Jump to content
RichardNixon

WN9 candidate prototype

Recommended Posts

Richard, what is the 'actual data' are you using to create the platoon formulas?  The API data is very limited, CC infrequent enough to provide very little insight, BIA depends on survival of the entire platoon, and TG can vary a lot depending on platoon makeup.  Those awards are also very dependent on the skill level of the platoon;  a unicum platoon rocking 80% WR will get them at a much higher level than a blue platoon pulling 65%, so accuracy would fall off pretty quickly as the skill of the platoon drops.

I just don't see how that data can give you a reasonable level of accuracy across the playerbase.

Share this post


Link to post
Share on other sites

My platoon ratio is 35% solo, 32% 2-man platoon 33% triple, so should be about right (say 1% deviation, its almost spot on, i think)

The color scale u showed is for both wn9c1 or wn9c2?

ps: super uni would be 2200, im 2198, about right imo, since i consider myself border case better as joe-avg-unicum, but also no true super unicum (same my winrate hoovers around unicum / super unicum)

ps ps: no idea why all your platoon ratings are so wonky, or are you all often platooning with much better / worse players :P

Share this post


Link to post
Share on other sites

@RN: with the recent changes you made, how do those graphs in your original post look now? The low std deviation of WN9c2 was impressive then, it would be good to see how it looks now. 

Share this post


Link to post
Share on other sites

What do you think about adding this platoon ratio as one of the main data? Like, on wotlabs.

Cause the disparity between solo and platoon difficulty in maintaining high WR is HUGE.

Share this post


Link to post
Share on other sites

If that's true over your whole career then you're platooning with players much worse than yourself. Your adjusted winrate is bang in line with your WN9 for a solo player.

The platoon ratios only work as a direct percentage if you played with players of very similar skill. If you platoon with inferior players, the ratios will come out much lower. If you platoon with better players, then the ratios will come out much higher. This is by design, because it's supposed to be used as a platoon padding indicator, although it's also the only way that it can work.

I should have explained this when I turned it into separate percentages, but I forgot

That's how I figured it worked.

You might want to rename the fields for clarity, e.g...

  • "2-person padding platoon" instead of "2-way platoon ratio"
  • "3-person padding platoon " instead of "3-way platoon ratio"

...and get rid of the use of percentages, unless a value of 20% for 3-way platoon ratio really means you're platooning in full platoons 20% of the time.

I also think you're trying to cut too fine a line by having separate 2-way and 3-way numbers.

The underlying thing you're trying to make visible is the extent of skilled platoon padding, right? If that's the case, just sticking with one metric / number is sufficient. E.g. just have one value called "padding platoon ratio".

Anyway, my point being that the name is really critical for context and correct interpretation.

All that said, from my looking up various players the platoon numbers from a relative perspective look right. Therefore, your formula and theory based on TG/CC/BIA seems to be working.

Nice work :)

As others have said there is something wonky with the platoon ratio.  For an almost elxcusive solo player 0.33 seems high.  Then again it's just internet pixel tank numbers.

0.33 is actually a low number.

It might also be reflective that when you do platoon, you do so with good players.

The API data is very limited, CC infrequent enough to provide very little insight, BIA depends on survival of the entire platoon, and TG can vary a lot depending on platoon makeup.

I just don't see how that data can give you a reasonable level of accuracy across the playerbase.

I believe you're misintrepeting the purpose of the platoon metric. It's more of a measure of platoon padding with skilled players.

Skilled padding platoons will get a lot of BIAs, because Unicums are smart enough to stay alive much more frequently than your average pubbie.

The platoon metrics look directionally accurate to me, when checking on multiple players.

Share this post


Link to post
Share on other sites

Winrate 64.6%
Adj. Winrate 64.2%

2-way platoon ratio 34%
3-way platoon ratio 49%

Pretty much correct, as I hate playing alone, and I usually play with people in the same skill level.
This game is the only game that have gone through my rather tolerant and stable mind, and made me kill my (old) desk.
 

:)

Share this post


Link to post
Share on other sites

0.33 is actually a low number.

It might also be reflective that when you do platoon, you do so with good players.

 

with the amount of games I play in a platoon it shouldn't effect it that much.  Id say in my past 60 days (1300 battles) I have platooned maybe 30.  To give you an idea of how much I actually platoon.  But you right out of those 30 battles I probably  earned 5-10 BIAs.

 

 

 

edit:  what's probably throwing it off is the amount of scouting I do.  Raising my WR without platooning/damage/kills.

Share this post


Link to post
Share on other sites

iY6rSqW.png

#1 Its nice seeing TDs being mor eor less back in line, wn8 pretty much prevents TDs from going over 5k no matter what you do.

#2 I actually don't see much with rSpot, i've checked many good players and it doesn't seem like there is a strong correlation at the top-most level

#3 54 seems to take a pretty big hit from c2, not sure why.

Share this post


Link to post
Share on other sites

Richard, what is the 'actual data' are you using to create the platoon formulas?  The API data is very limited, CC infrequent enough to provide very little insight, BIA depends on survival of the entire platoon, and TG can vary a lot depending on platoon makeup.  Those awards are also very dependent on the skill level of the platoon;  a unicum platoon rocking 80% WR will get them at a much higher level than a blue platoon pulling 65%, so accuracy would fall off pretty quickly as the skill of the platoon drops.

I just don't see how that data can give you a reasonable level of accuracy across the playerbase.

I didn't figure that it was possible for a while. There are two main tricks:

  • Using vast quantities of replays to determine how players platoon at each tier and skill level.
  • Using API per-tank achievements and then merging individual tank results by tier and skill rather than player.

There is an assumption that the EU replay and API samples have similar platoon rates. I should probably cross-check against different replay samples at least, although the numbers seem reasonable.

Medal infrequency and especially CC infrequency is an issue for the ratio calculation: By luck, a player might easily get 0 CCs when they should have got 2, or vice versa. If a player only has a handful of a particular medal then the calculation can be a long way off.

 

What's considered platooning with a "better player" by the ratio? One with higher WR, or one with higher WN?

Platooning with someone who gives you more BIAs or CCs than your frags & TGs predict. The central point isn't strictly set on an equal-skill player, but a player who someone with your skill level platoons with, on average. If you're good, this means a slightly worse player.

 

Share this post


Link to post
Share on other sites

What do you think about adding this platoon ratio as one of the main data? Like, on wotlabs.

The trouble is that it's very easy to intentionally avoid BIAs and CCs, so it only works as a private metric (on that note, I wish they'd put mileage in the API). It's also pretty useless for anything except overall metrics because the medal counts get too low.

It's also a pain explaining new metrics to people, although as taugrim explained, this is partly my fault. Still, on GD, no-one even understands adjusted winrate :(

 

#2 I actually don't see much with rSpot, i've checked many good players and it doesn't seem like there is a strong correlation at the top-most level

#3 54 seems to take a pretty big hit from c2, not sure why.

2. Spots would be useless as a primary metric, but when combined with frags they really do work well as a discriminator. You can think of both WN8 and WN9 as "damage modified by some other junk". Good players don't necessarily get good spots, but good players with good spots tend to win more games than damage alone would predict.

3. Spots are on the low side, but a lot of the difference is the nerf adjustment. If you played it since the big HEAT nerf then you should use the "WN9c2 unmod" value instead.

 

RN, could something similar to the platoon ratio with BIA, cc and TGs be done with scout and patrol duties?

Now that's a fun idea. The main issue is they're probably too rare to work on recent data, so it may need to be overall-only. Another issue is that due to map changes and variation in scout tank counts, it's possible that the chance of a patrol duty has dropped substantially over time. Note that this would also reduce the value of assisted damage as a metric, even if was available. Might be worth a look though.

BTW I get your current platoon ratios as 7% and 3%. The data favoured you a lot more than the model did.

Share this post


Link to post
Share on other sites

Can someone explain me the platoon-ratio?, I mean, Does platoon-ratio influence current WN8?, how? and, until what point?...............will influence WN9? if I understood correctly, It won't........Sorry but I read faster because I am at the office :(

many thanks

Share this post


Link to post
Share on other sites

Can someone explain me the platoon-ratio?, I mean, Does platoon-ratio influence current WN8?, how? and, until what point?...............will influence WN9? if I understood correctly, It won't........Sorry but I read faster because I am at the office :(

many thanks

read slower when you get home.

Share this post


Link to post
Share on other sites

Website updates:

  • Added static recent values. Should work for anyone on EU or NA with 10 or more battles over the last two weeks.
  • Updated WN9c2 & adjwin expected values with data from the last few weeks, as the previous run was a bit thin.
  • Cleaned up some minor stuff.

For reference, these are the tanks that currently get manually copied from others:

  • ISU-130, because no data. Took from Sturer Emil this time.
  • KV-220 test = KV-220
  • Pz V/IV alpha = Pz V/IV, but they're both very thin.
  • 57 100 = GF version, because WG fail at tank IDs.
  • A32 = A20, data shortage.
  • Jpn Tiger = VK36, SP IC = T71. Should have data in a couple of weeks.
  • A45 = Black Prince. Gonna be stuck there for a while.

Anything else should have legit recent values.

Share this post


Link to post
Share on other sites

My recent WN9c2 is 2255 for 181 battles. Nearest sample from noobmeter that I could find was 154 battles with 3458 WN8 and FV4202 as most played tank. Maybe you can add a recent WN8 and per tank stats to that so that we can compare.

 

There is one thing I do not get: It says "estimated Top Guns 146" even though I have 141 and have never played before any changes in medals (I'm pretty sure I started in 8.8 or 8.9). Maybe adjust that for the age of the account?

 

edit: Another thing: Did you alter something about the formula again? My account WN9c2 dropped by 60 points or so.

Share this post


Link to post
Share on other sites

lol my platoon ratio sum is above 100%. I platoon more games than i play. 

(It might be a bit screwy with me though because the people i platooned with if i did in my first 5k games were of similar skill, even then. I can't even remember if it was 2 or 3 man plats)
Might need a bit more tweaking. I know I've played 10-20% solo, and i do know I'm a baddie platoon padder as well.

Could just write me off as a statistical anomaly, which could be the case.

 

I can't comment on anything else as my recent gameplay has been slightly better than my overall.

Oh, and the adjusted win rate i reckon is pretty accurate. I'd call myself more of a 57-58% player than my overall states. My recents reflect that.

Share this post


Link to post
Share on other sites

My recent WN9c2 is 2255 for 181 battles. Nearest sample from noobmeter that I could find was 154 battles with 3458 WN8 and FV4202 as most played tank. Maybe you can add a recent WN8 and per tank stats to that so that we can compare.

I can't do any tank breakdowns on the recents because I only have 150MB of database capacity. Even RU would be marginal. I'll probably add recent WN8 though.

 

There is one thing I do not get: It says "estimated Top Guns 146" even though I have 141 and have never played before any changes in medals (I'm pretty sure I started in 8.8 or 8.9). Maybe adjust that for the age of the account?

Estimated top guns is actually a straight calculation from your per-tank frags. Comes out surprisingly close for most players. It's a demonstration that top gun probability (relative to frags) isn't much affected by playstyle or platooning.

 

edit: Another thing: Did you alter something about the formula again? My account WN9c2 dropped by 60 points or so.

It's mostly the expected values re-run. I need to figure out a solid normalization method and stick to it. Essentially I need to pick something to remain constant, for example tier 10 tanks with a defined popularity distribution. Tier 10 has changed substantially in the past (removal of tier 12 MM, general upwards migration), so I'm not too sure about that particular idea.

 

Could just write me off as a statistical anomaly, which could be the case.

Considering the numerous known sources of inaccuracy here, 110% vs 90% is an excellent estimate, especially for mostly 3-way. If the vast majority of the blue+ accounts were that close then I'd be ecstatic.

 

Share this post


Link to post
Share on other sites

 

edit: Another thing: Did you alter something about the formula again? My account WN9c2 dropped by 60 points or so.

Yeah same, around 50 difference(drop) from yesterday. I nearly got 100 more c1 than c2 now.

 

I'm interested in the final product indeed, and I do hope you find the right and most fitting values.

Share this post


Link to post
Share on other sites

Some quick feedback:

Huge difference between low and high tier tank impact. With only some small exceptions, all my stats on tanks below tier 5 take a huge boost. The only other tanks that are boosted are tier 5+ SPG. All other tanks above tier 5 take a hit. Of those, frenchy lights take the hardest hit. That's good, because I always thought my ratings in those were outrageously high. 

 

It seems to me, for now, that statpadding may be done mainly in low tier tanks, campy TD's and SPG's. Mobile meds and -some- lights take the hardest hit and are not as viable for padding anymore. HT's seem to be largely unaffected, with the exception of my Tiger I and Tiger II. 

Share this post


Link to post
Share on other sites

Some quick feedback:

Huge difference between low and high tier tank impact. With only some small exceptions, all my stats on tanks below tier 5 take a huge boost.

That's just a scale thing. WN8 has a much higher zero point and non-linear scaling, so bad players get higher WN9 than WN8, while good players get lower WN9 than WN8. You were bad when you played your low tiers, and so they go up.

A lot of less-popular low tier tanks were underrated in WN8, so it was possible to get very high WN8 values in them with vaguely competent play. Hence my 3400 WN8 in the Pz IIIA. Similarly, Kuroialty currently has 2070 WN9 (a little over unicum) and 3190 WN8 (over superunicum).

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...