bjshnog

⟪WN8⟫ Development / Resources

2,170 posts in this topic

I redid some analysis (by the way would love to see some analysis and conclusions from someone else), first off filtered players with less than 50 games on a tank, then did 1% top and 95% bottom filtering using WN7, BUT, set the top filtering minimum to 10 players, and the minimum amount of players analyzed per tank to 15.

 

Now analyzing the results... remember these are the top 5% players on each tank, so this is probably very close to measuring a true tank´s effectiveness (ie skill cieling). Therefore this is truly interesting data.

 

Table per-tier and per-class:

 

aj64.jpg

 

 

Some conclusions:

1-First off, as Folterknetch had foreseen, for top players, arty just has a lower cieling than tanks. 

2-TDs are top for WN7, as would be expected for a metric heavily weighed on damage and frags

3-Heavies are pretty close behind TDs in WN7.

4-Tier 4 tanks are TERRIBLE.

5-Lights are given the shaft by WN7, obviously. Specially from tier 3 to 6. While WN8 may not be able to effectively capture scout player skill from the available stats, we can sure as heck make sure they don´t drag a player´s stats down, like they do in WN7.

6-While TDs, SPGs and heavys are pretty balanced throughout the tiers, meds just get better as you go up in tiers...

 

Share this post


Link to post
Share on other sites

I redid some analysis (by the way would love to see some analysis and conclusions from someone else), first off filtered players with less than 50 games on a tank, then did 1% top and 95% bottom filtering using WN7, BUT, set the top filtering minimum to 10 players, and the minimum amount of players analyzed per tank to 15.

 

Now analyzing the results... remember these are the top 5% players on each tank, so this is probably very close to measuring a true tank´s effectiveness (ie skill cieling). Therefore this is truly interesting data.

 

Table per-tier and per-class:

 

 

Great job.My initial thought is I think this is going to be as good as it gets. Until great tankers start playing underperforming tanks to pad per-tank-wn8. Maybe? Will that force a change in the per-tank-wn8 coefficients in the future?

Share this post


Link to post
Share on other sites

Maybe... yeah... the thing is though, some tanks are just BAD BAD BAD. And you can´t get great stats on them no matter how well you play! :D

 

However, the effect you mentiopn definitely exists... more good players play OP tanks and less good players play bad tanks. This is also pretty clear from looking at the data.

 

 

I am thinking we can soften up the top tank stats and raise the bar a little bit on the absolutely trash tanks. This will prevent normal Joe from having his WN trashed by playing a Hotchkiss, and also prevent sealclubbers from jumping into an A-20, doing 1 frag per game and raising their WN8. :D :D :D

 

 

I am devising a formula that will adjust all scores, moving them slightly towards the middle, which should counteract this effect that we are talking about.

 

Another problem is that whenever new tanks come out, WN8 calculation will be trashed if we only use a per-tank stat line. I was thinking we could use a per-tier, or per-class stat line, or even per-class-per-tier statline like I posted above. and have each tank have modifiers to those modifiers. So when a new tank comes out and there is not enough data, the tank´s modifier is 0, so it will use the per-class-per-tier modifier. Thoughts?

Share this post


Link to post
Share on other sites

I am just thinking instead of using a multiplier to take the extremes closer to the median, I will analyze more closely the distribution for those tanks, maybe choosing different cutoff %s for those tanks will be enough to get data that is more closely related to the data for other tanks.

Share this post


Link to post
Share on other sites

edit - figured it out, top 5%

 

@Praetor:

 

 

 

The def number for artis are "to high" beside to t9, which fits. My explanation - if these number are from the top 5% players - what we see here is a combination of different things - but NOT random matches (platooned or not):

 

- excessive TCs and CW (and dont forget the last campaign), there SPGs get high Def values.

- low and mid tier arti players get way more Def because of the mobility, but that will drop. There SPGs also got nerfed afaik. Gun depression in low and mid tiers was also nerfed was SPGs

- t9 artis which are former t7s ("no" TCs/CWs) are within what I would expect from a good arti player

 

 

Spot with 0.1 overall fits on the other hand.

 

A while back I took a look at the data from the XVM-guys. The avg player in arti over all tiers had, as far as I remember, around 0.1 spots/game and and around 0.8 def/game.

Share this post


Link to post
Share on other sites

@Praetor:

 

pls clearify

 

"... first off filtered players with less than 50 games on a tank, then did 1% top and 95% bottom filtering using WN7, BUT, set the top filtering minimum to 10 players, and the minimum amount of players analyzed per tank to 15."

 

vs.

 

"Now analyzing the results... remember these are the top 5% players on each tank"

 

for me that is contradicting.

 

Are that the top 5% or avg tanking tomato table I m looking at here?

 

[99%-100%] <- filtered this out

[95%-98.9%] <- kept this

[0%-94.9%] <- filtered this out

 

At least, that's how i read it.

 

@Praetor

 

What is VBA-2.csv? The individual VBA's MB add up more than VBA-2.csv.

Share this post


Link to post
Share on other sites

Good progress is being made, I see. When we check the correlations, I suggest that we compare win rate to stat at each average battle tier, rather than total expected top stats and such. That way, we can see how each stat correlates to win rate at each battle tier and also probably make it even more accurate. :D

 

I think it might be getting excessive. :P

Share this post


Link to post
Share on other sites

[99%-100%] <- filtered this out

[95%-98.9%] <- kept this

[0%-94.9%] <- filtered this out

 

At least, that's how i read it.

 

@Praetor

 

What is VBA-2.csv? The individual VBA's MB add up more than VBA-2.csv.

 

What is MB? VBA-2 as compared to the single csvs, has simply been filtered for players with at least 50 battles on each vehicle. Also, the stats have been transformed to averages per battle.

 

edit - figured it out, top 5%

 

@Praetor:

 

 

 

The def number for artis are "to high" beside to t9, which fits. My explanation - if these number are from the top 5% players - what we see here is a combination of different things - but NOT random matches (platooned or not):

 

- excessive TCs and CW (and dont forget the last campaign), there SPGs get high Def values.

- low and mid tier arti players get way more Def because of the mobility, but that will drop. There SPGs also got nerfed afaik. Gun depression in low and mid tiers was also nerfed was SPGs

- t9 artis which are former t7s ("no" TCs/CWs) are within what I would expect from a good arti player

 

 

Spot with 0.1 overall fits on the other hand.

 

A while back I took a look at the data from the XVM-guys. The avg player in arti over all tiers had, as far as I remember, around 0.1 spots/game and and around 0.8 def/game.

 

Def numbers also look too high to me... but for all tanks, not just SPGs. Anyways, I don´t think it is worthwile to normalize defense, and maybe not even winrate and spots PER-TANK. Let me explain why:

 

1-Defense represents a minuscule part of the total WN score, even moreso if we reduce cap to 1.4 like planned, and weight from 100 to 85 per point.

2-I am not sure about the validity of the defense/cap numbers they just seem too high.

3-For winrate, you can take a look and see that it doesn´t vary much from tank to tank, variation is MUCH MUCH higher for platoons/non-platoons and TC/non-TC, so it really doesnt make much sense. Also, remember we use it as a proxy for intangibles, not for wins per se.

4-For spots, using per-tank values would reduce the score of players who play a lot of scouts. That wouldnt make much sense. Albeit, not normalizing spots still allows spot farming at low tiers... Hmmm...

 

So I think overall, I am in favor of simplifying things, and only use the EXPECTED per-tank formula for damage and frags. They are 80% of WN score, so I think that is where we should focus. Also, the VBA numbers look good IMHO for those stats, for each tank.

 

Regarding the damages, and frags from the table, keep in mind this is top 5% of the players who upload their stats to vbaddict. WN7 pretty much fails at properly measuring each tank, I think these kind of numbers (for example, for tier 10, 1.78 frags and 3235 damage, would represent probably around 2000-2100 WN8 (personal estimation).

Share this post


Link to post
Share on other sites

I am reconsidering some of the filtering, but its mainly a very subjective process, 1% off the top is too big though...I am reducing it.

 

To make this a more statistically solid process, I have asked Phalynx for player data (overall). I will use that data to comb through the filtering process better. What I will do is a ratio of WN7 on each tank to overall WN7. That should allow me to properly balance the filtering for each tank. For example, lets say A-20 has WN7 of 900 using the 95-99%, and the players who make up that portion of the playerbase have an overall WN7 of 1200 (ratio = 0.75)

If for pz38 NA the 95-99% WN7 is 1000, but the overall WN7 for those players is 1800 (ratio 0.55), then we are not comparing apples to apples.

 

Low ratios will suggest to filter out less top players, whille high ratios suggest more top player filtering is needed.

Share this post


Link to post
Share on other sites

I just had a better idea. Instead of filtering according to a % of the players that play that tank, I will reduce the top players filter, since it is only there for leaving out vbaddict errors and completely wild outliers I will take out the top3 players for each tank. Then, I will choose the number of players to analyze for each tank based on the difference of WN7 to the top 4 player (previously normalized of course).... brilliant! :D

Share this post


Link to post
Share on other sites

Okay analysis finished. I think it looks much better now, dont have time to make the pictures, gotta go now, but I did make a spreadsheet and put it on Dropbox. Per-tank, per-tier and per-tier-per class analysis.

 

Please need some feedback. Does anyhting look off? Keep the feedback coming please!

 

https://www.dropbox.com/s/g2cfyu9bg1wmu9e/Analysis1.xlsx

 

 

PS: What I did was use top cutoff of 3, chose number of players analyzed per tank according to WN variation from the top4 player. So I analyzed players with WN frm top 4 down to 75% of top4 wn score. This is for every tank. Then, I had to change the top cutoff on some tanks that were getting too few tanks analyzed, which just means that there were more outliers on the top (maybe pure gold flinging, TCs, etc. etc.).

 

PS2: I really need as many people as possible to take a look at this data and give it the "eye-test". I would really appreciate the contribution of anyone who can take a few minutes to look at the tables.

Share this post


Link to post
Share on other sites

Disclaimer: I dont have Excel so I imported the file to Google Spreadsheet so the problem could be on my end.

 

First, I dont see the per tank analysis. I see a per-tank tab and a per-tier tab. Both tabs show the same page showing per-class and per-tier analysis.

 

6ues1D4.jpg

rH9y8Un.jpg

 

Second, the numbers in the per-class heavy kill/death column are too large.

 

vfSqWf9.jpg

 

That's all I see so far on a quick first lookthrough.

Share this post


Link to post
Share on other sites

Second, the numbers in the per-class heavy kill/death column are too large.

 

vfSqWf9.jpg

 

That's all I see so far on a quick first lookthrough.

 

Yeah, in excel, it is only per tier. Also, the KILL/DEATH looks like a non-rounded DAMREC value.

Share this post


Link to post
Share on other sites

only use damage and frags, the other numbers make a wired impression in some cases.

 

btw - those numbers are from vbaddict. Could you ask that guy who is running it, who is mainly providing the data. Is it more a NA, EU or RU thing (numbers or % perhaps)?

 

 

Share this post


Link to post
Share on other sites

Where or how do I get the per vehicle type and tier stats to fill in that sheet?

Share this post


Link to post
Share on other sites

Fixed the spreadsheet.

 

In per-tier, it still has non-rounded DAMREC values in place of KILL/DEATH.

Share this post


Link to post
Share on other sites

What exactly is DAMREC?

 

Also, the top percentage needs to be much tighter at lower tiers. If it's top 25% at tier 10, then it should be about top 2% at tier 1.

Share this post


Link to post
Share on other sites

What exactly is DAMREC?

 

Also, the top percentage needs to be much tighter at lower tiers. If it's top 25% at tier 10, then it should be about top 2% at tier 1.

 

DAMREC is damage received.

Share this post


Link to post
Share on other sites

DAMREC is damage received.

 

I see.

 

On another note, I said previously that you would get quite a big boost in WN8 based on your winrate and the fact that you mainly play light tanks. I only just saw your average tier. :0

Share this post


Link to post
Share on other sites

Hmmm, yeah maybe tier 1 stats are a bit low, and the percentage of players being measured could be reanalyzed. Will look at that again, and see what I can come up with...

Share this post


Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.