Welcome to symthic forums! We would love if you'd register!
You don't have to be expert in bit baking, everyone is more than welcome to join our community.

You are not logged in.

Hey! If this is your first visit on symthic.com, also check out our weapon damage charts.
Currently we have charts for Battlefield 3, Call of Duty: Black Ops 2, Medal of Honor: Warfighter and Call of Duty: Modern Warfare 3

  • "leptis" started this thread

Posts: 282

Date of registration
: Mar 17th 2015

Platform: PC

Battlelog:

Reputation modifier: 8

  • Send private message

1

Sunday, November 6th 2016, 9:31pm

How Negative Model of Spread works (Accuracy)

The introduction of the negative model of spread in game has produced some quick conclusions:

1. Long bursts are better. Don't release the trigger.
2. Microbursts (2-3 bullets) are inefficient.
3. The bullets are increasingly more accurated.

and other ones like that.

Is this true?. Well, in general...not.

First let's look at the parameters of spread (ADS) by comparing negative and positive model (this has also had a major change by introducing a new FSM Spread). Many curves are overlapped because versions.

Spoiler Spoiler



In general, and seeing the average values (red and black lines), the negative model gives worst values of spread up to 5th bullet and reach the minimum value at the 7-9th bullet.

To analyze the geometry of the burst with negative model I have met several graphics of my simulator in a single control panel that is this: The Bursts Analyzer.

Here only geometry of burst is analized. No damage are applied.

Spoiler Spoiler



Well ... heaps of data.

We have qualitative bursts (upper left corner - clouds of hits), synthetic bursts with the main features (top center) , curves of horizontal (red shades) and vertical densities (green shades), horizontal cumulative vertical frequency (lower left) and main parameters per bullet (lower right corner).

The synthetic burst have to be explained. We have the maximum measured width of the cloud (remember that I simulate 2,000 bursts of 10 rounds): horizontal line; height: vertical line; oblique lines help to visualize the relationship between width and height . Horizontal line has vertical marks enclosing percentiles 68.26% and 95.44% of the bullets, which correspond to one and two times the standard deviation in a Normal distribution. Red vertical curve : spread applied to each bullet (horizontal dimension). An interesting line is gray one white dotted on the right that shows the maximun possible width of the hit clouds. As it is seen there is a gap between the measured width and the maximum extent possible. This indicates that probability of bullets in that gap is less than 0.025% (for 2,000 bursts).

For more details of the geometry of the cloud of hits can be seen

Position, shape and frequency distribution in a burst

Analyzed all LMGs with negative model it is found that the horizontal density of the cloud is very uniform in all bullets within the burst (including the second one) and tends to have a normal distribution from the 2-3th bullet with an effective density almost equal (it improves slightly for low recoil/spread weapons and is constant for more dispersive ones). Note the estability of standard deviation.

Spoiler Spoiler



This is because the spread is improving during the burst but the cumulative effect of the Horizontal Recoil compensates that resulting in a distribution almost equal with a very slight tendency to worsen from bullet 7-8th (due to constant spread and cumulative effect of the Horizontal Recoil).

An interesting graph is this, bullets that are in the range of body width at a specific distance (note that it is 50 m):

Spoiler Spoiler



That is, the maximum number of bullets that you could land in the body with perfect Vertical Recoil control. As you can see for this weapon at those distance the second bullet is worse than other ones, but similar as 9th and following bullets (due to the cumulative effect of the Horizontal Recoil). In other words; with this weapon at those distance it would be best two 7 bullets bursts that one of 14 bullets, opening the discussion of the optimal burst length for each weapon at a distance. Remember that re-start the burst longer than 7-8 bullets doesn't penalize the spread because you are in the minimum spread value and you have the advantage of a new accurate firts bullet (and surely of re-aiming)

You can see the differences of some weapons:

Spoiler Spoiler



I can advance that the same happens with microbursts (2-3 bullets). They have a range of validity. In a first approach are valid for highly dispersive LMGs and for all LMGs further a specific distance (shorter the more dispersive the weapon is). Maybe I will post about this.

The general conclusions are that Negative Model:

1. Produces an almost constant accuracy that worsens slightly with burst length after 7-9 bullets..

2. 7-9 bullets repeated bursts can produce better results than a maintained long burst at range (not in close range) depending on the weapon and specially with more dispersive ones.

3. Microbursts have a validity field, especially at range.

The quick conclusions in the beginning apply perfectly to Benet-Mercie Storm...no to all LMGs.


NOTE: Some more peculiarities are found when we apply damage (next).

This post has been edited 4 times, last edit by "leptis" (Nov 6th 2016, 11:42pm)


Miffyli

Symthic Developer

(6,088)

Posts: 3,650

Date of registration
: Mar 21st 2013

Platform: PC

Location: __main__, Finland

Reputation modifier: 16

  • Send private message

2

Sunday, November 6th 2016, 9:54pm

Ahmahgawd that second chart in second spoiler, asdlkamnflgksd I am even more convinced now you either are wizard with these colors and styles (or you are very patient when it comes to making these correctly).

But on the serious note: This indeed matches to what game seems to be after: instead of having tapfiring/short bursts as best option always, some weapons benefit from doing longer bursts but magdump is still not effective.

Two questions:
  1. What were the assumptions how player controls h recoil? Vertical can be learned and done "optimially", but horizontal is bit iffy.
  2. What distribution did you use for the spread? I think we had discussion about this but in Battlefield 1 the spread is not uniform inside circle, it has bit focus on the middle. See "Spread" here.
Links to users' thread list who have made analytical/statistical/mathematical/cool posts on Symthic:
  • 3VerstsNorth - Analysis of game mechanics in BF4 (tickrates, effects of tickrate, etc)
  • leptis - Analysis of shotguns, recoil, recoil control and air drag.
  • Veritable - Scoring of BF4/BF1 firearms in terms of usability, firing and other mechanics.
  • Miffyli - Random statistical analysis of BF4 battlereports/players and kill-distances. (list is cluttered with other threads).
Sorry if your name wasn't on the list, I honestly can't recall all names : ( . Nudge me if you want to be included

Posts: 7,793

Date of registration
: Feb 25th 2012

Platform: PC

Location: italy

Battlelog:

Reputation modifier: 19

  • Send private message

3

Sunday, November 6th 2016, 10:04pm

jesus!
that's way too much data, even if i tried i wouldn't understand all of that, it's almost like going back to school, my brain can't handle all of that information with barely any context at once, you need to give it in little snips that i can visualize better, graphs are decent but with so many specimens they just become confusing.

i mean, at least there's the "conclusions" part, but it's not like i understood why(the points told) that is...
"I'm just a loot whore."


stuff mostly unrelated to BF4 that interests nobody



bf4
on 13/05/2016
23rd M320FB user on pc(13/05/16)
rush mode score RANK:2794 TOP:2% OUT OF:215398
obliteration mode scoreRANK:994 TOP:1% OUT OF:159466
handgun medals RANK:2236 TOP:2% OUT OF:143874
longest headshot RANK:9512 TOP:4% OUT OF:257589
recon score RANK:10871 TOP:4% OUT OF:274899
general score per minute RANK:10016 TOP:4% OUT OF:294774

bf3
31/3/2012 4:58:

Headshot distance RANK:493* TOP:0%
Revives per assault minute RANK: 6019 TOP: 3%
Headshots / kill percentage RANK:25947 TOP:13%
MVP ribbons RANK:18824 TOP:11%

*= 6 if we not count the EOD BOT headshots

@kataklism

ARGUMENT DESTROYED 100

ENEMY KILLED [REASON] JSLICE20 100


WRITING SPREE STOPPED 500

link to full-size old avatar:
http://i.imgur.com/4X0321O.gif




  • "leptis" started this thread

Posts: 282

Date of registration
: Mar 17th 2015

Platform: PC

Battlelog:

Reputation modifier: 8

  • Send private message

4

Sunday, November 6th 2016, 11:12pm

Ahmahgawd that second chart in second spoiler, asdlkamnflgksd I am even more convinced now you either are wizard with these colors and styles (or you are very patient when it comes to making these correctly).

But on the serious note: This indeed matches to what game seems to be after: instead of having tapfiring/short bursts as best option always, some weapons benefit from doing longer bursts but magdump is still not effective.

Two questions:
  1. What were the assumptions how player controls h recoil? Vertical can be learned and done "optimially", but horizontal is bit iffy.
  2. What distribution did you use for the spread? I think we had discussion about this but in Battlefield 1 the spread is not uniform inside circle, it has bit focus on the middle. See "Spread" here.


LOL

On day in the future I will explain you how I can do that :). At the moment I follow the advice of a former boss: "Works may be good or bad...but always have to SEEM good". :) :) :)

I agree.. negative model revalues the long bursts. But it have some nuances. With Benet- Mercie Storm you can do it almost ad infinitum. But with more dirpersives LMGs it has a limit. It appear the "new microburst" of 7-8 bullet in the middle range. And at range (>50 m) with Madsen Trench... better 2-3 bullet microbursts.

Great issue the recoil control. I am planning a participative research about that. Perhaps next weekend.

Here recoil is not corrected. I think the horizontal recoil, due to its nature can't be corrected. The ramdom jumps to left and right don't follow any pattern. Only the cumulative effects of asymmetric horizontal recoil (deflection) can be corrected. And also the extremely rare cases of jumps left-left-left-left-left-left.... (that secuence - 5 jumps left- has only a probabilty of 1.56% in symmetric horizontal recoil and to have a significative value...less).

I use the model given in your link:

horizontal_dispersion = rand1^a * spread * cos(rand2)
vertical_dispersion = rand1^a * spread * sin(rand2)


That provide an uniform distribution of hits in the circle of spread (density per unit of surface). I have cheked it with 10.000 shots in vertcal and horizontal strips of the circle and it complies the Kolmogorov.Smirnoff test of uniformity.

If you calculate the density of "X" and "Y" coordinates (projections on axes) you get this:

Spoiler Spoiler



The very, very small bias to center is due to the circle geometry, not to density.

As you see 5% approx. in each class and I have used 24 classes soooo... as we say... white and in a bottle. :)

Incidentally, I think the explanation of spread is a bit confusing. Surely shotguns have barrel spread (point where effectively aim) like the other weapons (there is no reason why not). ...BUT when the pellets leave the gun have a dispersion exponent 1 which provides a strong bias to the center. There are two times to apply spread and can be different.

This post has been edited 2 times, last edit by "leptis" (Nov 6th 2016, 11:28pm)


Miffyli

Symthic Developer

(6,088)

Posts: 3,650

Date of registration
: Mar 21st 2013

Platform: PC

Location: __main__, Finland

Reputation modifier: 16

  • Send private message

5

Sunday, November 6th 2016, 11:16pm

BUT when the pellets leave the gun have a dispersion exponent 1 which provides a strong bias to the center. There are two times to apply spread and can be different.

Indeed, and this whole "two times the spread" for shotguns has been one of the tricky things to explain to everyone. I guess I should add a part about this on general info page.
And okey at the h-recoil.
Links to users' thread list who have made analytical/statistical/mathematical/cool posts on Symthic:
  • 3VerstsNorth - Analysis of game mechanics in BF4 (tickrates, effects of tickrate, etc)
  • leptis - Analysis of shotguns, recoil, recoil control and air drag.
  • Veritable - Scoring of BF4/BF1 firearms in terms of usability, firing and other mechanics.
  • Miffyli - Random statistical analysis of BF4 battlereports/players and kill-distances. (list is cluttered with other threads).
Sorry if your name wasn't on the list, I honestly can't recall all names : ( . Nudge me if you want to be included

  • "leptis" started this thread

Posts: 282

Date of registration
: Mar 17th 2015

Platform: PC

Battlelog:

Reputation modifier: 8

  • Send private message

6

Sunday, November 6th 2016, 11:21pm

I edited new precission about horizontal recoil. :)

  • "leptis" started this thread

Posts: 282

Date of registration
: Mar 17th 2015

Platform: PC

Battlelog:

Reputation modifier: 8

  • Send private message

7

Monday, November 7th 2016, 12:56pm

jesus!
that's way too much data, even if i tried i wouldn't understand all of that, it's almost like going back to school, my brain can't handle all of that information with barely any context at once, you need to give it in little snips that i can visualize better, graphs are decent but with so many specimens they just become confusing.

i mean, at least there's the "conclusions" part, but it's not like i understood why(the points told) that is...


lol

The path of science to move from "I think" or "it seems" to "IT IS" is not always easy. But when the path is completed everything seems clear and consistent.

Look at these images of density graphs, on the left a LMG with negative model and on the right an assault weapon with postive model. What do you see?:



The left weapon has a singular bullet (first) different and the rest is total uniformity -> Conclusion: accuracy is almost constant along the burst (controling the recoil up, obviously).

The right weapon has the same singular bullet (first) but successive bullets are becoming more and more dispersives. -> Conclusion: Do not shoot very long bursts.

And so...

At least I learned something from your post: I have to do more intermediate conclusions to "guide" reasoning, and not just "look at the image... and the conclusion isssssss....". :)

I will try to do better!!! :)

Posts: 7,793

Date of registration
: Feb 25th 2012

Platform: PC

Location: italy

Battlelog:

Reputation modifier: 19

  • Send private message

8

Monday, November 7th 2016, 2:45pm

Yup, that makes a lot more sense.
At least you're not like some programmers that for example say for adding a simple mod
"Yeah just add the mod"
how?
"C'mon it's easy just hack in the mainframe"
ok... and how do i do that?
"You'll get it, it's easy"

When really all you have to do is add a mod folder with your files in in the game folder, but not everyone knows that and i'll end up havingbto reinstall the game, pretty much the definition of "instructions unclear, got dick stuck in the fan"

Going back to the graphs, i assume that the lmg has lowar hitchance than the starting bullet because the next ones take in account recoil, while for the ar it's both sips and recoil, right? So in theory, if you used a bipod you could just fire till overheat and(as long as you don't overheat) you still wouldn't be doing anything wrong... right?
"I'm just a loot whore."


stuff mostly unrelated to BF4 that interests nobody



bf4
on 13/05/2016
23rd M320FB user on pc(13/05/16)
rush mode score RANK:2794 TOP:2% OUT OF:215398
obliteration mode scoreRANK:994 TOP:1% OUT OF:159466
handgun medals RANK:2236 TOP:2% OUT OF:143874
longest headshot RANK:9512 TOP:4% OUT OF:257589
recon score RANK:10871 TOP:4% OUT OF:274899
general score per minute RANK:10016 TOP:4% OUT OF:294774

bf3
31/3/2012 4:58:

Headshot distance RANK:493* TOP:0%
Revives per assault minute RANK: 6019 TOP: 3%
Headshots / kill percentage RANK:25947 TOP:13%
MVP ribbons RANK:18824 TOP:11%

*= 6 if we not count the EOD BOT headshots

@kataklism

ARGUMENT DESTROYED 100

ENEMY KILLED [REASON] JSLICE20 100


WRITING SPREE STOPPED 500

link to full-size old avatar:
http://i.imgur.com/4X0321O.gif




  • "leptis" started this thread

Posts: 282

Date of registration
: Mar 17th 2015

Platform: PC

Battlelog:

Reputation modifier: 8

  • Send private message

9

Monday, November 7th 2016, 9:00pm

Going back to the graphs, i assume that the lmg has lowar hitchance than the starting bullet because the next ones take in account recoil, while for the ar it's both sips and recoil, right? So in theory, if you used a bipod you could just fire till overheat and(as long as you don't overheat) you still wouldn't be doing anything wrong... right?


For the moment we don't know the impact that attachements do in the parameters of the weapon, especially what happens when the bipod is deployed.

In BF4 it had a huge impact: 90% reduction in Recoil Up, 75% in Recoil H, 85% in Recoil FSM, 95% in Spread Base and 90% in SpreadInc (LMGs).

Bipod turned LMGs into "sewing machines".

Surely the structure changes in BF1 because a reduction in negative SpreadInc worsens the weapon, not improving and we have to see what happens with the Spread FSM.

But I'll answer otherwise.

Here you have two weapons (no matter what) an accurate one and other dispersive in a 20 bullets burst . These are % of bullets within the body width at a distance, assuming perfect Recoil Up control.



With the firts you have a sustained 92% average in the last bullets. There is no incentive to stop because the first 8 bullets have an average of 86% and that is only because it includes the first bullet. Continuous burst.

The second one haven't such good results. The average of 10 to 20 is 64%. The first 8 bullets have an average of 74%. You have a 10% profit if you shoot in bursts of 8 bullet, and you can do it without Sppread penalty and without an appreciable loss of time because you do not have to wait to recover Spread. Continuous burst is not the optimal choice.

That is a question of marginal benefit.

If a bipod convert white weapon into yellow one ---> don't release the trigger.

And the improvement of the yellow weapon respect to the white one is only 30% in RecoilH and Spread ... imagine that could happen with reductions in the order of 70-90%.

My conclusion, I agree, BIPOD = CONTINOUS BURST in the regular distances of combat (not at 150 m)

Posts: 579

Date of registration
: Sep 3rd 2013

Platform: PC

Battlelog:

Reputation modifier: 8

  • Send private message

10

Monday, November 7th 2016, 9:38pm

This is because the spread is improving during the burst but the cumulative effect of the Horizontal Recoil compensates that resulting in a distribution almost equal with a very slight tendency to worsen from bullet 7-8th (due to constant spread and cumulative effect of the Horizontal Recoil).
So this is the key point, here.

The data is very convincing, but this effect of "cumulative recoil" just doesn't make much logical sense to me: how can horizontal recoil be cumulative? Can you try to explain it in layman's terms?