WN8: Summary

From WN Efficiency Wiki
Revision as of 10:41, 11 December 2013 by Phalynx (Talk | contribs)

Jump to: navigation, search

Intro

WN8 is the latest iteration of the WN8* (WNx) project lead by Praetor77.

WN8 seeks to measure the observable contribution to matches, across an account, and hopefully infers some information about the latent variable “skill”. WN8 should not be considered the end-all-be-all of skill evaluation. It is intended to be looked at next to win-rate, battle count, average tier, and will never replace the inspection of the Service Record or platooning with someone or watching their replays. This is because WN8 cannot capture things like timing, target priority, teamwork or decision making abilities.

Main differences with WN7

  • WN8 has much more weight on damage than on kills, which was reversed in WN7. What allowed us to put more weight on damage was that WN8 uses per-tank normalization instead of the average tier-based normalization WN7 used.
  • Per-tank normalization in WN8 now allows a fair comparison of players playing only highly effective tanks for their tier (think Wft100, Foch155, Hellcat, T49, T18, T29, the old Su-26, etc.) versus players who have played many lines, with many tanks.
  • Due to per-tank normalization, light tanks and SPGs are now more fairly compared to other tank types in WN8 (they received lower WN7 than other tank types).
  • No tier-based penalty is introduced. Low tier tanks are measured fairly against high tiers, whereas in WN7, you could raise your score artificially by playing a mix of low and high tiers, and were penalized if you only played low tiers.
  • Scale has changed. See below.


Per-tank stats explanation

Briefly, what we have done is used vBAddict database (more than 1 million tanks played by about 20000 players) to calculate balanced and comparable “expected” values for each tank. This table is available here: http://www.wnefficiency.net/wnexpected/

The number of games you have played on each tank allows us to estimate what your “expected” total stats should be for your entire account. Then your actual total stats are compared to the expected stats as explained below.

Interval/Ratio scale explanation

WN8 uses a different scale from WN7. WN7 values ranged from 400-2800, and we turned this scale into a ratio scale, which brings about numerous advantages. Debatably the most important one after improved accuracy in measurement is that we can now say that a 2400 WN8 player on average contributes twice as much for his team to win than a 1200 WN8 player. This was not true for any rating before WN8.


Brief explanation of the formula and how it was devised (Eureqa)

Using a subset of the NoobMeter.com database (115k players), we used their stats as input to a program called Eureqa which uses genetic algorithms to find the best “formula” to explain winrate using the rest of the stats. So, Eureqa assigns weights to each stat according to how important it is in determining wins. The final formula is:

WN8 = 980*rDAMAGEc + 210*rDAMAGEc*rFRAGc + 155*rFRAGc*rSPOTc + 75*rDEFc*rFRAGc + 145*MIN(1.8,rWINc)

Note: every rSTATc is the corrected ratio of STAT/expectedSTAT, and they are all capped to prevent, for example, farming spots on tanks with low expected value of spots, like arty.


Details

Interested in more details? Read the full document about WN8