Jump to content
You need to play a total of 20 battles to post in this section.
Kelorn

The Warships Podcast Sub Octavian QnA Transcript (4/27/17)

43 comments in this topic

Recommended Posts

209
[TSWAA]
Members
317 posts
6,813 battles

 

On April 27th, we at the Warships Podcast interviewed Sub Octavian about Warships Math. Today was the first time I had a chance to sit down and give you text transcript of the answers he gave. If you want more context, please check out the episode itself! Enjoy!

Does the Range Mod decrease dispersion?

  • No, Dispersion is linear

The dispersion coefficients appear to be based on a linear formula based on nation, can you help us understand this formula?

  • Point blank range doesn’t conform, then it steps to a linear formula each for horizontal dispersion (shown in client) and vertical dispersion (not shown). For example IJN bb’s have larger vertical dispersion.

It is assumed that the accuracy of a ship is based on the dispersion ellipse, with dispersion in game being horizontal dispersion and sigma being the vertical dispersion. Can you help us understand how this applies in game?

  • Sigma is the weighting of the shells toward the center. Horizontal and Vertical dispersion denote the maximum extents of possible shell landing. Sigma means aiming well pays off

Does the Aiming Systems Module affect only horizontal dispersion, or does it affect Sigma as well?

  • Aiming systems module only affects Horizontal and vertical dispersions, nothing affects sigma. Vertical dispersion is not shown in the client, not secret but would be too much information for the average player. Vertical dispersion varies by ship line, however it’s not used for ship to ship balance.

How is accuracy affected by "locking" on to a target and the camoflage/modules that decrease accuracy of shots fired at the ship?

  • Not locking on a target doubles the dispersion ellipse. Locking allows shells to go over obstacles and track your target. Bonus from camouflage/modules is far less than penalty for not locking.

Does the compressed ranges of engagements in World of Warships cause penetration values that are higher than might have been expected in a historical battle?

  • Ballistics are real world data or close. Ranges aren’t artificially compressed, players are playing the game close together. Visually shell arcs are compressed on the vertical axis to assist the player in aiming and seeing the path of shells. Accuracy in the game is much higher than the real world accuracy to allow players to have fun rather than constantly missing. Ranges are determined initially based on the visual restrictions of the fire control systems of the firing ship.
  • Wargaming is fine with the lack of plunging fire happening in the game. Ships need to be at 26-30km to overcome the auto-bounce mechanic unless overmatch is a factor.

When will the citadels be lowered in the Iowa/Montana? What is the process to making this happen?

  • Citadel change was based on community feedback. Game definition of a citadel is slightly different than the real life definition for balancing purposes. Lowering the Iowa/Montana citadel will improve the quality of life for USN BB captains without making them overpowered. They used auto-tests to fire thousands of shells to test the armor. Technically it’s an easy fix, but it needed extensive testing before release. Hoping to be released in 0.6.6. Good example of a fruitful collaboration of players and developers.

  • The definition of a Citadel in game: Most armored portion and what is protected. A little flexible for gameplay balance over historical accuracy.

Did the change to the Alabama citadel and how it's worked out in game affect the Iowa/Montana changes?

  • Wargaming is happy with how Alabama’s citadel change worked out and it helped them realize that Iowa/Montana will be ok.

We've been told there are two fire resistance coefficients per tier -- one for upgraded ships and one for stock ships. However, some ships just "feel" like they catch on fire more often. Is fire resistance a balance mechanic that can be tweaked beyond these coefficients for individual ships? If not, do premium ships count as stock or upgraded vessels for determining their Fire Resistance Coefficient?

  • Fire resistance is set per ship tier and hull and is not a balance mechanic. It grows slowly as your climb the tiers. Premium ships have the “upgraded” hull fire coefficient.

When determining if a vessel has line of sight to an enemy, what section(s) of the enemy ship must be seen? Will any section do or are there specific reference points used to calculate spotting mechanics?

  • Originally spotting mechanics were based on 3 points. Bow, aft, and top of the mast. However, it was too processor intensive. Spotting is set based on the top of the mast only. They’re considering returning to 3 point spotting with some optimizations.

How often does the program check to see if a vessel is spotted? There seems to be a noticeable delay between a vessel showing up on the minimap and being rendered in client.

  • Delay is the client rendering the model. It has been improved, but there’s room for more improvement to shorten the delay.

1.) What factors affect a ship's acceleration?

2.) If mass is a factor of acceleration, do installed modules / consumables / camouflage affect a ship's mass?

  • Acceleration is based on ship mass and engine power. Ship mass is set per hull and is unaffected by modules/consumables/camouflage. Royal Navy CLs get the propulsion module for “free”. Vanessa put in a plug for Graf Spee to get it too.

Do ships take more damage from flood if moving at speed vs not moving at all?

  • No, flooding damage is set per tick.

Does the propulsion mod help assist in slowing down as well as speeding up?

  • Yes, but it’s a tiny amount. Setting your engines to stop actually reverses the propellers until you stop. This is seen when your engine goes out and you slow down slower. The propulsion mod helps your engine start up, so you get a tiny benefit, but its barely noticeable when slowing down.

How much feedback to you get from the surveys? Do you feel it is a good statistical sample size to aid with balance?

  • Feedback from surveys is important both from emails and the ingame instant surveys. Email surveys are linked to account statistics. This lets them analyze surveys based on player experience and skill. Ingame surveys are amalgated and thousands of battles are considered and happiness can be correlated to # of DD’s, # of premiums, etc.

Can you explain the thought process on balancing ships and maybe give an example of the stages in which you balance a particular ship?

  • Conflict begins with Historical accuracy vs Gameplay. Ships are chosen to fill a certain role by developers and then inserted into the game and iterated for balance. Sometimes in game models have to adjust for balance purposes. Supertester/CC skill level is taken into account. Luckily some WG employees are not very skilled ;)

  • Testing phases: Supertest QA -> Supertest Balance -> Production Test on Live servers -> Ship goes live -> Live updates over time.

Are you able to explain any of the difficulties you face in balancing with regard to Random match play to Organized Competitive play?

  • Core mode is Random battles, biggest sample size and statistics, so ships are balanced around Random battles. However, WG understands that in organized competitive play treats ships in a different way. They intend to continue to balance around random battles, but monitor competitive play to see the “min/max” form of various ships. Balance purely on competitive play isn’t going to happen.

Does the team ever consider the skill required to play certain ships to their max potential, part of the balancing process?

  • Minotaur is the prime example. The ship performs differently depending on how good the player is. They consider diverse ships, some of which require high skill levels, to be good for the game.

  • They have advanced statistics, such as how HE performs from one ship vs AP to certain parts of each ship. They can organize by source of damage or class or type.

  • They see “heat maps” for level design to see where people go on maps over thousands of battles, including places of ship destruction.

 

  • Cool 3

Share this post


Link to post
Share on other sites
1,113
[CHASE]
Beta Testers, In AlfaTesters
2,973 posts
13,112 battles

I would have asked how HE and fire are considered to be balanced 

Share this post


Link to post
Share on other sites
2,887
[NSF]
Beta Testers, In AlfaTesters
5,304 posts
9,284 battles

That Alabama thread continues to pay off I see. Looks like they used her as the testbed for the Monty/Iowa citadel changes, and liked what they saw.

Share this post


Link to post
Share on other sites
1,363
[HYD]
Members
7,105 posts
5,289 battles

I would have asked how HE and fire are considered to be balanced 

 

I would like to ask you why you think they're not. 

Share this post


Link to post
Share on other sites
41
[ZAZEN]
Alpha Tester
404 posts
8,964 battles

Thanks for providing the transcript!

I heard the interview and I believe he also mentioned that during flooding a ship's speed is affected which I am not sure if new players know this.

 

Did the podcast end abruptly or did my download fail to get it all? Mine ended without closing remarks but cut him off mid sentence.

 

Share this post


Link to post
Share on other sites
209
[TSWAA]
Members
317 posts
6,813 battles

Thanks for providing the transcript!

I heard the interview and I believe he also mentioned that during flooding a ship's speed is affected which I am not sure if new players know this.

 

Did the podcast end abruptly or did my download fail to get it all? Mine ended without closing remarks but cut him off mid sentence.

 

 

Husband,

Yes he did say that, but it wasn't part of the original question =)

 

The episode got revised and a new version was uploaded on Friday of last week that was the correct length. Please feel free to check it out on our website, as I don't know if itunes/google play will update it properly!

Share this post


Link to post
Share on other sites
2,894
[HINON]
[HINON]
Wiki Lead, Beta Testers, Privateers
6,801 posts
5,248 battles

I like that someone put in that Graf Spee needed the propulsion mod for free. There is a historical precedent for it after all. :)

 

Share this post


Link to post
Share on other sites
3,879
[-K--]
Beta Testers, In AlfaTesters
8,172 posts
10,844 battles

There were a lot of question about dispersions, but not the correct ones, unfortunately.

 

Fost and foremost Sub should have been asked what's the exact connection between standard deviation of shells (say, for dispersion's horizontal projection) and this "sigma value". Another good question would have been the exact way they determine the cut-off boundary for the dispersion ellipse. Another good question would have been whether shell dispersion is always modeled as a normal distribution.

 

Appreciate the effort, but you didn't fully clarify the situation. Next time request community help for coming up with these questions, I'm sure plenty of people that are decently familiar with statistics and data analysis will be glad to help, myself included. 

Share this post


Link to post
Share on other sites
4,615
[-K-]
[-K-]
Members
6,121 posts
27,978 battles

Fost and foremost Sub should have been asked what's the exact connection between standard deviation of dispersion (say, for its horizontal projection) and this "sigma value". Another good question would have been the exact way they determine the cut-off boundary for the dispersion ellipse. Another good question would have been whether shell dispersion is always modeled as a normal distribution.

 

Appreciate the effort, but you didn't really clarify the situation. Next time request community help for coming up with these questions, I'm sure plenty of people that are decently familiar with statistics and data analysis will be glad to help, myself included. 

 

I'd also be interested and willing to help. However, I think this line of inquiry will rapidly (and may already have) run up against the boundary where WG starts getting concerned about their trade secret special sauce. We saw that for sure in discussions around the ballistics model, which they frequently refer to as "proprietary".

 

For what it's worth, Sub_Octavian has been pretty good about providing information when people can make the case for needing to know additional details in order to make in-game decisions, so if we can come up with some, we'll have better standing to make that kind of ask.

Share this post


Link to post
Share on other sites
209
[TSWAA]
Members
317 posts
6,813 battles

There were a lot of question about dispersions, but not the correct ones, unfortunately.

 

Fost and foremost Sub should have been asked what's the exact connection between standard deviation of shells (say, for dispersion's horizontal projection) and this "sigma value". Another good question would have been the exact way they determine the cut-off boundary for the dispersion ellipse. Another good question would have been whether shell dispersion is always modeled as a normal distribution.

 

Appreciate the effort, but you didn't fully clarify the situation. Next time request community help for coming up with these questions, I'm sure plenty of people that are decently familiar with statistics and data analysis will be glad to help, myself included. 

 

I appreciate the fact that you would like more information, heck i would like more information. That said, Sub made it pretty clear before the interview and during it that the information he gave on these subjects were as far as he could go about the details of dispersion. If you feel like you can "force" him to tell you what exactly is going on, please arrange an interview and ask your questions. 

 

But don't be surprise when he smacks you like he smacked Flamu.

Share this post


Link to post
Share on other sites
1,544
[PSA]
Members
5,118 posts
3,754 battles

Sigma has nothing to do with vertical disperson and Octavian firmly stated it above. Vertical dispersion DOES exist in the game as a value that's just not displayed in the port stats. But it is different from Sigma which is the tendency for shells to land where you aimed. It *literally* means standard deviation ... the tendency for a sample to either be focused closer to it's mean, or spread out.

 

https://en.wikipedia.org/wiki/Standard_deviation

Edited by KaptainKaybe

Share this post


Link to post
Share on other sites
4,615
[-K-]
[-K-]
Members
6,121 posts
27,978 battles

Sigma has nothing to do with vertical disperson and Octavian firmly stated it above.

 

The way he said it during the interview (of which this post is a summary) actually implied that sigma acts on both the horizontal and vertical axes.

Share this post


Link to post
Share on other sites
1,544
[PSA]
Members
5,118 posts
3,754 battles

 

The way he said it during the interview (of which this post is a summary) actually implied that sigma acts on both the horizontal and vertical axes.

 

No, he was quite clear. The maximum dispersion values, both horizontal and vertical, denote the maximum ranges the shells can deviate. Sigma denotes the weighting of those shells towards the center. I mean, that is literally what is written above. There is nothing remotely ambiguous about that. Sigma (standard deviation) has a very specific definition in statistics and I'm unsure why some people are trying to make it mean something different.

Share this post


Link to post
Share on other sites
209
[TSWAA]
Members
317 posts
6,813 battles

 

No, he was quite clear. The maximum dispersion values, both horizontal and vertical, denote the maximum ranges the shells can deviate. Sigma denotes the weighting of those shells towards the center. I mean, that is literally what is written above. There is nothing remotely ambiguous about that. Sigma (standard deviation) has a very specific definition in statistics and I'm unsure why some people are trying to make it mean something different.

 

My interpretation of Sub's words follows what Kaybe is saying. You can listen to the episode to hear sub for yourself, if you wish.

Share this post


Link to post
Share on other sites
4,615
[-K-]
[-K-]
Members
6,121 posts
27,978 battles

No, he was quite clear. The maximum dispersion values, both horizontal and vertical, denote the maximum ranges the shells can deviate. Sigma denotes the weighting of those shells towards the center. I mean, that is literally what is written above. There is nothing remotely ambiguous about that. Sigma (standard deviation) has a very specific definition in statistics and I'm unsure why some people are trying to make it mean something different.

 

"Center" is defined on both the horizontal and vertical axes. If sigma increases density toward the vertical center, then it is acting on the vertical dispersion. Not to be confused with the MAXIMUM vertical dispersion, which is what you're talking about.

Share this post


Link to post
Share on other sites
3,879
[-K--]
Beta Testers, In AlfaTesters
8,172 posts
10,844 battles

 

No, he was quite clear. The maximum dispersion values, both horizontal and vertical, denote the maximum ranges the shells can deviate. Sigma denotes the weighting of those shells towards the center. I mean, that is literally what is written above. There is nothing remotely ambiguous about that. Sigma (standard deviation) has a very specific definition in statistics and I'm unsure why some people are trying to make it mean something different.

 

Except that increasing statistical sigma (i.e. SD) would make dispersion worse, not better. 

 

 

I appreciate the fact that you would like more information, heck i would like more information. That said, Sub made it pretty clear before the interview and during it that the information he gave on these subjects were as far as he could go about the details of dispersion. If you feel like you can "force" him to tell you what exactly is going on, please arrange an interview and ask your questions. 

 

But don't be surprise when he smacks you like he smacked Flamu.

 

No, I don't suppose I can force an answer out of him. Still would be interesting to ask though. 

 

Right now it seems that "sigma value" is somehow connected to standard deviation of shell dispersions, while dispersion ellipse boundaries are not rigidly tied to the latter (and it's unclear how they are determined). Incidentally, this would mean that aiming mods and camo that affect only dispersion (that is, according to Sub, only affect the ellipse boundary) are essentially worthless.

 

It's ridiculous that WG refuses to share such essential data. They charge money for camo that decreases dispersion, and yet don't give a good description on what exactly is being affected. They sell flags that increase torpedo flooding chance by 15%, and yet players have no clue what the 15% are being added to. 

  • Cool 1

Share this post


Link to post
Share on other sites
1,544
[PSA]
Members
5,118 posts
3,754 battles

 

"Center" is defined on both the horizontal and vertical axes. If sigma increases density toward the vertical center, then it is acting on the vertical dispersion. Not to be confused with the MAXIMUM vertical dispersion, which is what you're talking about.

 

Where are you guys pulling this from the Octavian's words. He literally said "Sigma is the weighting of the shells toward the center". It has nothing to do with vertical center any more or less than it does horizontal center. It just means center, period. A better sigma value means the shells are likelier to go land where you placed your cursor, on BOTH axis.

 

As for the sigma value itself ... I was also confused at first, but I googled it, and it appears that the higher the value, the more likely it is be 'normal'. Apparently, there is this value called six sigma (6σ) that is considered the standard deviation that is tight enough during processes to keep the number of defective results at an acceptable level. By that logic, a sigma value of 2.0 is better than a sigma value of 1.8 or 1.5

 

Source:

http://www.sixsigma-institute.org/What_Is_Sigma_And_Why_Is_It_Six_Sigma.php

 

 

 

 

Share this post


Link to post
Share on other sites
4,615
[-K-]
[-K-]
Members
6,121 posts
27,978 battles

 

Where are you guys pulling this from the Octavian's words. He literally said "Sigma is the weighting of the shells toward the center". It has nothing to do with vertical center any more or less than it does horizontal center. It just means center, period. A better sigma value means the shells are likelier to go land where you placed your cursor, on BOTH axis.

 

As for the sigma value itself ... I was also confused at first, but I googled it, and it appears that the higher the value, the more likely it is be 'normal'. Apparently, there is this value called six sigma (6σ) that is considered the standard deviation that is tight enough during processes to keep the number of defective results at an acceptable level. By that logic, a sigma value of 2.0 is better than a sigma value of 1.8 or 1.5

 

Source:

http://www.sixsigma-institute.org/What_Is_Sigma_And_Why_Is_It_Six_Sigma.php

 

Okay, so first of all, we know that the WG use of the term "sigma" is not the literal same as the statistical definition, because normally higher sigma means "more spread out" whereas in this case we know it means "less spread out". They're related, but have some kind of inverse relationship. That's not the confusing part. We've known that with certainty for over a year.

 

The point that I'm making, and I don't know why people are contesting, is that higher sigma also means less vertical over/undershoot on average because it increases the tendency toward the vertical center. That is literally, by definition, affecting the vertical dispersion. Not the vertical MAXIMUM dispersion. The vertical dispersion pattern is affected by sigma.

Share this post


Link to post
Share on other sites
3,879
[-K--]
Beta Testers, In AlfaTesters
8,172 posts
10,844 battles

 

Where are you guys pulling this from the Octavian's words. He literally said "Sigma is the weighting of the shells toward the center". It has nothing to do with vertical center any more or less than it does horizontal center. It just means center, period. A better sigma value means the shells are likelier to go land where you placed your cursor, on BOTH axis.

 

As for the sigma value itself ... I was also confused at first, but I googled it, and it appears that the higher the value, the more likely it is be 'normal'. Apparently, there is this value called six sigma (6σ) that is considered the standard deviation that is tight enough during processes to keep the number of defective results at an acceptable level. By that logic, a sigma value of 2.0 is better than a sigma value of 1.8 or 1.5

 

Source:

http://www.sixsigma-institute.org/What_Is_Sigma_And_Why_Is_It_Six_Sigma.php

 

 

 

 

 

If the distribution is normal, then tighter sample placement towards mean denotes smaller SD (aka sigma). Period. This is stats 101. Which, incidentally, means that "sigma value" that WG uses is not SD (but seems to be somehow connected to it). 

 

I'll make a thread later today about this later today, with visuals and whatnot. If Sub didn't misspeak, then dispersion buffs are pretty meaningless, which is quite a revelation. 

Share this post


Link to post
Share on other sites
2,531
[BWC]
Beta Testers
3,432 posts
9,986 battles

Maybe this will help understand the dispersion/sigma issue.

 

Take a piece of paper, and draw a circle. Draw a crosshair through the center. Horizontal line is X, Vertical is Y.

 

If X and Y are equal, you get that circle. If you reduce Y and keep X the same, you get an ellipse along the horizontal line. Same if you do the reverse, but with an ellipse in the vertical direction.

 

So, that circle or ellipse is the maximum area your rounds will deviate from any shot aimed at the center of the circle. The X and Y values determine the shape of that area.

 

However, you don't have a scale to tell how big that area is. The shape may be the same, but having a diameter of 5 meters will produce completely different results than one of 500. Dispersion is the scale of this shape (probably measuring the X line), but not what determines the shape itself. For that, you'd also need the other (Y line), and WG is not giving out that second number.

 

Sigma is a value that determines how close your shots will average falling towards the center of that area rather than the outside. A set of attacks with the same sigma will have the same general pattern of shell hits within the shape determined by the Dispersion values, but the actual difference in scales of those two identical shapes will produce different results from the same sigma.

 

To illustrate what I mean, put three dots inside that circle. Those are where the sigma influences how your shots will fall inside the circle made by your dispersion. Make another drawing of the same circle with the same three dots in the same places in that circle as the first, but make the second circle half the size of the first (so it looks like a half-size copy of the first). That's what happens with smaller dispersion with the same sigma.

 

Hope this helps.

 

Share this post


Link to post
Share on other sites
3,879
[-K--]
Beta Testers, In AlfaTesters
8,172 posts
10,844 battles

Given that Flamu was banned for something rather innocent in my opinion -- is it okay to collect game data in client and analyze it? I'm not talking data mining, just set up a macro that would fire shells in training room at the same aim point and collect screenshots for each splash, and then do image segmentation and data analysis? I don't have much time these days, but dammit I might actually do it if nothing else then to uncover the true value of camo and aiming mod dispersion bonuses. 

Share this post


Link to post
Share on other sites
209
[TSWAA]
Members
317 posts
6,813 battles

Given that Flamu was banned for something rather innocent in my opinion -- is it okay to collect game data in client and analyze it? I'm not talking data mining, just set up a macro that would fire shells in training room at the same aim point and collect screenshots for each splash, and then do image segmentation and data analysis? I don't have much time these days, but dammit I might actually do it if nothing else then to uncover the true value of camo and aiming mod dispersion bonuses. 

 

LittleWhiteMouse and iChase are currently conducting (endless) tests on this and have been since last Friday

Share this post


Link to post
Share on other sites
3,879
[-K--]
Beta Testers, In AlfaTesters
8,172 posts
10,844 battles

A set of attacks with the same sigma will have the same general pattern of shell hits within the shape determined by the Dispersion values, but the actual difference in scales of those two identical shapes will produce different results from the same sigma.

 

SD (aka sigma) is the scale, at least in statistics. The "shape" of a normal distribution (i.e. probabilistic sample distribution) always stays the same, because AUC must be one. 

 

I'm now beginning to suspect that the ellipse boundary is not connected to SD at all, whereas before I assumed it was its multiple. Some WG guy just pulled these values out of the hat and hard coded them. That would explain why maximum horizontal ellipse boundary scales linearly with range for all same-nation BBs. 

Share this post


Link to post
Share on other sites
3,879
[-K--]
Beta Testers, In AlfaTesters
8,172 posts
10,844 battles

 

LittleWhiteMouse and iChase are currently conducting (endless) tests on this and have been since last Friday

 

Okay. Curious to see what they come up with. I'll gladly offer help with image processing/segmentation or statistical analysis if need be. 

Share this post


Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.

×