Jump to content


Verified Tanker [NA]
  • Content Count

  • Joined

  • Last visited

  • Days Won


Everything posted by Praetor77

  1. Thanks for the suggestions everybody. I tried a variance approach but the results were underwhelming. I really didn´t like what came out of that, so I stuck to the two point system. However, of course anyone is free to experiment with such a solution and mybe come up with something better. Also, the normalized per-stat based as bezz suggests would defeat the purpouse of using WN9 in XVM which is the main goal.
  2. Anyone who would like to pick up the torch is more than welcome to, of course. I will be around these forums helping out when I can.
  3. The brand is WN. the number is the version. I am undecided as to what se should call this new version. If we use WN8 however, confusion could arise since a lot of things changed, albeit much less than from WN7 to WN8 for example. Also, as I said, this will probably be the final version, as my time to dedicate to this will quickly drop to 0 in a few months, and there are no serious improvements we can make without some SERIOUS man hours, and which data suggest will result in very mild, IMHO nonsignificant, improvement. So as of this new version (Wn8b, WN9, WNFinal, WN9Final as some peop
  4. Easier solution? Remove credit rewards for doing nothing/near nothing. Other people have suggested something similar before. The problem is "team rewards", where people get a piece of XP/creds from the team damage. If you average 1000 damage in a tier 10 you should be losing A LOT of credits. Currently, doing 1000 damage with a prem account and not shooting prem rounds will barely result in a credit loss. This needs to change. In turn, this coupled with a little more expensive prem shells should put the meta back on track, IMHO. I agree with Rexxie. 183 is at least an 8 in my boo
  5. I only calculated rWINc to be able to use it for input into Eureqa and for calculating expected stats with the new method. It won´t be included in the formula. All the troubles I took to make the new rWINc is so that Eureqa has a better metric to correlate other stats to. No. The two linear formulas (0.1-1 expected stat, and 1-inf expected stat) join at 1 rSTAT. So the first formula is used for players with rSTATs up to 1 (exp1 stats), then the other formula is used. Welome back Crab! cWR works like a charm. Ill give you the formula later, it adjusts winrate according to tanks p
  6. What flint said, plus the mantlet has a shell magnet.
  7. Yes, calculation is 100% automatic by now. I could do monthly updates without much of a hassle. R is awesome. It would be possible to make a neural net or svm machine, but symbolic regression (Eureqa) has big advantages. It can also be considered "machine learning", and I already know how to use it, so...
  8. Yeah, I am doing the simulations and the WN9 stuff in a parallel manner. The simulations are dang interesting. Will trickle in results as I get them. Probably start a new thread on that. The simulation is giving me a lot of info that allows me to interpret and analyze the WN9 data better, and gives me ideas, like the rWINc stuff, which worked like a charm. Also, despite the changes, WN9 will be pretty similar to WN8. Major changes: Using WR to predict expected stats through linear regression instead of tankWN8/accountWN8 as per Gryphon´s suggestion. Two expected values instead
  9. IMHO, it´s not worth the huge amount of work that would entail. I just don´t have the time right now. I´m struggling as it is to finish WN9 (two value system) within a reasonable time constraint. Also, the samples would be tiny, and probably not large enough to build a solid model.
  10. This is partly spot on and partly off. I have been observing this "inflation" effect on the right hand side of the WN8 curve. Yes, platoons inflate WRs and cause skewing of the curve in Eureqa, with multiplication of terms allowing for fitting to those higher than expected WRs. You could call it "overfitting". However, I came up with a clever idea to eliminate this problem. What I did is first calculate a corrected WR, using the per-tank expected winrates for each tank. This returns a WR that is tank and tier independent. Once this is done, I simply did rWIN = (cWR - 49.5). This gives ave
  11. Exp values change, two tables, plus formula changes. So, I dunno, maybe calling it WN9 would be a good idea. And I honestly don´t foresee I will be able to dedicate as much time as I am now to WN development in the future, so WN9 might be the end of the road. Also, it´s a self-correcting system due to being able to perform easy and fast updates of exp tank table with a fully automated calculation, perhaps we could do it once per patch as I proposed before, or even monthly, since it is now much faster and effortless to calculate the exp values now.
  12. I really didn´t do an exhaustive analysis, because I really have better things to do with my time, I think Xelos is doing an in-depth analysis of the spotted damage data. Since historical spotted damage will never be available, I saw no sense in going into a detailed analysis for something that is never going to happen. I merely wanted to show that we are not missing out on that much due to not having spotted damage available (specially not for account-wide WN8 measurement). Also, as I said above, it´s a whole different picture for single-battle outcome prediction as opposed to account-wid
  13. I´m not all that familiar with PER, but PIE has a very strong correlation with most people´s percieved "skill" of most players, specially when you compare players amongst a single position. I think anyone would have a hard time dismissing it as useful in measuring a player´s contribution to their team´s winning chances. Top 25 PIE players as of today: Kevin Durant (OKC) LeBron James (MIA) Kevin Love (MIN) Chris Paul (LAC) DeMarcus Cousins (SAC) Al Jefferson (CHA) Carmelo Anthony (NYK) Blake Griffin (LAC) Dirk Nowitzki (DAL) Anthony Davis (NOP) Stephen Curry (GSW) Paul George (IND) LaMarcus
  14. For the single-game data, I analyzed lights vs non-lights, and the results were pretty similar regarding how influential damage is to winning. Spotted damage however, is quite a lot less influential in non-LTs. That being said, we only have aggregate data to calculate WN8 from, so we can´t have separate curves for each tank type for rDMG and such, but this information should already be included in the expected stats table. Also, I calculated how much the error in WN8 is for aggregate stats vs single-tank stats, and it is, IMHO, negligible:
  15. These are actually very interesting concerns, which will allow me to explain WN8 a lot better, I think. That was true for a binary classification system for single games. For overall account stats, the added benefit of the other stats in the formula are much, much larger. For account-wide stats, if we translate WN8 into a WR scale using a linear function and do the same with rDMG, the 95% confidence interval for each player is +-1.84% winrate for WN8 and 2.75% winrate for rDMG. That is a HUGE increase in accuracy, about 50%. Also, using more factors makes the formula more robust,
  16. Keep in mind Su122-54 is a RARE tank. Very few people have played it, and even less "purples" have played it, so the graphs don´t really mean much.
  17. Forget 268 if you want to play assault gun. It has less effective armor than 704, and now it has the same alpha, so...
  18. Basketball provides a better analogy I think. In fact, PIE and PER ratings are very similar to what WN tries to do.
  19. Interesting. However, quite a few of these tanks get crazy XP bonuses, like KV5, rhoomba. Also, they are all tier 8s. Doing damage to tier 10s gives crazy XP, while playing tier 10s always nets less XP compared to 8s. Anyways... I think the only fair comparison is with avg stats for those tanks.
  20. Sela, good to have you back, and did you remove safeshot? I find it useful to keep me from purpousefully shooting retards in the back of their turrets when they constantly keep blocking my shots.
  21. Nice. The only two major calculations missing on your table are vbaddict and XVM. Nice to know everyone is implementing the formula correctly then. I have played a shitload of different tanks. This makes it pretty clear to me that the problem with varying values is probably from handling players with missing battles on API. Right? Last I knew, M60 and M4E3A4 were causing problems due to API not counting battles on them. Orrie what comes up if you analyze AcesHighMDP, who has battles on both tanks, Hansli who has a LOT of battles on M4A3E4, and RodneyDangerfield who has quite a few battles
  22. There are some mods without which I could simply not play this game. Biggest one is server side blue sights. I play with 180-250ms ping, when I turn a corner, without blue server sights, I hit the wall 50% of the times, even if I wait a bit for the server to catch up. Solution is simple, WG should publish list of banned mods and be done with it.
  • Create New...