Jump to content
bjshnog

⟪WN8⟫ Development / Resources

Recommended Posts

And to be more specific, I gave Xelos data of uploaded Battle Results and Replays.

Link to post
Share on other sites

And to be more specific, I gave Xelos data of uploaded Battle Results and Replays.

Can you email me the same dataset please? Thanks!  :thumbup:

Link to post
Share on other sites

Already done yesterday. It consists of 30 files for each day of January. I've limited it to 20k battles per day, otherwise it would be around four times bigger.

Link to post
Share on other sites

No, people keep telling you that information is not available by the API.

 

The data he used is from vbaddict, which is using dossier uploads by willing volunteers and isn't available for a general user.

Stagnet - I knew that about the API, but that does not diminish the facts that a critical performance data point was not being fully considered or weighed.  I can just see Praetor cringe at that comment - sorry!

 

And I contend that XVM and its use of WN has changed general game play. Some of it good and some of it bad.

 

But we have lots of purple/good players stating that DPG and KPG is god. I strongly contend that it is not. I contend its spotting and assisted damage which results others high DPG and KPG = a high WR for the team.

 

And right now with the data we have available WN is only a piece of the skill measure - I contend that there is more. 

 

Lastly by making these statements I am not poo poo'ing the effort here to quantify a skill measure - in fact I have had my opinion changed by the WN guys on how they are doing the calculation and am very much intrigued by the whole effort.  :thumbup:

 

Money is in the mail - lol!

Link to post
Share on other sites

LOL! :D

 

Did you rescale/normalize the variables before the analysis (take each value and substract the mean for that variable, and then divide by standard deviation)?

 

It seems to me you didn´t, and that can mess up multiple regression analysis, since the scales are so different for damage as compared to, for example, spots, or frags. I glean this from the results, since it also states that 1 cap point contributes 10 times more than damage to a win. :)

 

Also, using crude damage is very counterproductive, since it lacks discriminative power. Using tier-based damage or better yet, rDAMAGE will allow a fair comparison (can´t find my graphs but for overall accounts, avgDAMAGE had around a 0.4 correlation to winrate while rDAMAGE has around 0.78. I suspect for single tanks using crude damage would have an even lower discriminative power, since doing 500 damage in a tier 1 makes a win highly probable, while in a tier 10 it makes a loss highly probable). I guess you could kind of do something similar for assisted damage by dividing it first by the average HP for the tier of tank being played. I guess the same thing could be done for damage instead of using rDAMAGE, to make a fair comparison.

 

Lastly, I am inexperienced in logit regression, can you input variables as categorical variables? If so, this needs to be done with tier instead of considering it as an explanatory variable. Same for tankID.

 

I would love to play around with this dataset, can you upload to mediafire or elsewhere and/or email it to me?

 

I didn't standardize and normalize to tier because I have tier categorical control variables and i have never heard that standardizing and normalizing can cause problems with coefs (just interpretation). I can try and see if this does anything.

 

With regards to your rdamage/crud damage; I think our datasets are very different, your working with aggregates? While I'm working with individual battle results so I don't think your suggestion would make any difference but I can see if I can do something comparable.

 

Yes, I have categorical variables in there, such as the tier categorical predictor variable.

 

Yes I fucked TankID up.  :thumbdown:

 

There are problems with the dataset, for example each line is being treated independently which it isn't because 29 other people contributed to that specific battle. There are a few ways I can deal with this (possibly treat it like a survey where each battle is a stratum... i doubt i have enough ram to do this, or break up analysis by tier/tank... and other ways).

 

Logit (logistic regression) is quite a useful when dependent variable is binary or ordered. I'm just using the tool that fits the data. Since win can only equal 0 or 1 it makes sense. I may try a probit link function to really stress test my computer.  :smiledown:

 

I don't want to hijack your WN8 thread so it may be better that I start a new thread when I get something more substantial.

Link to post
Share on other sites

Stagnet - I knew that about the API, but that does not diminish the facts that a critical performance data point was not being fully considered or weighed.  I can just see Praetor cringe at that comment - sorry!

 

And I contend that XVM and its use of WN has change general game play. Some of it good and some of it bad.

 

But we have lots of purple/good players stating that DPG and KPG is god. I strongly contend that it is not. I contend its spotting and assisted damage which results others high DPG and KPG = a high WR for the team.

 

And right now with the data we have available WN is only a piece of the skill measure - I contend that there is more. 

 

Lastly by making these statements I am not poo poo'ing the effort here to quantify a skill measure - in fact I have had my opinion changed by the WN guys on how they are doing the calculation and am very much intrigued by the whole effort.  :thumbup:

 

Money is in the mail - lol!

 

 

at least at posting BS you reached purple already

Link to post
Share on other sites

I didn't standardize and normalize to tier because I have tier categorical control variables and i have never heard that standardizing and normalizing can cause problems with coefs (just interpretation). I can try and see if this does anything.

 

With regards to your rdamage/crud damage; I think our datasets are very different, your working with aggregates? While I'm working with individual battle results so I don't think your suggestion would make any difference but I can see if I can do something comparable.

 

Yes, I have categorical variables in there, such as the tier categorical predictor variable.

 

Yes I fucked TankID up.  :thumbdown:

 

There are problems with the dataset, for example each line is being treated independently which it isn't because 29 other people contributed to that specific battle. There are a few ways I can deal with this (possibly treat it like a survey where each battle is a stratum... i doubt i have enough ram to do this, or break up analysis by tier/tank... and other ways).

 

Logit (logistic regression) is quite a useful when dependent variable is binary or ordered. I'm just using the tool that fits the data. Since win can only equal 0 or 1 it makes sense. I may try a probit link function to really stress test my computer.  :smiledown:

 

I don't want to hijack your WN8 thread so it may be better that I start a new thread when I get something more substantial.

 

Yeah, pls do start another thread! And nice to have someone else also crunching numbers!  :thumbup:

Link to post
Share on other sites

at least at posting BS you reached purple already

 

Ok - I see you're and arty player. How are you achieving damage without spots and assisted damage? Expand - why am I wrong. I have tried to explain my position, what is yours and why?

 

I am interested really - lay it on us!

Link to post
Share on other sites

But we have lots of purple/good players stating that DPG and KPG is god. I strongly contend that it is not. I contend its spotting and assisted damage which results others high DPG and KPG = a high WR for the team.

 

The chief problem with this contention is that not all maps have an equal potential for spotting and assist damage, but all maps do have the same potential for direct damage.

 

On actually quite a lot of maps the lines of fire are sufficiently short that there is simply no opportunity to spot for someone to damage targets, if they can shoot the target they're close enough to see it for themselves.

 

I suspect that if an analysis can be run on the 8.8+ dossiers that includes spotting damage it will be a minor term at best, simply because of the uneven opportunities presented to actually do it by the map pool.

Link to post
Share on other sites
Filtered the data from Phalynx as it was a HUGE dataset, picked at random 500k battles from tier 9 tanks (500k rows, although many tanks were in the same battles, does that make sense?). I realized normalizing scale is not necessary for inputing the data to Eureqa since the logistic function used to predict models for binary classification already squashes the result of each variable to a 0,1 range. Xelos, if you can do something similar with your variables before doing the logit regression that would be dandy! Otherwise, your resulting coefficients have to at least be multiplied by the mean for each variable, right?
 
Then I plugged all of that into Eureqa and ran 1.5e10 formula evaluations using AUC error and fitting damage, track_damage, radio_damage, kills, defense, spots, teamdamage to a binary win formula using a logistic function. (did not use cap, since it will be a strong predictor but only due to the fact that you can only get cap points while winning. Causality is completely reversed for this  stat, so no sense in including it).
 
Some preliminary results:
 
 
So, Gloating´s prediction was much more accurate than surfer´s. Sorry! :)
Link to post
Share on other sites

Do you haven any more of those juicy AUC numbers? i'd love to see how precise certain combinations and individual variables are.

Link to post
Share on other sites

Let's tell Tazilon.

 

On the subject of Tazilon:

 

snapback.pngTazilon, on Feb 20 2014 - 01:22, said:

It is east to pad WN8 with Scouts if you are good at passive scouting and can rack up large amounts of assisted damage.

Oh wait...Wn8 doesn't account for that.  That, plus a plethora of other design flaws makes Wnx a joke.  Ignore Wn8 and play the game.

 

My response:

 

The real joke is that WG API doesn't include spotting damage. Can't use what you don't have.

Stop being bitter and get over it. You're a broken record and it's getting real old.

 

That [edited]head neg'd me for saying that light tanks are really easy to pad WN8 in. I wonder if he's gonna go crying to the mods and get me moderated because "mommy...he's being mean."

 

http://forum.worldoftanks.com/index.php?/topic/333192-are-light-tanksscouts-bad-for-wn8/

Link to post
Share on other sites

On the subject of Tazilon:

 

snapback.pngTazilon, on Feb 20 2014 - 01:22, said:

It is east to pad WN8 with Scouts if you are good at passive scouting and can rack up large amounts of assisted damage.

Oh wait...Wn8 doesn't account for that.  That, plus a plethora of other design flaws makes Wnx a joke.  Ignore Wn8 and play the game.

 

My response:

 

The real joke is that WG API doesn't include spotting damage. Can't use what you don't have.

Stop being bitter and get over it. You're a broken record and it's getting real old.

 

That [edited]head neg'd me for saying that light tanks are really easy to pad WN8 in. I wonder if he's gonna go crying to the mods and get me moderated because "mommy...he's being mean."

 

http://forum.worldoftanks.com/index.php?/topic/333192-are-light-tanksscouts-bad-for-wn8/

 

It's not like Tazilon is anything better than a mediocre LT player who's just played LTs a LOT. And his understanding of the game is pubbie-level.

Link to post
Share on other sites

And he takes a lot of words to say nothing.

 

 

I didn't get good at scouting till I realized everything in Taz's guide was dum.

Link to post
Share on other sites

Do you haven any more of those juicy AUC numbers? i'd love to see how precise certain combinations and individual variables are.

 

Suggest to me some analysis and I will gladly perform it. I had already thought about doing another tier as control, and seeing how this works for light tanks (remember I did tier 9 which has no LTs).

Link to post
Share on other sites

It might be nice to see how T8 lights compare to that T9 baseline (as they have the same minimum tier).

 

Yeah that was my next move. :)

 

PS: Xelos, you don´t need to scale or transform the variables as long as you stay within a tier. Otherwise, you will somehow need to normalize damage, damage_track and damage_radio by tier somehow. I suggest dividing by average HP for that tier (excluding arties).

Link to post
Share on other sites

Awe shit - I felt so good for a minute! lol

 

I was hoping the result would result in more scouting play - oh well one can hope.

 

Question have you looked at total assisted damage for the team rather than individual - if possible!

 

You see what I mean - if you take the total assisted damage for the team does it have an effect. VBaddict does have uploaded battle stats - even by map.

 

This maybe just to complicated.

 

As I kick a rock, I just find it hard to believe that spotting and assisted damage is not a major contributor to WR. I realize this is based on perception rather than numbers!

Link to post
Share on other sites

Awe shit - I felt so good for a minute! lol

 

I was hoping the result would result in more scouting play - oh well one can hope.

 

Question have you looked at total assisted damage for the team rather than individual - if possible!

 

You see what I mean - if you take the total assisted damage for the team does it have an effect. VBaddict does have uploaded battle stats - even by map.

 

This maybe just to complicated.

 

As I kick a rock, I just find it hard to believe that spotting and assisted damage is not a major contributor to WR. I realize this is based on perception rather than numbers!

 

I think the problem is that mostly anyone can "spot", and most players do. Spotting damage doesn´t necessarily stem from good play. Also, you depend on your team´s ability to deal damage.

On the other hand, only good players can consistenly deal damage to the enemy team, while the average pub struggles to do this.

 

Although anecdotal, who hasn´t had a game in an LT where you are perma-spotting 4 or 5 enemy tanks for 2 or 3 minutes, either form an active or passive scouting, and absolutely no one on your team can land hits on them? Nevertheless, I would suspect that good players (lets say 1900WN8 and above players) probably have increased covariation between assisted damage and winning. However, for non-light tanks at least, the link between assisted damage and wins is pretty frail.

Link to post
Share on other sites

I see that - begrudgedly!

 

lol: "Although anecdotal, who hasn´t had a game in an LT where you are perma-spotting 4 or 5 enemy tanks for 2 or 3 minutes, either form an active or passive scouting, and absolutely no one on your team can land hits on them?"

 

Or the other way when no one is scouting - except the other team and invizi shots are coming from all directions and your team melts in the first 2 to 3 minutes!

 

The bottom line and others(I) have mentioned it here - is that it is truly amazing at how good scouting can completely change the outcome of a battle, its a shame there is not a way to score that as a indicator of skill.

Link to post
Share on other sites
 

I see that - begrudgedly!

 

lol: "Although anecdotal, who hasn´t had a game in an LT where you are perma-spotting 4 or 5 enemy tanks for 2 or 3 minutes, either form an active or passive scouting, and absolutely no one on your team can land hits on them?"

 

Or the other way when no one is scouting - except the other team and invizi shots are coming from all directions and your team melts in the first 2 to 3 minutes!

 

The bottom line and others(I) have mentioned it here - is that it is truly amazing at how good scouting can completely change the outcome of a battle, its a shame there is not a way to score that as a indicator of skill.

100% agreed. This first part here is pure guesstimation: Problem is that it seems assisted damage stat is just too noisy, like cap. Sui-scouting can rack up assisted damage. Yoloing can rack up assisted damage. At the same time, a very carefully and poperly executed scouting run can sometimes rack up small numbers of assited damage due to team inability to deal damage. If a stat depends on your team to give you the win, then it will probably not be as decisive as damage, which you can do yourself, since in 90% of games, you will have targets to shoot at even if you don´t spot yourself.

 

 

My analisis for 100k tier 8 LT games:

 

The most accurate solution using one variable is again with damage_dealt, which correctly predicts 70.4% of games, although down from 72.3% in tier 9 tanks.
 
Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.

×
×
  • Create New...