To do this, we multiply the number of games he has on each tank, by the "expected" stat for each tank on the Tank table. Once we do this for damage, frags, spots and defense, we then divide his total stats by the "expected" stats, this ratio results in the player´s rSTATS. The player´s rSTATS are then used to calculate rSTATSc, which contain a correction using a baseline for each stat.
To calculate rSTATSc, a baseline value is substracted from each rSTAT.

rFRAGc = rFRAG - 0.32
rDMGc = rDMG - 0.36
rSPOTc = rSPOT - 0.39
rDEFc = rDEF - 0.24
Finally, using rSTATSc for each player, the formula is this:
(0.25*rFRAGc+0.70*rDAMAGEc+0.10*MIN(1.0, rSPOTc)+0.10*MIN(1.6, rDEFc))*2000
or, distributing:
400*rFRAGc+1400*rDAMAGEc+200*rSPOTc (capped at 1.0)+200 * rDEFc (capped at 1.6)

CptSkyhawk #119 Posted Sep 24 2013 - 04:51

    Captain

  • Beta Testers
  • 10228 battles
  • 1,835
  • [TRUTH] TRUTH
  • Member since:
    01-06-2011

View PostPraetor77, on Sep 24 2013 - 04:05, said:

We are still open for criticism and such. The idea is to create a table of comparable stats for each tank, using vbaddict´s dosier database. This table was built using the top 50% of players who play each tank (using a ratio to measure this, WN8ontank/WN8account), and using all that data to do a linear regression to a player of about 1200 WN8. That should make the stats for each tank nicely comparable amongst each other.
Once this table was finalized, I recalculated player WN8, then recalculated tank stats, etc. three times.
The resulting table is posted here:
http://forum.wotlabs...e-13#entry58083

Once that table is done, what we do to calculate a player´s WN8 is first calculate his "expected" stats. To do this, we multiply the number of games he has on each tank, by the "expected" stat for each tank on the Tank table. Once we do this for damage, frags, spots and defense, we then divide his total stats by the "expected" stats, this ratio results in the player´s rSTATS. The player´s rSTATS are then used to calculate rSTATSc, which contain a correction using a baseline for each stat.
To calculate rSTATSc, a baseline value is substracted from each rSTAT.

rFRAGc = rFRAG - 0.32
rDMGc = rDMG - 0.36
rSPOTc = rSPOT - 0.39
rDEFc = rDEF - 0.24
Finally, using rSTATSc for each player, the formula is this:
(0.25*rFRAGc+0.70*rDAMAGEc+0.10*MIN(1.0, rSPOTc)+0.10*MIN(1.6, rDEFc))*2000
or, distributing:
400*rFRAGc+1400*rDAMAGEc+200*rSPOTc (capped at 1.0)+200 * rDEFc (capped at 1.6)

I'm doing something wrong, I'm sure.  Using the old formula, Telia would be around 1600, with the first formula, he would be 321, and the second he would be 1117?  Have I done the math wrong?

=0.25*($R$322-0.32)+0.7*($S$322-0.36)+0.1*MIN(1; ($T$322-0.39))+0.1*MIN(1.6; ($U$322-0.24))*2000
=400*($R$322-0.32)+1400*($S$322-0.36)+200*MIN(1; ($T$322-0.39))+200*MIN(1.6; ($U$322-0.24))

(R is Frags, S is Damage, T is Spots, and U is Defense)

ConchitaWurst #120 Posted Sep 24 2013 - 05:01

    Major

  • Players
  • 11165 battles
  • 3,624
  • [EBOLA] EBOLA
  • Member since:
    04-09-2013

View PostCptSkyhawk, on Sep 24 2013 - 04:51, said:

I'm doing something wrong, I'm sure.  Using the old formula, Telia would be around 1600, with the first formula, he would be 321, and the second he would be 1117?  Have I done the math wrong?

=0.25*($R$322-0.32)+0.7*($S$322-0.36)+0.1*MIN(1; ($T$322-0.39))+0.1*MIN(1.6; ($U$322-0.24))*2000
=400*($R$322-0.32)+1400*($S$322-0.36)+200*MIN(1; ($T$322-0.39))+200*MIN(1.6; ($U$322-0.24))

(R is Frags, S is Damage, T is Spots, and U is Defense)

under the current (wn6/7) overall is 1401 60 day 1702 and 30 day 2050 give or take...

are you using the numbers for my total, 60 or 30?