Android25

WNR - Discussion

54 posts in this topic

I've had several people pm/contact me either asking to help or proposing ideas for WNR but they can't post in Mathlab. This topic is simply meant to extend

in case anybody has any ideas they'd like to share.

Now that I've read a lot of WN9 development I'd like to partially extend WN9, but as only a recent metric that is able to utilize assisted damage with multi-linear regression obtained assisted damage expected values for all tanks.

Share this post


Link to post
Share on other sites

One thing I really want to ask: a lot of of people lately have been referencing WGR as the most accurate rating now, since 9.18. now, it has its issues, but would it be possible to take the WGR formula, reverse engineer it, remove/improve the bad/questionable bits from it (like battle count) and reimplement it?

Share this post


Link to post
Share on other sites

Wasn't battle count effect pretty much nullified after a couple thousand battles? I remember the initial introduction had a large drop off until about 10k battles where it didn't make any effect in PR and in subsequent patches they toned it down a lot.

I find it good enough as a rough idea though based on battle count and my own improvement as a reference point but it still isn't super accurate. 

Share this post


Link to post
Share on other sites

PR as overall, WN9 as recent, problem solved?

Share this post


Link to post
Share on other sites
24 minutes ago, MagicalFlyingFox said:

Wasn't battle count effect pretty much nullified after a couple thousand battles? I remember the initial introduction had a large drop off until about 10k battles where it didn't make any effect in PR and in subsequent patches they toned it down a lot.

I find it good enough as a rough idea though based on battle count and my own improvement as a reference point but it still isn't super accurate. 

IIRC it was 12K battles where it stopped making a difference. I dont know much else, i had a quick look at the formula a few days ago and i think there was one other questionable part to it, like survival rate or something, cant remember.

Share this post


Link to post
Share on other sites
8 minutes ago, Assassin7 said:

IIRC it was 12K battles where it stopped making a difference. I dont know much else, i had a quick look at the formula a few days ago and i think there was one other questionable part to it, like survival rate or something, cant remember.

Yeah i remember them reducing the threshold and the effect of it along with the hit-rate removal after a large outcry. Not sure what the threshold is now though. 

Share this post


Link to post
Share on other sites

PR rating heavy favours damage dealers, so the more high tiers / TDs / Autoloaders, the better, not saying this is bad, but this is a flaw

But this is regardless of rating hard to avoid, since it then turn (again) into dmg vs winning, spot, cap and def points contribute to winning, but how much? a good defender kills the other guy before they reach the flag, while im not getting much spots an do win.....

Share this post


Link to post
Share on other sites

We know the WGPR rating formula and removing battle count modifier is easy. There's mention of it... somewhere in this subforum. WN9 is considerably better than WN8, but still requires frequent updates to the expected values chart, unless it's standardized like WN8 was yesterday.

I still have to play with data I don't yet have, but I would really like to try incorporating assisted damage into the formula. Possibly even less track assist damage. It's (generally) easy to sit in the back and farm damage, but if you required some assisted damage, especially if you multiplied assisted by damage, I think you could weed out those that farm damage vs those actively contributing to damage.

Share this post


Link to post
Share on other sites

WGR is stupidly easy to pad. Play unranked teambattles and you can reach 15k with a tryhard team (they need to be lower PR than you though, but as this goes on it's not very hard, RU server has several people above 14k as of now, and 15k was reached). Just look at NewMultiShow: obviously an inferior player in randoms to A LOT of people by literally every single important metric, yet he's almost 1k points in PR ahead of everyone else on EU (last time I checked). 

All you need to do is get a bunch of 2k PR people on TS with you and you can watch your PR skyrocket while grinding a shitton of credits since SH boosters work for TBs, especially now since TBs are the best way to grind credits after SH hours. The point with this is that you just by playing with the right people you will inflate the metric. Sure W/R can be inflated as well, but padding WR in platoons won't give you extra WN8 just like that. WGR doesn't just take random battles into account which is the obvious flaw here. 

 

 

On another note that I'm curious about, if there were to be a rating (besides the WG one) for ranked battles, would it be the same values? I can't imagine tanks performing the same in 15v15 T10

Share this post


Link to post
Share on other sites
55 minutes ago, Kolni said:

WGR is stupidly easy to pad. Play unranked teambattles and you can reach 15k with a tryhard team (they need to be lower PR than you though, but as this goes on it's not very hard, RU server has several people above 14k as of now, and 15k was reached). Just look at NewMultiShow: obviously an inferior player in randoms to A LOT of people by literally every single important metric, yet he's almost 1k points in PR ahead of everyone else on EU (last time I checked). 

All you need to do is get a bunch of 2k PR people on TS with you and you can watch your PR skyrocket while grinding a shitton of credits since SH boosters work for TBs, especially now since TBs are the best way to grind credits after SH hours. The point with this is that you just by playing with the right people you will inflate the metric. Sure W/R can be inflated as well, but padding WR in platoons won't give you extra WN8 just like that. WGR doesn't just take random battles into account which is the obvious flaw here. 

 

 

On another note that I'm curious about, if there were to be a rating (besides the WG one) for ranked battles, would it be the same values? I can't imagine tanks performing the same in 15v15 T10

Well today I learned. So much for unpaddable metrics.

Share this post


Link to post
Share on other sites
4 hours ago, Luna said:

Well today I learned. So much for unpaddable metrics.

It comes full circle! From Russians finding the need to develop efficiency because win rate could be manipulated with tank companies to this...

Share this post


Link to post
Share on other sites

Ok so I may be alil confused, but if WN8 is going to someday become WN9, and it will have tables to update all the time, same as WN8 did. Why are there plans to moving to it if the tables themselves have become the problem? I have not been keeping up with a lot of this stuff because I stopped playing for a while. So forgive my ignorance if I am missing part of the conversation.

Share this post


Link to post
Share on other sites

WN9 implementation isn't going to happen. The goal is to make a metric which we can leave on its own essentially forever bar any big WG changes, hence WNR. 

 

Share this post


Link to post
Share on other sites

PR is also randomly adjusted by team battles for different people. Personally my pr rises or falls by 40-80 points on average with every team battle win or loss. I've had nights where I gain 200-300 pr and nights where I lose that too. I got up to 9150pr one night only to drop 200 the next. It's very frusterating not being able to play team battles becuase my pr is in extreme limbo. On the other side I've seen players with a lot of team battles literally not lose or gain pr for hundreds of team battles. I'm sure I could get up to 10k pr if I had a good enough team battle team and we played every night, even though it wouldn't be a real 10k pr.

Share this post


Link to post
Share on other sites

I'm realizing since the server will be pulling tank data constantly at a rate of about 750,000 accounts per day, you could just calculate expected values every day rolling old and new data together. This would eliminate any noticable jumps in table changes every 30 days and people's values would hover at more constant values (if there's no change in skill) while consistently correcting for "flavor of the patch" tank padding.

Share this post


Link to post
Share on other sites

If this is now the table Wotlabs is using, when does it get put into actual usage? I am seeing a difference between my data on the website and the data that my after battles tats are producing using the new CSV,JSON from this latest iteration. I am only curious to see what this WNR does to what my overall looks like.

Share this post


Link to post
Share on other sites

@Maggz

This rating would have to be calculated and stored on a central server. I have the server prepared but need to mine the initial data from the API. Once the data is mined, a formula with similar checks to WN9 but with the addition of assisted damage (hopefully) will be created through Eureqa, as well as a formula to calculate expected tank values, which should be more straightforward.

Expected values would be available to the internet through an API, so any website could track users and provide WNR stats. The api would also be available for XVM to request stats of players on the server (since I doubt XVM will want to hold the change values necessary to calculate each player's WNR by itself). In a perfect setup, xvm would send a key, the id of the player the request is for, and the id's of the other players in the match. If the player making the request has activated their statistics on the server (much like XVM already does) then the server would reply with the data. This would provide ad revenue to keep the server running. The website the server runs through (and that you would activate statistics on) would also have it's own stats tracking setup.

All of this so far is theoretical as I can't do anything until I've mined the data of every active player (which I'll be doing over the next 2 weeks), and then mined it again some 30 days later for delta values.

Share this post


Link to post
Share on other sites

Very cool stuff. Would love to see a transparent metric that included assist damage and minimized farmed i.e. assisted damage.  

A few weeks ago I told a redline camping Lorraine to move up and support and he told me to piss off as I was a tomato.  I looked at his stat and he was 47% WR player with .52 spots per game and less than 600 games all played in Lorraine. Got a private message bragging about his 1800 WN8 rating and how I needed to learn to play. Fully admit I could be better at trading...but this is the reason we need to move on from WN8. 

Share this post


Link to post
Share on other sites

Thank you for clarifying Android25, I was I guess assuming that this Was currently tied into the current published wn values posted up the site here. Which as I see now I am incorrect. I am very glad to see how well the new formula works. Good luck in this.

Share this post


Link to post
Share on other sites

the whole issues with the WG PR are why I suggested reverse engineering it and adapting it to our needs, without all the other questionable parts to it

Share this post


Link to post
Share on other sites
2 hours ago, Assassin7 said:

the whole issues with the WG PR are why I suggested reverse engineering it and adapting it to our needs, without all the other questionable parts to it

RkzNiAN.png

win – winrate (from 0 to 1)
surv – survival rate (from 0 to 1)
dmg – average damage per battle
bc – total amount of battles played
bc8.8 – amount of battles played from the point of patch 8.8 release
xp8.8 – average XP per battle (without premium account, from the point of patch 8.8 release)
radio8.8 – average damage done after your spotting (from the point of patch 8.8 release)
track8.8 – average damage done after you detracking the opponent (from the point of patch 8.8 release)

 

Unfortunately there's nothing in it that we can't already just get from the API

Share this post


Link to post
Share on other sites

Can we get assisted damage from the API now?

Share this post


Link to post
Share on other sites

Average assisted spotted and track damage per game since the release of 8.8 are both in the API. It's not listed per tank, so you would have to take it incrementally (hence recent) and use enormous amounts of data to multi-linear regress the expected assist damage per tank.

 

I take it back; average xp per battle without premium is not something you can get from the API. It's also one of the least useful stats :/

Share this post


Link to post
Share on other sites

Average base XP? Wouldnt that be the best single indicator of performance though?

Share this post


Link to post
Share on other sites
9 minutes ago, Assassin7 said:

Average base XP? Wouldnt that be the best single indicator of performance though?

Team performance factor ruins it. You get a coefficient based on the amount of damage your team caused to the enemy team. That's why in games where you do great but the rest of the team absolutely blows it and gets almost no damage, your steller loss game is still only a few hundred xp. I could look into charting it but I suspect it will just be too full of noise.

Here are all the factors http://wiki.wargaming.net/en/Battle_Mechanics#Tank_Experience_and_Credits

Share this post


Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.