bjshnog

⟪WN8⟫ Development / Resources

2,170 posts in this topic

I have re-analyzed the data, and it will be impossible to properly choose players in a fair way for calculating each tank´s numbers without overall player information. I have requested Phalynx that data. He sent me some info, but it is apparently wrong. I kept seeing 1300 and 1200 WN7 players in the top 50 for several tanks, which was weird. I hand checked, and almost all of them are 1800-2000 WN7 players, so it seems the player info Phalynx sent me was wrong, although every single per-tank data I hand checked seemed fine.

Share this post


Link to post
Share on other sites

@Praetor:

 

Now that you mention it - vbaddict had problems a few months back corectly integrating WN-Rating ...

 

 

Edit: Is their forum down - cand find the thread/forum anymore?!

Share this post


Link to post
Share on other sites

I have re-analyzed the data, and it will be impossible to properly choose players in a fair way for calculating each tank´s numbers without overall player information. I have requested Phalynx that data. He sent me some info, but it is apparently wrong. I kept seeing 1300 and 1200 WN7 players in the top 50 for several tanks, which was weird. I hand checked, and almost all of them are 1800-2000 WN7 players, so it seems the player info Phalynx sent me was wrong, although every single per-tank data I hand checked seemed fine.

 

https://www.dropbox.com/s/wnxbzralb56zrvu/add2.csv

 

My attempt at adding the data from vb1.csv through vb16.csv. Is this what you need?

 

Share this post


Link to post
Share on other sites

uploaded mine from both my Pc and my laptop- so there are two.

Share this post


Link to post
Share on other sites

https://www.dropbox.com/s/wnxbzralb56zrvu/add2.csv

 

My attempt at adding the data from vb1.csv through vb16.csv. Is this what you need?

 

Nice juicebar! :thumbup:  Now just to implement the WN7 formula and voila! What did you use to do the math? R?

 

It seems Phalynx was calculating overall user WN7 by just doing an average of WN7 of each tank played, instead of doing a weighted average using # of battles on each tank. o_o

 

Another bad thing is, he deletes users who have not uploaded their dossier for over 90 days.

Share this post


Link to post
Share on other sites

Nice juicebar! :thumbup:  Now just to implement the WN7 formula and voila! What did you use to do the math? R?

 

It seems Phalynx was calculating overall user WN7 by just doing an average of WN7 of each tank played, instead of doing a weighted average using # of battles on each tank. o_o

 

Another bad thing is, he deletes users who have not uploaded their dossier for over 90 days.

 

I use python and sqlite. I learned python to do gambling analysis. I learned sqlite because I needed a database for my recording software, which I used for live sex webcam shows. So I guess you can say gambling and porn is helping the development of WN8. :D

Share this post


Link to post
Share on other sites

I learned sqlite because I needed a database for my recording software, which I used for live sex webcam shows.

 

o_o

Share this post


Link to post
Share on other sites

We have yet to rebalance kills vs damage, as well as other stats. I believe it was Folterknecht who suggested tying them together somehow, to "punish" kill stealers who do barely any damage and only get the last shot. I'm trying to work out a way to do that, but it's pretty tricky.

 

Also, on farming damage in light tanks: I think Praetor77 suggested dragging the "expected" stats closer to the average. I think this would only be best for balance for tanks with lower expected damage. Although, as I write this, it occurs to me that a scalar function, based on the ratio of expected damage in that tank to expected damage for its battle tier overall, could be used.

 

(From that sarcastic WN7 failure post.)

Share this post


Link to post
Share on other sites

Tying them together by normalizing frags and damage to expected values, multiplying that, and then multiplying by a constant, actually makes the frag importance WORSE. This is due to frags varying about twice as much as damage. I think the best solution is simply to change frag/damage relations (presently 40%/40% of WN7), and make damage weigh twice as much, like this (damage 55%, frags 25%):

 

nopg.jpg

 

Player A is an "average" player, lets suppose they are playing tier 8s, the same tank. Player B farms frags at the expense of some damage. Player C would be actually an example outstanding game for Player 1, focusing on dealing damage. And player D is an outstanding frag farming game. (games like C and D occur, even if you are not trying to "farm" anything, just doing what you can to win, but they illustrate a point). You will notice how tying frags and damg make things worse, while simply increasing the weight of damage solves the problem.

Share this post


Link to post
Share on other sites

juicebar since you are obviously awesome, could you retrieve only new players (that were not in the old dataset) from this data Phalynx just gave me?

 

They are in the dropbox  named as 2013-09-03. Huge thanks mate! Oh, it would also be aswesome if you could do the "add" thing you did before, adding the stats for every tank but only for the new players?

Share this post


Link to post
Share on other sites

juicebar since you are obviously awesome, could you retrieve only new players (that were not in the old dataset) from this data Phalynx just gave me?

 

They are in the dropbox  named as 2013-09-03. Huge thanks mate! Oh, it would also be aswesome if you could do the "add" thing you did before, adding the stats for every tank but only for the new players?

 

I just downloaded vBAddict_Dossiers_2013-09-03_batch_0.csv.zip. For some reason, I can't unzip it. Can you upload it as just a straight csv file?

Share this post


Link to post
Share on other sites

I still can't unip vBAddict_Dossiers_2013-09-03_batch_0.csv.zip. I notice it's only 86.66KB while the other files are larger than 1.83MB in size.

Share this post


Link to post
Share on other sites

Tying them together by normalizing frags and damage to expected values, multiplying that, and then multiplying by a constant, actually makes the frag importance WORSE. This is due to frags varying about twice as much as damage. I think the best solution is simply to change frag/damage relations (presently 40%/40% of WN7), and make damage weigh twice as much, like this (damage 55%, frags 25%):

 

nopg.jpg

 

Player A is an "average" player, lets suppose they are playing tier 8s, the same tank. Player B farms frags at the expense of some damage. Player C would be actually an example outstanding game for Player 1, focusing on dealing damage. And player D is an outstanding frag farming game. (games like C and D occur, even if you are not trying to "farm" anything, just doing what you can to win, but they illustrate a point). You will notice how tying frags and damg make things worse, while simply increasing the weight of damage solves the problem.

 

I beg you to just find R between the different stats and winrate, then adjust for them (I have already come up with a good way to do this). Don't use an arbitrary percentage for it, based on what you think is correct. I already have an idea for how to tie them together effectively, but it's not complete yet. When I say R, I mean for a very large (maybe 5000-10000) random sample, if not every single data point.

Share this post


Link to post
Share on other sites

We don't just need to tie Damage and Frags, though. We can do it between all of the stats based on the correlation coefficients between each pair.

 

As for the coefficients for the value of each stat, I will shortly put the idea in a Word document, along with a few ideas for tying the stats together.

 

NOTE: These formulae will only work after R has been found for each of the 4 stats AND win rate. So, can we find R between these stats for a large sample of players already? The method by which we should do this is to compare player stats with the expected top stats for their weighted average tier (NOT expected for their tanks), which will need to be resurrected for the purposes of the creation of WN8, but should not remain within the formula. This is to show how much each stat contributes to the win in the big picture, while avoiding a per-tank bias. Then, we compare those resulting stats to win rate, then each other.

Share this post


Link to post
Share on other sites

When using per-tank stats, R does not give statistically significant numbers. For each tank, each player has 50-200 games. Winrate just varies too much for such small sample rates. You have players with 50 games, very low stats and 80% winrate, and many players who clearly play solo, with top notch stats but 60% winrate.

 

R is about .45 for damage and frags, and even less for the other stats.

 

Correlating to winrate will not help us here, I am afraid.

 

What I did was use player overall WN7 to make sure players of approximately the same level are being taken into account for each tank. The values I came up with for each tank are very similar to the top 100 displayed in XVM in-game, except for bizarre and clearly weird values XVM gives for example, for Batchat, M48A1, etc.

 

 

I would definitely like everyone to give the data the eye test, and look for weird, unbalanced, outlier tanks. Anything that comes to mind, pls post it.

 

https://www.dropbox.com/s/axpjw9r08oautub/results.xlsx

 

 

 

PS: shnog ur idea is worth trying, maybe putting together a DB using wotstatgrabber.

Share this post


Link to post
Share on other sites

R does not give statistically significant numbers. For each tank, each player has 50-200 games. Winrate just varies too much for such small sample rates. You have players with 50 games, very low stats and 80% winrate, and many players who clearly play solo, with top notch stats but 60% winrate.

 

R is about .45 for damage and frags, and even less for the other stats.

 

Correlating to winrate will not help us here, I am afraid.

 

I specifically said that I don't mean for each tank. Say the stat we are comparing is damage. We take the total damage of the player and divide it by the top expected damage for their WAT (data which we can interpolate). From that point, we compare the resulting amount (K1) to the player's overall winrate. Then, we can compare the different K values, K1, K2, K3, K4.

 

We can just do this for all players over 1000 battles total or so.

 

EDIT: This is giving me a headache...

Share this post


Link to post
Share on other sites

Sounds good. We would need a decent (aprox. 50k) player DB though, which we don´t have. If we can get HIbachi to modify wotstatgrabber and give us the data we need, we could use it to put together a DB. In fact, using the per-tank values on the spreadsheet I posted, we could directly calculate expected damage and frags for each player.

Share this post


Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.