Jump to content
Gryphon_

WN8 Expected Values Update

Recommended Posts

Problem is that reward tanks arnt played by the overall server population, but only a small subset opposed to regular tanks, which makes these numbers not really comparable.

 

The expected values calculation uses a linear regression to find the line of best fit of player performance in tank vs player performance overall. The expected values are set so that the line always goes through the user vs tank rSTAT (1.0, 1.0). In the case of tanks like the 907 (and M60) most of the datapoints are above that point, but as long as the line has a high adjusted r squared then the fit is good, so the ex values are fine -  even if most players who play those tanks exceed them, not only in that tank, but in their other tanks too.

 

Here is the plot for the M60 so you can see;

 

 

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

 

Share this post


Link to post
Share on other sites

Gryphon I didnt have stats in school  to a degree that I can argue with you, if you start throwing "numbers" around.  :$ 

But please could you also post the graph for the M48. The thing with the M48/M60 is that they are "identical", so you should have a control group e.g. if expt. damage for those two tanks differs by 50-100 points or more, there is something wrong - plain and simple.

Just looking at the graph above makes me think the lower end is pretty thin, when it comes to data points.

 

But I m not so sure when it comes to tanks like 907 or Fail-Löwe, at least the Fail-Löwe has no equal among the regular tanks. Not sure how the 907 plays compared to the other russian t9-10 meds.

And with skill scaling unsolved, you could "potentially" have an other SPG or LT desaster there, just for a single tank and not a class.Ofc it wont be of the same magnitude in those cases.

 

As a said repeatedly, I m not a fan of blind faith in numbers.

Share this post


Link to post
Share on other sites

Of course, the scaling problem is still there with WN8. Maybe that will get accounted for (along with a few other things) if there ever is a WN9.

 

Here is the M48 graph. 

 

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

Share this post


Link to post
Share on other sites

When IS-5 comes on reg test, we can gauge which. Initially though my feelings are if the gun's d25t it's is-6, if it's bl-9 (or bl-9x) it's IS-3. 

Share this post


Link to post
Share on other sites

Its time for another update, and I have it almost ready to go, but I want to make sure that this update is good through 9.7..

 

Could someone please list the tankids for all the new tanks on test in 9.7, and also their tier and type? 

Share this post


Link to post
Share on other sites

Any tanks added except the French ones?

Haven't really followed the updates lately.

 

 

Edit:

French Tanks:

Name                  ID     Tier  Type
FCM 36                15169  2     LT
Renault R35           15937  2     LT
Somua S35             13121  3     MT
SARL 42               14913  4     MT
Renault G1            4417   5     MT
AMX 30 1er prototype  15681  9     MT
AMX 30 B              15425  10    MT

Share this post


Link to post
Share on other sites

Values update version 19 is done, the excel file showing all the changes is here

 

All the tanks coming in 9.7 have been added

 

Some of the tanks that came in 9.6 were so new that I didnt have enough data to accept the change hat the system calculated, so those tanks (T55A, T28 Concept, etc) were left alone for now

 

I've sent the values to Phalynx to update wnefficiency.net, and all the websites should flow them from there

Share this post


Link to post
Share on other sites

I was looking over some of the values (including the changes, which I'm mostly happy with!) and saw that some values are quite nonsense, at least to me. For example, take the E8 vs Fury. The E8 has an expected dpg of 695, while the Fury has an expected damage of 740, which is a sizable difference. There is no real reason the Fury should have a higher expected value, the only reason it'd ever get that is because of flawed data. The same goes for the type 62 vs wz-131 (one is vastly better, yet only 10 more expected damage) and fv4005 vs 215b 183 (one is probably the worst t10, yet has some of the highest expected damage in the game). I also noticed the STA-2 has a higher expected than the STA-1 (even if extremely slightly), which also seems wrong, considering it's mostly the same exact tank except slower and with a significantly worse gun.

 

It seems like the values start getting funky when the user population is significantly different from the norm. Is this a problem, or is it just an accepted issue with WN8?

Share this post


Link to post
Share on other sites

I was looking over some of the values (including the changes, which I'm mostly happy with!) and saw that some values are quite nonsense, at least to me. For example, take the E8 vs Fury. The E8 has an expected dpg of 695, while the Fury has an expected damage of 740, which is a sizable difference. There is no real reason the Fury should have a higher expected value, the only reason it'd ever get that is because of flawed data. The same goes for the type 62 vs wz-131 (one is vastly better, yet only 10 more expected damage) and fv4005 vs 215b 183 (one is probably the worst t10, yet has some of the highest expected damage in the game). I also noticed the STA-2 has a higher expected than the STA-1 (even if extremely slightly), which also seems wrong, considering it's mostly the same exact tank except slower and with a significantly worse gun.

 

It seems like the values start getting funky when the user population is significantly different from the norm. Is this a problem, or is it just an accepted issue with WN8?

This has been repeatedly brought out, and repeatedly ignored.

Share this post


Link to post
Share on other sites

Its an issue with WN8, because the values are not derived directly from the tank, but from the results of the players playing the tanks. It gets bad when you look at tanks that arent available to all - M60 vs M48, Obj 907 vs T-62a. On the other hand, the current system does compensate for differences in tanks that should be about the same but arent - pick any 2 tier ten TDs.

 

To solve this problem I would like to move (one day) to a new system of expected values where there is one set of values for each type and tier: so, for example, all tier ten mediums would have same values. I've calculated the values and they are ready to go anytime, but the community and my WN8 peers are worried about change. If we did go to averaged expected values, some tanks would change a lot, some a little, but the good news is that they would never change again. Pros and cons of this approach are as follows:

 

Pros:

Values don't change so I dont have to keep recalculating them

No need to constantly research values for new tanks

Players' WN8 scores dont change every 3 months

No inexplicable differences in values between tanks that are basically the same as far as MM is concerned

Website administration of WN8 system gets a little easier as its static

 

Cons:

Value change for some tanks will be bigger than normal when average values first introduced

Players WN8 will change more than normal when average values first introduced

If WG doesnt balance their tanks it will show

Some tanks will be easier to pad than others

Some tanks wont get played much, others will get played a lot

 

Still, we are in no hurry, at least while I'm still around

Share this post


Link to post
Share on other sites

> one set of values for each type and tier: so all tier ten mediums would have same values, for example

 

An obvious exception arises -- tier 8 mediums.  2k dpg in a T69/416 with MM weight X is quite different than 2k dpg in a .6X t44/etc.  This would also make some tanks have whacky values, especially those much better than their peers (walker).  Or simply those tanks most focused on damage dealing (50 120) as opposed to other roles (vk45).  Finally, this would probably incent padders to play low tiers where they are compared more to shitters instead of tier 10, where the average is a green.

 

It seems to me a three point system is better, and expected values based on actual bottom ~20, middle, top 20% or something.

Share this post


Link to post
Share on other sites

That seems like it would be a completely different rating system, Gryphon. The whole point of WN8 was to be per-tank I believe; changing it to average values makes it more along the lines of previous WNx versions, though including a per-class average would certainly make it more accurate. If you have the values could you post/pm them to me? Interested to see how much they would change. Might be able to put together a script that would compare differences in WN8 of players/vehicles.

Share this post


Link to post
Share on other sites

That seems like it would be a completely different rating system, Gryphon. The whole point of WN8 was to be per-tank I believe; changing it to average values makes it more along the lines of previous WNx versions, though including a per-class average would certainly make it more accurate. If you have the values could you post/pm them to me? Interested to see how much they would change. Might be able to put together a script that would compare differences in WN8 of players/vehicles.

 

That is here with v19 values in pink, averaged values in green, and the delta in the middle

Share this post


Link to post
Share on other sites

That is here with v19 values in pink, averaged values in green, and the delta in the middle

Thanks! Perusing through, a couple tanks get shafted pretty badly (Mutant 6, Jg 88 ) and most of the old light tanks have expected damage increase while the newer (and better) lights have their expected damage decrease. Could you create an average for these two groups?

 

I'll see if I can write a program to see how people's actual WN8 would change.

Share this post


Link to post
Share on other sites

I have PM'd Never, no response yet. 

 

Pretty much all the sites out there do a weekly pull from wnefficiency .net - xvm certainly does - so the list of those who need to be notified is very short.

Share this post


Link to post
Share on other sites

I have PM'd Never, no response yet. 

 

Pretty much all the sites out there do a weekly pull from wnefficiency .net - xvm certainly does - so the list of those who need to be notified is very short.

OK. I pm'd Never too (still had a PM from last time).

I know noobmeter didn't update last round, so if there's an autopilot, it's not working. Then again, neither is the pilot.

Share this post


Link to post
Share on other sites

I'll have that program done tomorrow, interested to see how much people's overall WN8 will change with the different expecteds. Could you update the excel with the missing tank please? I needs it for program ^.^

Share this post


Link to post
Share on other sites

This is what I am using:

 

593 M2 (UK) 1 2 1_2 uk 593 1.61 278.45 2.41 1.58 60.89 -0.01 -4.44 -0.61 0.00 -2.69 593 1.60 274.01 1.80 1.58 58.20

Share this post


Link to post
Share on other sites

as per all the other discussions going on:

To what extent can we conservatively group tanks, and then treat each group as a unit for calculating expected values? Lumping the tier X SU MTs is not terribly controversial (though not without caveats). Mixing Soviet tanks and their Chinese copies brings up the question of how well WG compensated for gun depression.

I know something similar (class/tier) is planned for WN9; what happens if we start with the restricted case of nearly identical tanks here in wn8?

A few groups I'd like to see numbers for, if it's feasible and others agree that it's worth checking:

121 china Medium Tanks 10 0.92 1763.51 1.40 0.67 48.16

Object 140 ussr Medium Tanks 10 1.01 1838.23 1.66 0.89 49.62

Object 430 ussr Medium Tanks 10 1.00 1793.24 1.63 0.96 50.32

Object 907 ussr Medium Tanks 10 1.19 2019.94 1.63 1.09 52.98

T-62A ussr Medium Tanks 10 0.93 1704.46 1.60 0.77 48.64

M48A1 Patton usa Medium Tanks 10 0.88 1721.88 1.56 0.70 47.96

M60 usa Medium Tanks 10 0.80 1639.44 1.46 0.71 46.49

Panther II germany Medium Tanks 8 0.81 1080.04 1.31 0.88 51.96

Panther mit 8,8 cm L/71 germany Medium Tanks 8 0.91 1201.63 1.11 1.18 50.98

T-34-85 ussr Medium Tanks 6 1.01 736.47 1.44 0.96 55.13

T-34-85M ussr Medium Tanks 6 1.07 746.22 1.33 1.00 56.29

Type 58 china Medium Tanks 6 0.99 697.96 1.58 0.96 53.06

M4A3E8 Fury usa Medium Tanks 6 1.03 740.20 1.36 1.03 54.66

M4A3E8 Sherman usa Medium Tanks 6 1.01 694.97 1.53 1.00 55.12

M4 Sherman usa Medium Tanks 5 1.38 645.32 1.53 1.03 57.78

Sherman III uk Medium Tanks 5 1.31 616.04 1.55 1.31 58.56

Share this post


Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.

×
×
  • Create New...