Jump to content
RichardNixon

WN9 candidate prototype

Recommended Posts

14 hours ago, GotchMaster said:

I can't argue against the data as the majority of the data comes from players playing the tiers they are comfortable/enjoy playing.  The best players typically don't go back down tiers to club low tier shitters all day.  They gravitate to the higher tiers as that is where most of the talent is.  This is why green 50% T6 players go to T10s and turn into red 44%ers.

Difficulty is actually orthogonal to skill scaling: It's a lot easier for (almost[1]) everyone to win games at tier 5 than tier 10, but if you take a terrible tier 10 player and a good tier 10 player and put them in tier 5 tanks, the relative difference in their damage output is unchanged. There's an intuitive idea that a good player would hit some sort of performance ceiling against weak players, but it's not actually true.

Differences in tier (and tank) difficulty are handled by the expected values. The tier 10 average expected value is equal to the tier 10 average performance, but at tier 5 it's 75% higher.

[1] Extremely strong players have little winrate variation with tier. This is because the negative effect of playing against better players is countered by the higher matchmaking influence at higher tiers.

 

14 hours ago, GotchMaster said:

I can't see the correlation here aside from the fact that good players typically pay attention so would naturally be good at an easy to play class(arty).  But players who can perform in arty are not necessarily good tankers, which is especially evident at T10.

I'm guessing that you're talking about winrates again. A terrible player will get much better winrates in arty than other classes at tier 10, but that's true even if they're similarly skilled at all classes. At tier 10, arty is a low skill scaling class and a low-influence class, so the winrate spread is much narrower than for other tanks. It's the same reason that even the best arty players struggle to top 60% solo winrate at tier 10.

The correlation's probably down to mechanical skill having weak influence even in proper tanks. Map knowledge is somewhat transferrable, and tactical decision making is mostly down to learning ability.

On the other hand, there are very few players with a high proportion of their battles in arty, so it's possible that the correlation only operates in one direction: Players who are good at other tanks are generally good at arty, but not vice versa.

 

14 hours ago, GotchMaster said:

If you used aggressive tier cap scaling like I mentioned above would it turn out any differently?  As in using the best 65%(or whatever % of games you choose) plus removing most of the clubbers game with caps, valuing T10s the most, etc.

No, because most clubbers either perform similarly at lower and higher tiers, or play so few higher tier games that the adjustment would be minimal. For example, capping total tier 1-6 games to 5k wouldn't make much difference if a player never played more than 1k battles in tier 7+ tanks. Capping and weighting are generally soft corrections, even when applied aggressively.

 

15 hours ago, GotchMaster said:

Take td42 as an example "arty god", he did well even in WN8.  That account is arty only and has an account WN9 of 1373.1, which beats the top NA re-roll accounts by a good margin.

TD42's all historical data problems (the CGC results are acceptable), although still interesting. According to my records, the M53/55 has never been nerfed and the top 100 RU players don't come close to TD42's results in it, so there's something odd going on there. It looks like there was a strong advantage for arty on NA for a limited period. Maybe an MM issue caused by low tank counts, or something to do with the old camping meta: Arty perform much better when the games are longer.

That sort of issue is a good argument for stripping arty from account WN9, but I'm not sure that the practical consequences are positive. Might be a polling issue.

 

2 hours ago, GotchMaster said:

There are plenty of players who can and will pad with arty using the new WN9, more so than they could otherwise, and have stats to back that up.

From what I've seen, active arty padding in WN9 is only possible by time travelling. The 261 appears to be a minor exception, probably because most people play it with the wrong ammo. Some of the French mid tier HTs and TDs are likely to have similar problems.

 

Link to post
Share on other sites

Please do not be swayed by appeals to emotion. You've got the metric pretty well sorted and now it's just people arguing to have their pets included.

Link to post
Share on other sites
53 minutes ago, RichardNixon said:

Difficulty is actually orthogonal to skill scaling: It's a lot easier for (almost[1]) everyone to win games at tier 5 than tier 10, but if you take a terrible tier 10 player and a good tier 10 player and put them in tier 5 tanks, the relative difference in their damage output is unchanged. There's an intuitive idea that a good player would hit some sort of performance ceiling against weak players, but it's not actually true.

Differences in tier (and tank) difficulty are handled by the expected values. The tier 10 average expected value is equal to the tier 10 average performance, but at tier 5 it's 75% higher.

[1] Extremely strong players have little winrate variation with tier. This is because the negative effect of playing against better players is countered by the higher matchmaking influence at higher tiers.

I was referring more to the fact that the data you have is flooded with average players making themselves look good or be relevant(as they don't have the talent for 10s).  The players who are really good at T10s or high tiers really don't contribute to the data pool for the lower tiers after the initial learning/grind.  Because the data is based off the majority(average slob), the value of T10 WNx has always been skewed.   Take a really good re-roll for example.  These players specifically play for stats just like a lot of the population but the difference is that they are past the learning stages but now dropping down tiers to pad really hard.  Sure they play T10s and can pull great numbers, but almost all of them pad using lower tiered tanks(lower than T10) for amazing numbers.  If I personally went down to lower tiers to pad(soloing and throwing win-rate to the curb) it would almost always be higher than anything I could pull in T10s.  I do want to mention that this iteration of WNx does scale much better with 10s than the others, particularly with the TDs.

I can see the win-rate point as long as we are talking specifically about soloing which I think we are.  Platooning and tank selection certainly throws a wrench into it all.

53 minutes ago, RichardNixon said:

I'm guessing that you're talking about winrates again. A terrible player will get much better winrates in arty than other classes at tier 10, but that's true even if they're similarly skilled at all classes. At tier 10, arty is a low skill scaling class and a low-influence class, so the winrate spread is much narrower than for other tanks. It's the same reason that even the best arty players struggle to top 60% solo winrate at tier 10.

The correlation's probably down to mechanical skill having weak influence even in proper tanks. Map knowledge is somewhat transferrable, and tactical decision making is mostly down to learning ability.

On the other hand, there are very few players with a high proportion of their battles in arty, so it's possible that the correlation only operates in one direction: Players who are good at other tanks are generally good at arty, but not vice versa.

Was specifically referring to general competence or proficiency for anyone playing arty.  That yes the only correlation I could find was that talented tankers(any class but arty) are generally able to perform in arty but the opposite was not necessarily true.

54 minutes ago, RichardNixon said:

No, because most clubbers either perform similarly at lower and higher tiers, or play so few higher tier games that the adjustment would be minimal. For example, capping total tier 1-6 games to 5k wouldn't make much difference if a player never played more than 1k battles in tier 7+ tanks. Capping and weighting are generally soft corrections, even when applied aggressively.

I should have been more specific and thrown out numbers so an aggressive capping would be very harsh on lower tier games.  I was also under the impression it would be scaling depending on how many games a person has.  More along the lines of if a person has 10k total battles then only 1k(10%) low-tier(1-5 or whatever) battles could count.  That also would vary if you are doing it by total battles or per tank.  Those numbers are super raw and just an example.  Going this route would make it very hard for low tier padders to accumulate account WN9.  WG has a similar strategy for PR just involving damage which stagnates the lower tier clubber rating.

55 minutes ago, RichardNixon said:

From what I've seen, active arty padding in WN9 is only possible by time travelling. The 261 appears to be a minor exception, probably because most people play it with the wrong ammo. Some of the French mid tier HTs and TDs are likely to have similar problems.

Have to disagree on this one.  My rudimentary peek at spencer's damage site showed otherwise.

26 minutes ago, StormCrowReaperManyHats said:

Please do not be swayed by appeals to emotion. You've got the metric pretty well sorted and now it's just people arguing to have their pets included.

Yes that is it.  My extremely emotional thoughts have gotten the best of me and spilled right into this thread.  I certainly stand to gain a tremendous amount by swaying all that listen and getting my way with everything WN9 has to offer.  You caught me.

Link to post
Share on other sites
3 hours ago, GotchMaster said:

I was referring more to the fact that the data you have is flooded with average players making themselves look good or be relevant(as they don't have the talent for 10s).  The players who are really good at T10s or high tiers really don't contribute to the data pool for the lower tiers after the initial learning/grind.  Because the data is based off the majority(average slob), the value of T10 WNx has always been skewed.

No, expected values aren't averages (this is basic stuff and shouldn't still be coming up), so even the WN8 expected value method didn't have that problem. You can actually filter the data down to good players who play both tiers regularly and the tier scaling is the same. That's a primitive version of what the method actually does.

There were some other flaws with the WN8 expected value method that did make tiers 5-9 easier to pad: Recency bias, stock bias and crew skill bias. The first two are fixed in WN9, and crew skill bias is a compromise. Hence the tier 5 expected values are ~15% higher than WN8.

Crew skill bias is the remaining problem for low tier padding, but the potential advantage is quite small, even at tier 5: Roughly 5% at 400k earned XP, or 10% at 1200 XP. This can be adjusted for directly (and it is for expected value generation), but the adjustment can't work with recent WN9, and it doesn't correct for high tier -> low tier crew-retraining strategies, so it's probably not worth the code or the extra data column. The caps are a simpler alternative with much the same effect.

 

3 hours ago, GotchMaster said:

I should have been more specific and thrown out numbers so an aggressive capping would be very harsh on lower tier games.  I was also under the impression it would be scaling depending on how many games a person has.  More along the lines of if a person has 10k total battles then only 1k(10%) low-tier(1-5 or whatever) battles could count.  That also would vary if you are doing it by total battles or per tank.  Those numbers are super raw and just an example.  Going this route would make it very hard for low tier padders to accumulate account WN9. WG has a similar strategy for PR just involving damage which stagnates the lower tier clubber rating.

Note that even the harshest capping or weighting doesn't touch people who don't play higher tiers. Capping's only useful against people who play a mix, and setting the tier 5 cap that low is brutal on f2p players. I can accept 300 battles per tank at tier 5, but not much lower. At that level, high-tier -> low-tier crew retraining won't be very effective.

WG-PR punishes clubbing directly instead: If you're a sufficiently good player, you'll get less PR from a low tier tank than a high tier tank. The equivalent for WN9 would be increasing the wn9scale values at lower tiers, but it's a value judgement and trashes half the point of the metric.

There are other more complex adjustment methods, but I doubt the sealclubbing problem is worth it. WN8 rewards sealclubbing by 15% on top of crew retraining strategies, and sealclubbing is still pretty rare. Most of the players with a high battle count at low tiers are just too bad to handle the higher tiers.
 

3 hours ago, GotchMaster said:

Have to disagree on this one.  My rudimentary peek at spencer's damage site showed otherwise.

As the question is current padding potential, you want to look at recent DPG, not 150 battle.

I'm not seeing anything too worrying, although there may be a more general issue with AP arty. I'll look into it. May notch up the skill scaling a bit if I can find an excuse.

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

No, expected values aren't averages (this is basic stuff and shouldn't still be coming up), so even the WN8 expected value method didn't have that problem. You can actually filter the data down to good players who play both tiers regularly and the tier scaling is the same. That's a primitive version of what the method actually does.

There were some other flaws with the WN8 expected value method that did make tiers 5-9 easier to pad: Recency bias, stock bias and crew skill bias. The first two are fixed in WN9, and crew skill bias is a compromise. Hence the tier 5 expected values are ~15% higher than WN8.

I wasn't saying the expected values are averages.  I was saying that the data that is being used is tainted/misinterpreted and for that reason it has always been easier to pad lower tiers.  It looks better in WN9 but the problem still exists.  I am not offering definitive technical advice to solve the problem, just pointing it out.  I mentioned rerolls as they are a good example without sifting through accounts that have been through the ringer.  The best tanks for WNx are almost always below T10 and remain so.

4 hours ago, RichardNixon said:

Note that even the harshest capping or weighting doesn't touch people who don't play higher tiers. Capping's only useful against people who play a mix, and setting the tier 5 cap that low is brutal on f2p players. I can accept 300 battles per tank at tier 5, but not much lower. At that level, high-tier -> low-tier crew retraining won't be very effective.

WG-PR punishes clubbing directly instead: If you're a sufficiently good player, you'll get less PR from a low tier tank than a high tier tank. The equivalent for WN9 would be increasing the wn9scale values at lower tiers, but it's a value judgement and trashes half the point of the metric.

There are other more complex adjustment methods, but I doubt the sealclubbing problem is worth it. WN8 rewards sealclubbing by 15% on top of crew retraining strategies, and sealclubbing is still pretty rare. Most of the players with a high battle count at low tiers are just too bad to handle the higher tiers.

I'm not sure why you are trying to spare the f2p players when it comes to a skill metric.  They still have the low WG PR and in game stats just the same.   Isn't it just by the nature of f2p that they really can't compete optimally anyway?

 

4 hours ago, RichardNixon said:

As the question is current padding potential, you want to look at recent DPG, not 150 battle.

I'm not seeing anything too worrying, although there may be a more general issue with AP arty. I'll look into it. May notch up the skill scaling a bit if I can find an excuse.

Yep of course looking at recent DPG not 150b.

Link to post
Share on other sites

Comparing my 'padded' WN8 with my current WN9 I would say WN9 is more in agreement with my actual skill level. So please push for implementation so I can start padding my WN9.

Link to post
Share on other sites

If my opinion is worth anything, I've read through pretty much everything you published regarding WN9, how it works, its limitations, etc. and I think you have reached a very nice point.

The only aspect which I think is still somewhat debatable is the influence of artillery on the metric. As pointed out by a few above, perhaps being a good player in artillery allows for a higher WN9 ceiling than being good in tanks, and as such, one could maybe "pad" in artillery more easily than in other classes. I suppose this is due to the fact that proportionally much fewer players are good at arty than good at tanks, which creates very low "scaling" values for them. Perhaps a simple solution would be to lock the *scale values for artillery at 1.0 (which is, if I understand correctly, "average").

Whatever you decide to do about artillery, I think WN9 is already the best metric available at the moment, and as such, should be implemented.

Link to post
Share on other sites
17 hours ago, GotchMaster said:

I wasn't saying the expected values are averages.  I was saying that the data that is being used is tainted/misinterpreted and for that reason it has always been easier to pad lower tiers.  It looks better in WN9 but the problem still exists.

It's not a general problem. This chart compares experienced players with 200+ recent battles each in tier 5 and tier 10. Tier 5 XP range is cut to 50-100k for comparison:

wn9 count low10 low25 median high25 high10
200 31 0.725 0.866 0.941 1.093 1.181
300 215 0.791 0.899 0.994 1.09 1.202
400 699 0.803 0.894 0.998 1.081 1.168
500 898 0.819 0.901 0.981 1.064 1.149
600 570 0.844 0.924 1.004 1.079 1.169
700 174 0.873 0.933 1.002 1.069 1.159
800 34 0.873 0.929 0.984 1.106 1.184

The medians are very close to 1.0 with a 75k XP midpoint, so it's working as intended. However, the variance here is quite wide, and that's not primarily down to crew skills. There will be some selective ammo/food spending, but some players may simply be ~10% better at tier 5 (or at least, the tanks they're playing there), while others are 10% better at tier 10. It's not a problem unless you insist that only tier 10 skill matters.

Note that if you use dpgwhores as your primary source then you'll get selection bias. Players who perform extremely well in a tank are likely to be genuinely better and/or throwing credits at it, in addition to exploiting crew skills. Other players with similar general skill will do nothing special in the tank.

As far I can tell, the main issue at tier 5 is players with thousands of battles in a single tank (usually T67). That's fixed with a moderate cap. I'll probably switch the cap to something like tier*40 + tier*tier*totalbat/2000, which reduces everything below tier 10.

 

17 hours ago, GotchMaster said:

I'm not sure why you are trying to spare the f2p players when it comes to a skill metric.

Because I want a more accurate representation of their skill, or at least performance. If the metric throws away most of their elited battles in favour of their grind battles then it won't do that.

Class padding analysis to follow. Post was getting too long.

Link to post
Share on other sites

Here's a padding comparison for arty and other classes. Each table is filtered to players with 10-30% of their recent battles in that class, but with no XP limits:

https://docs.google.com/spreadsheets/d/1VNpJjJCqQfcRqafe95YzNfF0InEmKjkokyBY2L5dnRA/edit?usp=sharing

Observations:

1. Meds, heavies and TDs have their median very close to 1.0 for 300+ WN9 players, which is ideal. Variance is fairly low, so apparently skill in these classes correlates well.

2. SPGs have a low median (around 0.9) at all skill levels, so most players are anti-padding in their SPGs, as intended. Variance is higher (which means weaker correlation with other classes), but the top 10% in each skill band are still padding less WN9 in arty than the top 10% in other classes. The variance distribution is near symmetrical, so bad players are often even worse at arty.

3. LTs have a high-ish median (around 1.04) for 300+ WN9 players, and so most players are padding in them. This is probably because having >10% of your battles in LTs is relatively unusual, so the filter selects the better LT players. The variance for LTs is also on the high side, so a fair proportion of LT players are padding significantly.

4. Sub-300 WN9 players don't fit well. This is probably because this test includes stock grinds, which are typically a lot tougher for heavies, and sub-300 WN9 players are too dumb to use free XP correctly. Alternatively the linearity assumption may not hold at the bottom end, but it doesn't matter much either way if it's not skewing the rest of the skill range.

5. Class variance is higher for weaker players, with a similar pattern for all classes. Weak players also have more playstyle variance, so that makes sense. Good players figure out the right way to play each class. Bad players pick something randomly.

If the primary concern is padding, LTs are currently more of a problem than SPGs, and should probably get +5% on their expected values. I'm still considering dumping SPGs from account WN9 for other reasons though.

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

Here's a padding comparison for arty and other classes. Each table is filtered to players with 10-30% of their recent battles in that class, but with no XP limits:

https://docs.google.com/spreadsheets/d/1VNpJjJCqQfcRqafe95YzNfF0InEmKjkokyBY2L5dnRA/edit?usp=sharing

Observations:

1. Meds, heavies and TDs have their median very close to 1.0 for 300+ WN9 players, which is ideal. Variance is fairly low, so apparently skill in these classes correlates well.

2. SPGs have a low median (around 0.9) at all skill levels, so most players are anti-padding in their SPGs, as intended. Variance is higher (which means weaker correlation with other classes), but the top 10% in each skill band are still padding less WN9 in arty than the top 10% in other classes. The variance distribution is near symmetrical, so bad players are often even worse at arty.

3. LTs have a high-ish median (around 1.04) for 300+ WN9 players, and so most players are padding in them. This is probably because having >10% of your battles in LTs is relatively unusual, so the filter selects the better LT players. The variance for LTs is also on the high side, so a fair proportion of LT players are padding significantly.

4. Sub-300 WN9 players don't fit well. This is probably because this test includes stock grinds, which are typically a lot tougher for heavies, and sub-300 WN9 players are too dumb to use free XP correctly. Alternatively the linearity assumption may not hold at the bottom end, but it doesn't matter much either way if it's not skewing the rest of the skill range.

5. Class variance is higher for weaker players, with a similar pattern for all classes. Weak players also have more playstyle variance, so that makes sense. Good players figure out the right way to play each class. Bad players pick something randomly.

If the primary concern is padding, LTs are currently more of a problem than SPGs, and should probably get +5% on their expected values. I'm still considering dumping SPGs from account WN9 for other reasons though.

Thanks for looking into it.

I understand that most players "anti pad" in arty, i.e. they are performing worse in arty than in other classes (and get a lower WN9). 

This however is not the issue I forsee.

 

Specifically because of the above, scaling values for artillery are very low. This means that if a player were to be really good at artillery, he could get a very, very high WN9; higher in fact than by being really good in other tank classes.

 

As such, much like good players today are able to pad WN8 in a T-62A or an E50 because the vast majority of people suck at it, people would be able to pad WN9 in arty, because pretty much everyone sucks at it.

Right now, a large proportion of people good enough to pad WN9 in arty either don't play it at all (because hate or fear of falling WN8) or don't play it enough to actually get good at it (hence the "anti-pad" effect).

 

The concern is not with padding itself, as I couldn't care less. However, it stands to reason that there could be a significant chunk of morons who live only to pad WNx who spam HEAT T-54 or APCR E50 right now who would start playing artillery so they can become good enough to pad in it.

 

In other words, padding itself is not an issue, as demonstrated by the data you gathered. However, low expected values + very low scaling for artillery would likely be an incentive for more players to spam arty until they get better and are able to pad in it, which is just spreading the cancer.

 

I myself am not convinced this is an actual problem, because "arty padders" would be quite easy to spot, and would likely be called out by good clans and such, much the same way Hellcat padders were in the day.

However, I do think that something that could potentially increase artillery player population should be looked at carefully.

 

Removing artillery from WN9 is an ideal solution to counter this and other potential issues. However, as much as I can't stand the very concept of artillery, it seems a little unfair to leave them out completely.

Would it be complex / difficult to have a separate "WN Arty" be calculated on the same bases as WN9? I wouldn't advise on putting any emphasis on it, as I for one would be pretty damn disgusted to even have the word "Arty" in my signature.... However it could be available separately if it were ever relevant to know how someone fares in artillery alone.

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

Specifically because of the above, scaling values for artillery are very low. This means that if a player were to be really good at artillery, he could get a very, very high WN9; higher in fact than by being really good in other tank classes.

That's only true if every current arty player is doing it wrong (including those who play arty relatively well compared to their other tanks), and that there's actually some huge hidden potential. The far more likely explanation for the results is that arty is bad and scales badly with skill.

It's similar to the argument that damage metrics are worthless because you can get much higher (solo) winrates with much lower damage. It's conceptually possible, but there are no examples anywhere in the player base. You'd expect someone, somewhere to have taken the high road.

Note that arty is already manually adjusted. Expected values get increased by 10% (it's not a coincidence that the arty median comes out 10% low), because that's how much better arty was before personal missions. Skill scaling gets increased by 5% because of the small disconnect between arty skill and skill in other tanks, and that value appears to be in the right ballpark. If players do somehow figure how to pad arty then it's a trivial adjustment to the next expected values version, but I doubt it'll happen.

 

26 minutes ago, NThirtyTwo said:

Removing artillery from WN9 is an ideal solution to counter this and other potential issues. However, as much as I can't stand the very concept of artillery, it seems a little unfair to leave them out completely.

The only metric in question is account WN9. Recent WN9 has to include arty, and there's no reason not to provide per-tank values.

Removing arty from account WN9 would be a huge advantage for most current arty players. This isn't just because arty currently anti-pads, but because >50% of arty have historical nerfs, and account WN9 assumes historical maximums. Your options are limited to arty released after 8.6, excluding the FV304 and the CGC. Hence the trashy historical data is a pretty strong argument regardless of whether you love or hate arty.

Link to post
Share on other sites

Well then, I stand corrected.

The logic behind Arty's removal seems pretty solid then... Are you going to remove it? Or just launch WN9 as it is right now?

 

Link to post
Share on other sites
3 hours ago, NThirtyTwo said:

The logic behind Arty's removal seems pretty solid then... Are you going to remove it? Or just launch WN9 as it is right now?

I'm going to poll it, as it's difficult to weight the general-metric argument against the accuracy issues.

Removing arty brings up a second question. There are relatively few other tanks which were significantly nerfed, and most of those by only ~10%. Therefore it starts to make sense to use the current rather than historical maximum expected values for account WN9. That means players are free to play tanks like the T57 or 268 without murdering their stats, although it adds some potential future incentive for playing currently-overpowered tanks like the 50/51 or E5.

Link to post
Share on other sites
5 hours ago, NThirtyTwo said:

The logic behind Arty's removal seems pretty solid then... Are you going to remove it? Or just launch WN9 as it is right now?

48 minutes ago, RichardNixon said:

I'm going to poll it...

Sorry, but polling this here is like  "Do you want to pay taxes?"

 

Link to post
Share on other sites
Just now, Graf_Starhemberg said:

Sorry, but polling this here is like  "Do you want to pay taxes?"

If you asked it like that, then yes.

I'm more concerned that if I give people sufficient information, they'd vote to keep arty in the metric on the grounds that the current implementation usually punishes arty players.

Link to post
Share on other sites
On 7/22/2016 at 7:47 PM, RichardNixon said:

If you asked it like that, then yes.

I'm more concerned that if I give people sufficient information, they'd vote to keep arty in the metric on the grounds that the current implementation usually punishes arty players.

Current implementation also punishes players that play nerfed tanks (maxhist), doesnt it....?

Link to post
Share on other sites

Yeah ok, not going to poll the arty question. Bad data is bad data, so arty gets dropped from account WN9. Four fairly significant changes will be pushed shortly:

  1. Arty removed from account WN9 because the data sucks.
  2. Scout tanks get +3% exp and +5% scale to fit the profile of regular scout players.
  3. Account WN9 cap changed to tier*40 + tier*tier*totbat/2000 to reduce exploitation of lower tiers.
  4. Account WN9 switched to current rather than historical expected values.

Overall distribution of account WN9 is very similar, so the base multiplier doesn't need changing. Some players will do a bit better (players with a lot of arty and battles in nerfed tanks), and some will do a bit worse (players with a lot of light tanks and heavily padded mid tiers). Players who only play arty get N/A.

Effect of new cap:

  • 6k battles: 275 battles per tier 5 tank, 700 battles per tier 10 tank
  • 20k battles: 450 battles per tier 5 tank, 1400 battles per tier 10 tank

 

2 minutes ago, Vasth said:

Current implementation also punishes players that play nerfed tanks, doesnt it....?

Current implementation punishes players who play tanks after they were nerfed. New implementation rewards players who played tanks before they were nerfed, although limited by the cap: Players with 10k battles in the pre-nerf Hellcat won't benefit much.

Link to post
Share on other sites

Ok, pushed the changes to the site:

http://jaj22.org.uk/news.html

One additional change: I halved the weight of nerfed tanks in account WN9. That reduces the potential rewards from playing tanks when they're overpowered, without directly hurting players who play post-nerf tanks.

Link to post
Share on other sites
Quote

25/07/16

  • Account WN9 changes:
    • Changed weight cap to tier*(40 + tier*totalbat/2000).
    • Removed SPGs (from account WN9 only).
    • Current rather than historical expected values are now used.
    • Halved weight (and weight cap) of nerfed tanks.
    • Adjusted implementation, description & FAQ to match.
  • Expected values 0.4: Increased exp & scale of scout tanks.
  • Switched WN8 over to v27.
  • Changed WG-PR unicum equivalent on experimental stats page.
  • Added question about per-tank scale on FAQ.

One one hand, I like these changes and think they make the metric better. On the other, they put me dangerously close to purple and I really don't want that kind of attention in pubs.

Link to post
Share on other sites
5 hours ago, Animactus said:

One one hand, I like these changes and think they make the metric better. On the other, they put me dangerously close to purple and I really don't want that kind of attention in pubs.

I have reached the point of no return this weekend... I expect to start taking extra artillery fire starting next week or so when people's XVM update.

Also, everything @RichardNixon does increases my WN9 :O 2-3 points here, 5-6 there, 3-4 left, 4-5 right. I suppose being almost 100% non padded has its benefits!

 

I might be wrong, but I think this recent update pretty much takes care of the last bit that people were concerned about regarding WN9? Is it ready to go?

Link to post
Share on other sites
1 hour ago, Kuroialty said:

So this is no longer a universal metric?

Account WN9? You could argue that it wasn't universal anyway, as it only uses random battles. The choice is whether to generate a really bad number for the handful of arty-only players (6 on EU), or no number at all. They can still use recent WN9, which will give them a sensible result.

 

11 hours ago, Animactus said:

One one hand, I like these changes and think they make the metric better. On the other, they put me dangerously close to purple and I really don't want that kind of attention in pubs.

I wouldn't worry. No-one will use it anyway :music:

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...