Jump to content
Everpest

Help me to understand Win8 math...

Recommended Posts

First, let's start with the basics.  I like the general concept behind Win8 - trying to index your performance (per tank) to the average performance (per tank) of the entire gaming population.  Granted there are some issues related to non-quantitative indicators of player skill, like scouting, spotting, etc, but I like where Win8 is headed.

 

The problem I am having with the math is that Win8 is a table-driven system, and the table (version14) does not appear to me to be balanced - giving skewed player results and Win8 scores.

 

Win8 indexes individual performance for any one tank to the gaming population as a whole.  So, using basic math, if you averaged all player's Win8 scores for any one tank, the average should come back to 1000.  YOUR score might be higher or lower than 1000 for that tank... but the total score (of all players) SHOULD BE 1000.

 

This is NOT the case.  Some tanks have much higher than 1000 Win8 scores.  Some have much lower.  If you play THESE TANKS, your Win8 scores will be inflated, on average.  If you play the lower score tanks, your Win8 scores will be deflated, on average.

 

As an example,  the VK 72.01 (K) currently has a Win8 score of 2,177.61.  This is for ALL PLAYERS over the last 30 days - total of 47k battles.  How is this possible?  Shouldn't the score be 1000 - since some players should be above average, and some should be below average?  What this is suggesting is that if you roll one of these tanks, you can expect to get a Win8 score 2x what is expected.

 

Likewise, on the other end of the scale, the Renault FT has a Win8 score of 175 (last 30 days played, 19.5 k battles).  Shouldn't these ALSO have a score of 1000 - so that you might be better or worse than average, but ALL players totalled come back to the mean.  (In this case the data might be skewed somewhat since this is a noob tank, and the Win8 tables are based only on players with 1k + battles).

 

There are a lot of other inconsistencies with the Win8 data table.  Looking at Win Rate data, the average win rate for all tanks in the table is 53.4%.  This is FAR above the average win rate for even experienced players.  On the high end of win rates is the Cruiser Mk. III - with a win rate of 64.03%.  This suggests that if you aren't getting a 64% win rate with your Cruiser in pubs, you are a below average player.  Based on actual game results, the 30 day win rate for Cruiser Mk III's is actually the 3rd lowest for ALL T2 tanks - 42.5% (109k battles).  On the other side of the Win8 table we have the STB-1, arguably the best T10 medium in the game.  On the Win8 table it has a win rate of 47.3%.  Yet looking at recent actual battle results, the STB-1 has a 30 day win rate of 54.0% (2nd highest T10 win rate in the game).

 

Since actual and table win rates are so out of wack - you would expect Win8 score to be out of wack as well... and they are.

 

The Win8 score for an STB-1 is 2044.6 (versus an expected 1000).

 

The Win8 score for a Cruiser Mark III is 289 (versus an expected 1000) (again low tier tanks may be skewed because of new players - but not THIS much).

 

Help me with my math if I am wrong somewhere...

 

 

 

 

 

Link to post
Share on other sites

This is mostly because of the data set used to draw the expected values. The data set used is indeed above average. Link to the wiki here. Reading between the lines, it means that only half decent players are included. There is no value in measuring against a deep red. At the same time, the average win rate is 53% might be partially because of the number of games in each tank. There are more games in the tanks with lower win rates than tanks with higher win rates perhaps? IS-7 for instance has a lot of games. Few claim that it is the epitome of OPness.

 

Cruiser III data is skewed by it's immense seal clubbing ability with that deadly 40mm. A half decent player can stomp in this tank, and those who prefer this tank over the T18 tend to be a little more capable.

The same can be said for STB-1. The first to get the tank will be those who splurge gold for free xp (and may or may not know how to play the game) and no one will really know how to make the tank perform at its best for a little while, then unica will get their hands on the tank, and players will start accommodating themselves to the tank. Remember also that good crews don't just appear and take a little grinding. No skills on a tier 10 medium is a huge disadvantage. In general, a tank if left alone would should a downward trend in WR after introduction as successively less skilled players unlock it, but the initial tick will be lower, and WN8 was released right after the Japanese Tanks patch was.

That said, enough with the explanations. The above link provides more information if you so desire. And I do wish they would eventually update the database to fix the STB-1 thing as well.

Link to post
Share on other sites

I am no expert on wn8, so correct me if I am wrong, but IIRC expected is not the same as average.

 

Edit:  Thanks for confirmation

Edited by KnobbyHobbGoblin
Link to post
Share on other sites

If you meet the expected values for WN8 in any given vehicle, your rating is 1565.  Most players do not.

 

For limited offer tanks like the VK72.01, WN8 is way off.  The expected values are way below what the top 50% of players using that vehicle are doing and that's because it is a new vehicle and there's not a lot of data out on it as yet.  Couple that with the fact that the only players that get that tank happen to be very good (in general) and you can see why it's average WN8 is so high.

 

A few other vehicles (and I've pointed this out before) have expected values less than their server averages -- especially in the higher tiers.  This will no doubt be adjusted in future iterations of the WNx system, but for now, I'd recommend playing the Object 416 or E-50 ... a lot :verysmug:

Link to post
Share on other sites

Excellent.  I didn't know about the 1565 - which helps explain a lot.

 

There are still issues on the tank by tank level, and I'm going to crunch some numbers on the expected versus actual results.  Sounds like there are people way ahead of me on all of this, but I'll do it for my own self-enlightenment.

Link to post
Share on other sites

Wait. Brain now hurts.

 

 

 

expected values for WN8 in any given vehicle, your rating is 1565

 

Why is this value "expected" if according to the scale 1565 is "Good"?

 

Shouldn't the average be the expected?

 

And in that light, what does "average" WN8 (like me) actually mean about a players relative skill to other players?

Link to post
Share on other sites

I'm sure here "expected damage" means "damage expected from one who plays at 1565 WN8", and it's just a base point.

Link to post
Share on other sites

I'm sure here "expected damage" means "damage expected from one who plays at 1565 WN8", and it's just a base point.

 

I went and read up a bit on it. I was making a couple of incorrect assumptions.

 

The expected value is (loosely and with other parameters) based on a scale that doesn't include the average. It includes the median AND ABOVE for per tank performance as the "baseline" so if I'm an average player, I should (and do) play below the expected value since it begins at the median.

 

The rest is pretty complicated - in fact I'm not sure I truly understand it - but now it makes a lot more sense.

 

I'm ~1,100 WN8 player overall. I'm "average" in "most" of my tanks but have not moved to or above the median.

 

And as a funny (horrible) I'm worth approximately 1/3 of a Garbad on any team, in pretty much any tank I've played vs. him on any team in any tank he's played.

 

Garbad is quite literally a platoon of 1. Which explains a lot.

 

Very cool system.

Link to post
Share on other sites

I guess if you look at it that way, unicum platoons are platoons of 9. The idea of that is somehow very funny.

 

At any rate, WN8 grossly under-represents Garbad's abilities. It's only good up to a point.

Link to post
Share on other sites

I guess if you look at it that way, unicum platoons are platoons of 9. The idea of that is somehow very funny.

 

At any rate, WN8 grossly under-represents Garbad's abilities. It's only good up to a point.

 

Absolutely agreed. He's worth at least 5 of me any day, ASSUMING that all the other players on his side are my level or worse. But again, only up to a point. Garbad + 14 reds vs. 15 of me... not going to go well for Garbad 90% of the time because there is a definite scale. If I'm worth 2 oranges, or 3 reds... well, you get the idea.

 

It's not perfect, but it definitely gives me a better way to look at team compositions, particularly if there's a super-purple WN8 guy on my team. Normally I just try to stay the fuck out of their way. Having this bit of information about relative skill means that I might do more for my teams (hypothetically) by spotting for him and taking shots for him even at the cost of my own tank... so long as I can do that by staying the fuck out of his way.

 

Overall, a very enlightening way of looking at stats that gives me a simple (if not 100% accurate) picture of how to behave in games that feature blue+ players on either side.

Link to post
Share on other sites

Sorry for unburying this topic but it seems to be the best place to ask my question.

The WN8 formula - as described here: http://wiki.wnefficiency.net/pages/WN8

is perfect when I'm trying to calculate WN8 value for a single tank.

I got the average stats - I can get the expected stats from the table, join them together and calculate WN8 for a tank.

But how (and at which stage) should I combine values for multiple tanks - in order to calculate player's WN8?

Should it be weighted average as in example (lets assume I've only driven two tanks so far) :

100 battles on tank A with calculated WN8 = 1000

300 battles on tank B with calculated WN8 = 1400

Player's WN8 = (100*1000 + 300*1400 / (100+300)) = 1300

Link to post
Share on other sites

I believe it should be a weighted average of the expected values of all your tanks going into the formula with your overall stats (DPG, KPG, etc), so it can evaluate your overall performance by what tanks you drive. I might be wrong, but I'm pretty sure this is how it works.

Link to post
Share on other sites

Ok either my mods are all screwed up.... or I am.

 

I play a session with two games.  I get two WN8 scores - one 1000 and one 2000.

 

The WN8 for my session should be 1500, correct?

 

I know this sounds stupid, but I am getting lots of weird results with mods "supposedly" tracking WN8 but giving very strange results.

Link to post
Share on other sites

I believe it should be a weighted average of the expected values of all your tanks going into the formula with your overall stats (DPG, KPG, etc), so it can evaluate your overall performance by what tanks you drive. I might be wrong, but I'm pretty sure this is how it works.

 

This is the problem I am having atm.  It should NOT be a weighted average.  It should simply be an average of all WN8 scores - because WN8 ALREADY takes into account tank by tank variances.

 

So if I play a T8 and get a WN8 of 1600, it is the same as playing a T10 and getting a WN8 of 1600.  The WN8 calcs for each of those tanks is based on expected damage, kills, etc, BY TANK - so the scores are identical.

 

If I play one game and get a WN8 of 1000, and play a second game and get a WN8 of 2000, my WN8 for that session SHOULD BE 1500.  It should not matter whether I play different tanks, different tiers, etc.  The tank balancing takes place WITHIN WN8 calcs.  Once you have your WN8 you are comparing apples to apples - or at least that is the entire justification for why WN8 is set up the way that it is.

 

If this were NOT the case, there would be huge bias in total WN8 scores towards high tier tanks and tanks with huge expected damage output.  You could be the best player in the world with T5's - with a T5 WN8 of 5000, for example, but if you were only a 1000 WN8 player in T10s, your total WN8 could be like 1100 (assuming you had the same number of battles in each tier).  This is clearly WRONG - or at least is in my opinion.

Link to post
Share on other sites

Well, I just found the following little item buried way at the bottom of a FAQ page on the WN Efficiency Wiki. If this is true, it basically throws WN8 out the window, in my opinion.  If it is using aggregate stats (instead of per-tank stats) each time you play one game it is incrementing your total stats by the expected value of that single tank - so a T10 tank battle will get 20x the weight of a T1 battle (because it is incrementing your expected damage by around 3k for a T10 battle, versus around 150 for a T1 battle).  You can have 5 amazing T5 battles in a row, and if you have 1 bad T10 battle, all that work will be thrown out the window. Can someone confirm this?

 

Is WN8 calculated on a per-tank basis?

No, WN8 is calculated based on the only data available through the API, which is games played in each tank, and aggregate stats for damage, kills, spots and defense. However, from a 20000 player vBAddict database, the average error introduced to WN8 by using aggregate data instead of averaging per-tank WN8 (weighted by battles on each tank) is 2.15% of the total WN8 score, while 95% of players have an error under 6.3% of their total WN8 and 99% of players have errors under 9.2% of their total WN8.

Link to post
Share on other sites

Well... I am getting so frustrated I am about to give up on trying to figure out WN8.

 

I just started this morning with a blank slate.  Played two games with the SAME TANK.  First game WN8 1000.  Second game WN8 1800.  Average WN8 for session... 1200.  WTF?

 

I understand the theory of WN8 and agree with it.  However the implementation seems completely arbitrary and FUBAR at the moment.

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.


  • Recently Browsing   0 members

    No registered users viewing this page.

×
×
  • Create New...