Jump to content
RichardNixon

WN9 candidate prototype

Recommended Posts

Platoon ratio 1.88. Rather high, although I did play a lot of platoons with my friends, back in the days :)

It also suggests, that I heavily platooned with better players (because of the medals BIA:Topgun), but that is definitely not the case, they all have and had much worse stats than me.

I generally lack kills, but I have very many confederate medals. I don't know why.

Share this post


Link to post
Share on other sites

No one is complaining about  drop because by going back to per tier with this scale we're back to WN6/8 which were also per tier. Just with all the positive and negative lessons of WN8.

 

All that said, what are your thoughts on completely ignoring historical data and making WN9 for recent only. The nerf buff cycle and meta shifts have really made old data ugly.. If we choose to go recent only then we can leave overall account rating to Personal Rating and just have WN9 for the current meta and relative tank power.

Share this post


Link to post
Share on other sites

All that said, what are your thoughts on completely ignoring historical data and making WN9 for recent only. The nerf buff cycle and meta shifts have really made old data ugly.. If we choose to go recent only then we can leave overall account rating to Personal Rating and just have WN9 for the current meta and relative tank power.

Spotting damage and tracking assists can also be incorporated into WN9 if historical data isn't required. Just a note.

Share this post


Link to post
Share on other sites

Spotting damage and tracking assists can also be incorporated into WN9 if historical data isn't required. Just a note.

Yes, make it recent only and include assistance damage.  Historical data is going to be more and more of a problem as time goes on.  For example, just think how the coming arty changes will screw with stats.

Share this post


Link to post
Share on other sites

Spotting damage and tracking assists can also be incorporated into WN9 if historical data isn't required. Just a note.

Absolutely. I meant to write that but actually feel asleep in the middle of composition and woke up 11 hours later. Perhaps I wasn't entirely lucid.

Share this post


Link to post
Share on other sites

Super Interesting...

57.5 actual win rate

57.8 adjusted win rate

1.78 platoon ratio

Personal Rating: 9037 (WOOT! UNI IN ONE CATEGORY!!!)

wn8 = 2107

wn9c1 = 1807

wn9c2 = 1786

Seems that for my account, WN8 = WN9. At least by my understanding of how they might scale, the values represented by both are all pretty much equal be it WN8 or WN9

Share this post


Link to post
Share on other sites

The platoon ratio is perhaps the most odd one of the bunch. While I do platoon, it tends to be only with 1-2 others. Otherwise, Im playing solo probably 2/3's of the time

 

*Edit*
Im also not a very good kill-secure'er, which can skew? the platoon ratio. Many of my session will have a kill average of 1.5-1.7 per VBAddict. Top-Gun's are just a rare medal for me

Edited by TheMarine0341

Share this post


Link to post
Share on other sites

Recent only would be good, but recent should be more than 1k games.  More like 5k.

An alternative would be to use the patch that introduced assisted damage into the available stats as cut-off point (this would give us roughly 2 years of data for each account). We would finally get a formula that could deal in an adequate way with scouts.

But I would assume such a solution would get you a massive backlash from old overall unicums who want to keep their club as exclusive as possible (which I can understand). In consequence, such a new rating might fail to gain the acceptance of the entire community.

Share this post


Link to post
Share on other sites

An alternative would be to use the patch that introduced assisted damage into the available stats as cut-off point (this would give us roughly 2 years of data for each account). We would finally get a formula that could deal in an adequate way with scouts.

But I would assume such a solution would get you a massive backlash from old overall unicums who want to keep their club as exclusive as possible (which I can understand). In consequence, such a new rating might fail to gain the acceptance of the entire community.

Unicum = top 1%, not x recent.  What it means is that in wn8 terms 2400 is unicum but if we use recent its probably more like 3k wn8 is unicum due to power creep.  That will make a lot of kids who think they have improved angry, but percentile is what matters.

Share this post


Link to post
Share on other sites

Unicum = top 1%, not x recent.  What it means is that in wn8 terms 2400 is unicum but if we use recent its probably more like 3k wn8 is unicum due to power creep.  That will make a lot of kids who think they have improved angry, but percentile is what matters.

Fine by me. But you know yourself that it never stays that way. Once the cutoff is made, people will find the best padders and creep their way back into the unicum range by working the formula. You may have initially 1% unicums but a year later you probably will be closer to 1.5%.

Share this post


Link to post
Share on other sites

My stats are pretty wonky, so I would suggest that this new metric is not very good for assessing up and coming players.  For better and for worse.  My WR takes a big hit because I've only recently started playing the top tiers.  It is mentioned that it's not suitable for low game counts, but is 7K games really a low count?  My platoon ratio is listed as low (.66), even though I'm pretty sure I platoon regularly and it's inflated my WR.  My stats are kind of all over the place to be honest and my Tier X WR gets inflated.  It seems weird to me.

And this might be a little off-topic, but garbad has repeatedly said that winning is all that counts.  I agree.  So "Plain old useless winrate" is ultimately the stat that counts most.  The rest of this stuff is just adding context to that stat.  I'd suggest multiple stats, each with their own contextual limitations, might be a better way of doing that (e.g., Wn8, avg. tier, platoon %) then trying to use a single stat.  If you want to avoid paddable tanks, develop formula that account for tank type and non-linear damage curves.  Ultimately though, I think Gharirey is right in that padders will find a way to pad.  Hence the utility of multiple statistics as a barrier to that.

Share this post


Link to post
Share on other sites

For me, this site is one of the most useful to come in years:

https://clantools.us/servers/na/players?id=1000125653&stats=pubs

It shows me my pub stats without clam wars or TC.  Sadly, its not quite perfect (some very old clam fights are not included) and it still can't parse out platoons.  Wn8 and such (to me) are only useful to screen out platoons.  If someone could somehow win 70% doing 200 dpg, they would be the best player on the server.  DPG has no value on its own, only as a means to winning.  And all DPG is not equal.  Still, everyone knows dpg and such have some impact, so we assume that more dpg = more skill.  The problem is that this tilts the other way -- you have dpg autists who maximize dpg and are indifferent to win rate as long as its not too putrid (under 58%).  That's where we are now -- a lot of the padders are just whoring dpg because they can always seem to find a way to stretch it a bit further even if the server best win rates are effectively capped at ~65-70%.

Share this post


Link to post
Share on other sites

There is no such thing as a perfect metric: If people actually just padded WR as the only meaningful thing, you'd just have tier 10 cap rushes every battle instead. Even WG had to admit that some measure of kills and damage were needed, hence why cap victories almost exclusively provide a negative exp modifier today. But as this is a PvP game, I'd rather have a few damage whores that win 1 or 2 percent less than optimal over cap padders. Damage and kills should be the primary objective to achieve a win, since this is more akin to a 3rd person shooter than an RTS.

This is also one of the main reasons why watching competitive WoT is such a fucking snooze-fest: It's basically an RTS, where the units are too precious to lose, so everyone poke, camp and do nothing for as long as possible. It's better to hope the opponent makes a mistake than to make one of your own. I'd hate for pubs to turn into that, although it would likely never happen.

Share this post


Link to post
Share on other sites

I now looked up some of `special players`

Pljuk, 9k games in SU-26, 
win%  : 65.54%
PR    : 1744
wn8   : 2115
WN9 c1: 1985.4 
WN9 c2: 2130.3 

http://www.noobmeter.com/player/eu/partidista/501859776/

Partidista, 6k games in T67, 5k in SU-26 4k in VK16 and 3k in chaffee
win%  : 60.15%
PR    : 1725
wn8   : 2130
WN9 c1: 2039.5
WN9 c2: 2102.6 

Tazilon, 48k games in VK2801
win%  : 60.15%
PR    : 1856
wn8   : 2407
WN9 c1: 1998.9 
WN9 c2: 2218.4 

Valthornet, 24k games in Type 59
win%  : 56.56%
PR    : -----
wn8   : 2255
WN9 c1: 1780.9 
WN9 c2: 1730.1  

My own: (most played tank IS3 with 780 games and in total 293 different tanks played with a fairly even tier / class distribution)
win%  : 65.7%
PR    : 2064
wn8   : 3006
WN9 c1: 2198.4
WN9 c2: 2186.9  

So it seems also these outliers are falling reasonable within the margin (wn9 is about as faulty as Wn8) however those with a ton of low tier crap tanks seem to (still) get (a bit) too high rating.

As solution, i was thinking of the following: Would it be possible to cut of games if a tank is played more as say 5% of total amount of games? So incase of Valthornet who played 50k games of which 24k in type,  cut of 22.5k type games, and calculate the overall stats base on that, so hes overall is based on only 26k normal games +2.5k type games. That way  peolpe with massive amount of games will get more relevant stats.

The 5% might be placed lower (4%) or higher (6%) depending on outcome. I dont know for example how many people have more as 5% of their games in 1 single tank. Perhaps also exclude tier 9 and 10 tanks (from beiing capped), since they are better balanced, and will thus create less wonky stats (Morgotz, Trakais and Ew_ all have 20% of their games in E50m, but their overall stats are roughly the same, so here its not much of a problem.

But by hardcapping extreme outliers (like cap points are capped) the overall accuracy should become better (??) (or is it not possible to cap games played?)

ps: i myself would go even furter, and also completly drop tier 1-2-3 tanks for the following reasons:
- they now get special mm, newbie protection, making old results slowly no longer comperable with new
- its often pure seaclubbing with tanks like T18, and has nothing to do with `skill`
- they get special mm, only +1/-1, no heavy tanks, many spary guns
- many bad players and terrible crews + no equipment, leading to a huge gap even with equal player skill.

Same can also be said for tier 4 (and 5 and 6 and...) BUT:
- Tier 4 is a terrible tier
- Tier 4 has normal mm, +2/-1
- Tier 4 is the first tier where you need to play more as 2 games to get to the next tank
- Tier 4 can be used for missions and specials + some rewards can only be obtained tier 4 and higher
- Tier 4 has tds, heavy tanks, arty, no machine gun sprayers, its in other words the first `normal tier`.

Im quite sure that by dropping tier 1-2-3 entirely the overall accuracy becomes a bit better and it will also stop rewarding murder killing newbies with a 4e crew skill cruiser. Because 90% of the people driving low tier crap tanks are doing it for demm numbers...

@Garbad: i thought it was not possible to cut off stats at a certain date, otherwise i would fully support picking patch day 8.6 as starting point, and dont include any stats before (the only problem is, all current ``original unicums`` will get a stat boost aswell, my ``recent`` 21k games are 3300 WN8, so when the old games get dropped, i get a 300 point increase, and so will almost all other players who never rerolled (and played since beta)

Edited by GehakteMolen

Share this post


Link to post
Share on other sites

Very interesting.  I think the WN9 c2 calculation may be bugged, though - mine has changed more that 200 points up or down based on playing 5-6 games.  It was in the low 1700s yesterday, then mid-1800s after last night, now back to low 1700s.  Unless you've changed this several times?

Share this post


Link to post
Share on other sites

TD42> Straik.

TD42's stats

WN9 c1 3335.3 well then....

WN9 c2 4958.8 Lol k......

Oops, nerf adjustment to the overall stats was backwards (also relevant to GehakteMolen's examples). I think some of my arty nerfs might be a bit undershot, though, notably the SU-26 and M7 Priest which already have ridiculous modifiers. How on earth did it take WG so long to nerf that shit?

Does TD42 have a non-arty account?

 

DPG has no value on its own, only as a means to winning.  And all DPG is not equal.  Still, everyone knows dpg and such have some impact, so we assume that more dpg = more skill.

It's not solely an assumption. Damage really does correlate exceptionally well with solo winrate. Towards the unicum mark, the standard deviation for tank-adjusted damage is only 6% against tank-adjusted winrate. This suggests a difference of +12% damage for pure damage padding, vs -12% for an optimal play-to-win style, with around 1% of unicums over each limit.

I think people overestimate the effect of damage padding because they include goldspam (which actually boosts winrate along with damage), tank-specific WN8 padding and WN8 non-linearity.

It's not easy to improve on adjusted damage as a metric with what's available. WN8 is far worse (~8% sd), primarily due to the single-point expected values. WN9 improves the standard deviation only to ~4.5%, because spots, frags and defence points are weak discriminators. Assisted damage and maybe a usable armour metric would help, although the worst-correlated class is actually artillery, even excluding the pre-8.6 stuff.

 

General notes on points that various people brought up:

1. You can't use assisted damage for recent stats either, because the API only provides it as an average: There's no indication of many games it's calculated over. It's solely useful for players who started their accounts after 8.8 or 9.0 or whatever, otherwise it's just noise. WG do have better stats. They just won't put them in the API.

2. You can't directly control the weighting of tanks in an overall or recent metric unless tanks/stats is used, which is a far slower API than account/tanks and hence problematic for stat sites. WG could provide a cut-down version of tanks/stats with reasonable performance, but apparently they're not interested. I'll provide a weighting scheme if anyone wants to use tanks/stats for overall or recent metrics: Something like x1.3 weight per tier is probably fine. Fancier stuff such as giving nerfed tanks lower weight in the overall calc is also possible.

3. The last time dumping the overall stat was suggested on WoTLabs, the idea was shouted down and the thread closed. Removing it entirely doesn't really solve anything though, except for persuading people that maybe recent stats are more valid. Recent stats acquired a bad reputation because they were so easy to pad with WN8: 1000 games should be good for WN9.

 

Very interesting.  I think the WN9 c2 calculation may be bugged, though - mine has changed more that 200 points up or down based on playing 5-6 games.  It was in the low 1700s yesterday, then mid-1800s after last night, now back to low 1700s.  Unless you've changed this several times?

Changed it again today :P

Don't take too much notice of the overall value for the moment. I might just switch it for a tanks/stats version, and the scale will change.

 

Someone requested a link to the nerf values. It's right there in the javascript: http://jaj22.org.uk/expvals_v20.csv. If you want something that's readable in a browser I chucked the relevant values here as well:

https://docs.google.com/spreadsheets/d/1aBbeJgvH5bETAabx-2YURIXF-gMqLwdFppDDZg8s47o/edit?usp=sharing

Share this post


Link to post
Share on other sites

 

It's not solely an assumption. Damage really does correlate exceptionally well with solo winrate. Towards the unicum mark, the standard deviation for tank-adjusted damage is only 6% against tank-adjusted winrate. This suggests a difference of +12% damage for pure damage padding, vs -12% for an optimal play-to-win style, with around 1% of unicums over each limit.

You have to estimate solopub win rate; how can you possibly conclude you have done so without error?  So far in this thread, I've had the lowest rate of platoon padding...and I almost certainly have played more than half of my games in a platoon with 2 other supercums.  There is simply no way I platoon the least of supercum; your platooning estimate is clearly wrong.

Besides, even the technique of the damage padders shows another error.  Tip #1 is stay alive.  Why?  Not to win close games, but because the difference in an average game and a good game is snapping up meaningless mop up damage.  By their own admission, the damage doesn't create the win, the win creates the damage and they stay alive to exploit it.  Your stats can't tell this.  Tip #2 is to be consistent.  Yet we all know ~25% games are auto win, ~25% are auto lose.  50% of the time, your performance makes no difference to winning, yet it makes a tremendous difference in stats.  Likewise, the best chance to win a very improbable game is often a risky gamble...that damage padders won't take, because better a consistent damage loss than a small chance of a win.  Again, your stats can't tell this.

Or look at me vs kewei.  He often exceeds my DPG by up to 50%, yet with both of us having gone head to head over 1,000 games we are within 1% of each other.  Either we have to conclude that I am some kind of freak of nature who is the exception to the DPG = skill rule, or DPG isn't everything.  For me its obvious, because I can watch guys and tell if they are good at this game or good at doing damage.  And also I'm not the only one -- crab as I recall got good win rates with crap dpg.  So did illusion.  And very few (kewei excluded) of the top damage dealers are the best winners in a tank.  They are often acceptable (63-65%) but the best on the server should be better than a routine unicum doing 33% less dpg.

And finally, taking your own figures.  95% of unica are within +/- 12% of the mean dpg for a given win rate.  Let's say an IS-6, and assume 2500 dpg = 65% win rate.  So we are looking at 5% of unica getting a 2200-2800 gap, ~25% difference in DPG and the same win rates.  And some will be even more.  Literally everyone will swear the 2200 player is worse than the 2800 player...and be wrong 5% of the time.

Share this post


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...