Jump to content
Gryphon_

Estimation of WN8 Expected Values for Rare and New Tanks

Recommended Posts


Every three months I take data from approximately 25,000 accounts provided by vbaddict and process the data to update the expected values for WN8. WN8 was intended to be dynamic; new tanks arrive, older ones get buffed and nerfed, and sometimes tanks swap tiers. The methodology for the original build of WN8 expected values is here and the methodology for the updates is here. Both are widely misunderstood:

  • Expected values are not averages
  • Expected values are set such that the 'benchmark' WN8 player - who has an 'overall' of 1565 - should be able to hit the expected values in every tank once he/she has played greater than 50 battles in the tank
  • The update method uses a linear regression to see the relationship between the rSTATs of all players playing the tank and the rSTATs of those same players playing ALL their tanks
  • The linear regression is done using the statistics software application 'R' and the result can be shown graphically
  • If the expected values of a tank are correct, the red line should pass through (1.0, 1.0). if it doe snot, an adjustment to the expected value for that rSTAT is calculated to raise or lower the red line such that it will pass through that point. 
  • This process is repeated for all tanks every 3 months - or sooner if needed. 

%20rDAMAGE%20_vs_%20user_rDAMAGE%20_12_1

 

The process has some well known issues:

  • The expected values are derived from how all the players are playing the tank, rather than from the properties of the tank itself. Two almost identical tanks in different lines of the tech tree can, and do, produce slightly different values.
  • There is no data on new tanks so the initial values are estimates, either type/tier averages, or very similar tanks. 
  • Any update process that relies on a linear regression needs sufficient data to give a result that has high accuracy. Some tanks are rare (or new) and data for those tanks is slow to arrive and normalize (need high battlecounts per user per tank). 

To mitigate these problems the community is invited to comment on the expected values for those tanks where data is sparse. Starting with the very worst, rarest tanks, I will post the current expected values for them. You are invited to discuss the values and whether they should be changed, and if so why. 

 

The strongest argument is: tank A  = tank B; in these cases we will probably decide to use same values for both.

 

The next best argument is: tank A is better than tank B for <reason> but not as good as tank C for <reason>; in these cases we will probably decide to average B and C to get values for A. 

 

You get the picture. Your opinion is welcome, but its the basis for your opinion that matters most.

 

Here is the first batch (below). If this goes well, there will be further batches. For reference on other tanks values, the 'change' spreadsheet from v18 to v19 is here. The current values are the values in the GREEN panel. 

 

Finally, when discussing values, use the Code function in the Editor to get a legible table like below

id	title	                type	tier	eFRAG	eDAMAGE	eSPOT	eDEF	eWIN	accounts
55841	T95E6	                2	10	0.81	1664.00	1.51	0.71	46.50	1
53793	T95E2	                2	8	0.86	1118.55	1.58	0.99	51.70	6
1361	Stuart I-IV	        1	3	1.08	252.87	2.53	1.27	57.56	7
54033	Pz.Kpfw. V/IV Alpha	2	6	0.66	454.44	1.63	0.62	49.62	7
60945	T 55A	                2	9	0.94	1457.35	1.80	0.81	51.34	7
56609	T28 Concept	        4	7	1.07	1089.15	0.81	1.13	53.42	11
593	M2	                1	2	1.61	278.45	2.41	1.58	60.89	12
59665	Grosstraktor - Krupp	2	3	1.39	323.51	1.64	1.19	56.43	13
60689	StuG IV	                4	5	1.34	591.84	0.60	1.27	56.88	15
13905	FV4005 Stage II	        4	10	1.20	2370.10	0.45	0.47	46.52	16
52065	STA-2	                2	8	0.89	1196.34	1.34	1.05	51.48	16
1617	Grant	                2	4	1.04	388.59	0.86	0.85	54.04	34
13137	FV4004 Conway	        4	9	0.87	1610.79	1.04	0.69	50.07	36
60433	Pz.Kpfw. II Ausf. D	1	2	1.30	231.95	2.35	1.15	58.88	42
51201	KV-220 Beta-Test	3	5	1.49	676.28	1.39	1.54	59.92	59
54273	SU-76I	                4	3	2.35	475.54	1.07	1.80	62.87	59
63553	AMX Chasseur de chars	2	8	0.95	1244.04	1.43	1.43	53.04	99
 
Link to post
Share on other sites

Would we see any benefits from switching over from a linear regression to a polynomial? Tanks with different "skill ceilings" would be fitted better assuming that relative performance is not a fixed ratio everywhere. I can see curvature in that M48 graph - it looks like the values under 1.0 RDamage are consistently low and those above end up diverging high. 

Link to post
Share on other sites

Thank you, Gryphon for your continued work and taking into account the criticisms of the not always too eloquent community.

Link to post
Share on other sites

Seconding Miros, I appreciate your work Gryphon_ and have to thank you for taking the time to do it. Some of us are not very patient people when it comes to certain stuff, so I must apologise for seeming frustrated.

Link to post
Share on other sites

The CW rewards are missing.

 

 

 

The tanks that I have found (I didn't check all of them) that need adjustment:

 

Obj. 907 = average of Object 140, Object 430 and T-62A (or just the stats of the best of the 3)

M60 = M48

T-55A =< T-54

ISU-130 = ???

 

M60 is currently too low. Object 907 is currently too high. T-55A is currently too high. ISU-130 is roflcopterwtfomg too high

Link to post
Share on other sites

CW rewards need to be there.

 

Also, I don't think we need to do the tanks that are available to the public like the Conway, CDC or FV4005, Not yet atleast.

Link to post
Share on other sites

I'm guessing that the T23E3 probably scales better with the T21, because they get the same matchmaking, have similar speed, and have the same gun.

T21 see's tier 9's though.

 

The comparison is solid, but the battle tiers is an issue.

Link to post
Share on other sites

@Ollie Tabooger

 

*Disclaimer* I have not played either the T21 or the T23E3. My input is only a counterpoint so we can find the best solution.

 

Playing Devil's advocate here:

 

The T21 and the T23E3 are similar tanks but have very different roles. Again, I have zero experience with either of these tanks but is it fair to compare a LT that gets +3MM to a MT that gets preferential MM? (not sure about the T23E3's MM, pl0x confirm/deny).

 

A T21 that might spend it's time scouting might get much lower dmg values than a med that is actively looking for damage/flanking opportunities.

 

Would the Panther M/10 be a more suitable comparison? Might run into small sample size here (as it isn't too popular and is going to be removed).

 

Just my pubbie .02

 

Edit: Ninja'd

2nd Edit: T21

Edited by FavreFan4ever
Link to post
Share on other sites

Here's the Obj. 140, Obj. 907, and Obj. 430 comparison table only including differences for convenience purposes for deciding the damage if and only if the Obj. 907 is added to the list:

YhKjTUz.png

Draw your own conclusions.

Link to post
Share on other sites

T21, not T71, Favre. I do not believe the T23E3 and T21 share enough in common to be considered at all similar, though. Different matchmaking, hugely different stats.

 

I thought we were supposed to be discussing the tanks of the current batch?

Link to post
Share on other sites

ISU 130 is likely the worst T8 TD.

 

Furthermore, it's owned by what, 30 people on NA?  I nominate it as a prime example of "better off ignored."  If that is not an option, give it the average of the worst two T8 TDs.  Can't really pad in the thing, no matter what the numbers are set at.

Link to post
Share on other sites

Just to be clear, the accounts column you've provided represents what, exactly?  The number of accounts you've got that have usable data for that tank, or the number of accounts that remain after you've filtered down to what you're actually using.

 

 

Pz. II Ausf. D - I'm going to make the case that this tank most likely performs between the Pz. II and the T7 Car.  The vehicle is not a perfect match with the Pz. II, being most different in health pool, overall armor, and top gun, though gold spam in the Ausf. D can mostly compensate for the gun.  I would put the Pz. II as being better than the Ausf. D.  The obvious worse tank to compare the Ausf. D to is the Pz. I, but that tank and the Pz. II share so many qualities that are different from the Ausf. D (DPM, top speed, radio) that it doesn't allow the Ausf. D to be an appropriate middle for the two.  Instead, the T7 Car shares the speed, DPM, and radio problems that the Ausf. D has, despite some more obvious differences.

 

M2 - This tank pretty much is the M2 LT.  There are some differences (and I think the brit version is better), but nothing else is close enough to being like the M2 LT for the M2 to perform as the average of two.

 

FCM 36, Renault R35 - Not on your list, but these tanks will hit the live server soon.  If you are having trouble getting data on the M2, I don't imagine you'll find more success on getting data on either of these.  Right now, I think these have been given expected value estimates that don't make any sense.  These tanks basically are the Hotchkiss 35, which has much different expected values than what these two tanks have been assigned.

Link to post
Share on other sites

I thought we were supposed to be discussing the tanks of the current batch?

 

Correct

Link to post
Share on other sites

ISU 130 is likely the worst T8 TD.

Furthermore, it's owned by what, 30 people on NA? I nominate it as a prime example of "better off ignored." If that is not an option, give it the average of the worst two T8 TDs. Can't really pad in the thing, no matter what the numbers are set at.

Yep. About the only thing going for it performance-wise is camo and gun depression. It also doesn't seem to be in the API. Edited by Tedster59
Link to post
Share on other sites

Just to be clear, the accounts column you've provided represents what, exactly?  The number of accounts you've got that have usable data for that tank, or the number of accounts that remain after you've filtered down to what you're actually using.

 

 

Pz. II Ausf. D - I'm going to make the case that this tank most likely performs between the Pz. II and the T7 Car.  The vehicle is not a perfect match with the Pz. II, being most different in health pool, overall armor, and top gun, though gold spam in the Ausf. D can mostly compensate for the gun.  I would put the Pz. II as being better than the Ausf. D.  The obvious worse tank to compare the Ausf. D to is the Pz. I, but that tank and the Pz. II share so many qualities that are different from the Ausf. D (DPM, top speed, radio) that it doesn't allow the Ausf. D to be an appropriate middle for the two.  Instead, the T7 Car shares the speed, DPM, and radio problems that the Ausf. D has, despite some more obvious differences.

 

M2 - This tank pretty much is the M2 LT.  There are some differences (and I think the brit version is better), but nothing else is close enough to being like the M2 LT for the M2 to perform as the average of two.

 

FCM 36, Renault R35 - Not on your list, but these tanks will hit the live server soon.  If you are having trouble getting data on the M2, I don't imagine you'll find more success on getting data on either of these.  Right now, I think these have been given expected value estimates that don't make any sense.  These tanks basically are the Hotchkiss 35, which has much different expected values than what these two tanks have been assigned.

 

The total number of accounts in the dataset, and yes, some will fall out in the filters.

 

Thank you for laying out the rationale of your recommendations so clearly, you have set the standard.

Link to post
Share on other sites

Hey look the PZ V/IV A was fixed. Dia is going to be promoted from perma blue to perma purple!

907 still hasnt been fixed tho ;~;

Hey Gryphon_, what are the current plans for reward tanks?

Link to post
Share on other sites

T95E6: Lets ignore this for now, only WG employees/testers have access to this tank and we have no indication of it being released anytime soon. Nobody has played it so we aren't able to make any comments.

 

FV Conway, AMX CDC, FV 4005, Grant, STA-2, SU-76I: These tanks are not really special tanks and do not fall under the category that Xen specified in the other thread, they should be excluded from this group.

Link to post
Share on other sites

Hey look the PZ V/IV A was fixed. Dia is going to be promoted from perma blue to perma purple!

907 still hasnt been fixed tho ;~;

Hey Gryphon_, what are the current plans for reward tanks?

 

We will work our way up through the tanks with little to no data, to get into the swing of working on this in a constructive manner. Next batch onwards, we will be looking at tanks where there is a reasonable amount of data, but some of you might mistrust the analysis, and want to explain why

Link to post
Share on other sites

 

FV Conway, AMX CDC, FV 4005, Grant, STA-2, SU-76I: These tanks are not really special tanks and do not fall under the category that Xen specified in the other thread, they should be excluded from this group.

 

No, I believe the expected values of tanks that are are still new need some looking at as well. Looking at the values of different tanks, it seems that new tanks often get unreasonably high expected values. But adjusting all those values is probably too much work. But maybe leaving new tanks with placeholder values a while longer until the available sample is more conclusive would be a good idea.

Link to post
Share on other sites

It's hard to quantify but I think a good conservative estimate (as in better slightly too high than too low) for the 907 would be to just equate it to the 140's expected stats.

 

Compared to the 140:

- 1 degree less depression

+ 100 more hp

+ statistically better hull due to spaced armor + weird angles.

- substantially worse gun handling

+ slightly better dpm

- really awful ammo rack layout

+ forward turret makes corner fighting slightly easier

 

I think it comes out slightly behind the 140 mainly because of gun handling and depression. The gun handling more than negates the DPM advantage and the one degree really makes the 140 a lot more versatile and able to hull down better. On the other hand, 100 extra HP is nothing to sneeze at.

 

I think the 907 outclasses the T-62A mainly due to speed, hull armor, 50 extra hp and a smaller profile (harder to hit + better camo), which make it more flexible; the T-62A really only has a turret going for it, which is difficult to use consistently with 5 degrees of depression.

 

Regardless, I feel like the 907 is a lot closer to the 140 in overall capability than the T-62A.

Link to post
Share on other sites

Yeah, ISU-130 a shit.  All the decent players I've talked to find nothing really redeeming about it and I'd have to agree.  The pen is too low to engage many targets from the front reliably and it's too big and fat to relocate.  It does have decent depression and alpha/dpm, but it's far from enough to make up for the other failings it has.  It doesn't have the armor, penetration, mobility, or potential for abusing the vision system the other tier 8s have.

 

SU-76I doesn't really have other similar tanks at its tier, but I'd rate its strength somewhere around what the PaK40 has.  Might have to make something up for this if not enough people bought it.  WR can be padded in this easy, so it shouldn't be ignored in counting WN8.

 

Something to consider about the Pz II D is that unlike the Pz I and the T7 combat car, it does not get pref mm.

 

KV-220 beta is the exact same as the normal KV-220.

 

The StuG IV is certainly a lot like the StuG III, though one gets a better gun and the other gets better MM.  Since the gun will always be there for you, I'd say the StuG IV should have slightly lower expected values than the StuG III.

 

T95E6: Lets ignore this for now, only WG employees/testers have access to this tank and we have no indication of it being released anytime soon. Nobody has played it so we aren't able to make any comments.

 

FV Conway, AMX CDC, FV 4005, Grant, STA-2, SU-76I: These tanks are not really special tanks and do not fall under the category that Xen specified in the other thread, they should be excluded from this group.

I'd personally put the SU-76I with the other rare tanks since it wasn't even on sale for 24 hours IIRC.  Agreed that T95E6 should be ignored.

Link to post
Share on other sites

FV Conway, AMX CDC, FV 4005, Grant, STA-2, SU-76I: These tanks are not really special tanks and do not fall under the category that Xen specified in the other thread, they should be excluded from this group.

I dunno, available for only 3-4 hours sounds pretty special to me. It's also quite OP for a premium tank. But comparing the SU-76 and SU-76i the expected values are okay, in my opinion.

The SU-76i has basically the same gun as the normal SU-76, but it has special matchmaking. Because of that the average kills and damage should be higher. it can deal damage in basically every match.

 

 

KV-220 Beta Test

It's exactly the same as the normal KV-220.

 

T95E2

Almost the same gun (but HEAT as gold) as the Super Pershing, but no armour. It does have the speed and mobility of a medium though. However, the T95E2 sees tier 10.

Current expected stats for the Super Pershing and T95E2 are almost the same. Seems okay to me.

 

Pz V/IV Alpha

Exactly the same as the Pz V/IV. However, most games played in the Pz V/IV (Alpha) are done in the 'old' Pz V/IV. It has received significant buffs and nerfs over the past few years.

It's still a Pz IV turret and gun on a Panther chassis. It's slow, has no workable armour (lolturret), no DPM, no gun depression and it gets set on fire or ammoracked every other match.

 

tl;dr: Tank is shit. Expected values are probably okay.

 

Another question on the Pz V/IV, Gryphon. A few patches ago I grinded 3 MoE on the Pz V/IV (Yeah, I like to hurt myself) and with the next expected values update the expected damage jumped around 100 points (if memory serves me right). Is it possible that I was responsible for that jump in expected damage?

If yes, is it possible to lock the Pz V/IV stats so that my stats won't eventually become the expected stats, because I'm the only one playing it?

Link to post
Share on other sites

1361 Stuart I-IV

Similar to the M3 Stuart, stats seem fine.

 

1617 Grant

Similar to the M3 Lee. Trades a higher pen gun for worse mobility. The damage for the Grant seems to high. Gold ammo and only selected players going for the line might play a roll.

 

52065 STA-2

As said before, toned down STA-1. Worse pen, worse mobility. It's the stock STA-1 with better turret and second gun.

Gunwise comparable to the Pershing. From that and grinding the tank, the damage seems a tad to high.

 

54033 Pz.Kpfw. V/IV Alpha

Merge the dataset with the Pz.Kpfw. V/IV. Calculate expected value from that. Seeing User and Dia_ are/have been playing it a lot might distort the data

 

51201 KV-220 Beta-Test

Merge the dataset with the KV-220. Calculate expected value from that.

 

59665 Grosstraktor - Krupp

Saw it once or twice in battle. Doesn't really have leverageable capabilities. Isn't available anymore? Damage might be to high because it's a seal clubbing tank.

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...