These forums have been archived and are now read-only.

The new forums are live and can be found at https://forums.eveonline.com/

Issues, Workarounds & Localization

 
  • Topic is locked indefinitely.
 

Energy weapon falloff accuracy
Author
Previous Topic Next Topic
Eremitae
Ministry of War
Amarr Empire
#1 - 2015-04-12 01:41:31 UTC
In short - i've noticed recently that energy weapon accuracy in falloff is quite terrible. I've decided to test it - to count the numbers of misses at optimal+falloff range. After several hours of tests(of course neither me nor my target were moving) i got some data - 75% of attacks missed the target, although according to evewiki the number of misses shouldn't exceed 50%. I was testing only small pulse lasers, but i think this can also be applied to other types of lasers.
Sobaan Tali
Caldari Quick Reaction Force
#2 - 2015-04-13 01:29:57 UTC
The 50% is never guaranteed. Hell, technically, 99.99% isn't even guaranteed either. On top of that, that's 50% chance of a hit for each shot on an individual bases (individual being the key word here), not "50% of all shots will make a hit". That's just how practically everyone assumes "chance" and "odds" work. Arithmetic is a heartless ***** like that.

Of course, computer driven RNG sucks as is anyways, so don't ever rely on it like that. It'll save you some headaches...and a few Aleve tablets.

"Tomahawks?"

"----in' A, right?"

"Trouble is, those things cost like a million and a half each."

"----, you pay me half that and I'll hump in some c4 and blow the ---- out of it my own damn self."
Eremitae
Ministry of War
Amarr Empire
#3 - 2015-04-13 01:54:02 UTC
But still on large numbers of shots it shouldn't deviate that much(almost 13k shots in my case).
Torgeir Hekard
I MYSELF AND ME
#4 - 2015-04-13 09:15:49 UTC
This is strange. Kadesh Priestess recently (about a week ago) tested range and tracking calculation mechanics for large ships using gatling pulses on falloff and didn't notice any deviations from the expected distribution.

How were you testing this?
Also can you try testing it with larger lasers to reduce positioning error influence on the outcome?
Sobaan Tali
Caldari Quick Reaction Force
#5 - 2015-04-13 12:07:28 UTC  |  Edited by: Sobaan Tali
Don't know what to tell you then. Anyone in exploration running anoms and sigs will tell you the same thing, that you just can't look at the concept of chance in a game like that and expect it to work that way. Fact of the matter is that the chance of 50% in this -- and any similar -- encounter is handled on a case-by-case system, where each shot is neither affected by any previous instance nor has any affects on any proceeding shots; just because the first five out of ten shots could miss doesn't mean all or even most of the remaining shots land a hit should that scenario occur. Each shot is its own event strictly handled and calculated independantly of all other shots.

You're on the right track, but you aren't thinking about it like a computer using RNG to generate simulated odds does.

"Tomahawks?"

"----in' A, right?"

"Trouble is, those things cost like a million and a half each."

"----, you pay me half that and I'll hump in some c4 and blow the ---- out of it my own damn self."
Eremitae
Ministry of War
Amarr Empire
#6 - 2015-04-13 13:32:44 UTC
Torgeir Hekard, i just fired at jita monument from optimal+0.9falloff range for several hours and counted hits and misses in log after that. No moving, to make sure that nothing will affect the results except falloff accuracy.
Torgeir Hekard
I MYSELF AND ME
#7 - 2015-04-13 13:36:01 UTC
Sobaan Tali wrote:
. Fact of the matter is that the chance of 50% in this -- and any similar -- encounter is handled on a case-by-case system, where each shot is neither affected by any previous instance nor has any affects on any proceeding shots; just because the first five out of ten shots could miss doesn't mean all or even most of the remaining shots land a hit should that scenario occur.


Ookay, go to wolframalpha and calculate the probability of 3/4 shots of 13000 missing with a single miss probability of 1/2.
Hint: I think it should be a CDF of binomial distribution with n=13000, p=1/2 and k=3250.
Hint hint: it gives me 3,48e-741. That's not bad RNG. That's either a methodology error, or the guy just won all the lotteries in the Universe.
Logan Revelore
Symbiotic Systems
#8 - 2015-04-13 15:54:59 UTC
Sobaan Tali wrote:
The 50% is never guaranteed. Hell, technically, 99.99% isn't even guaranteed either. On top of that, that's 50% chance of a hit for each shot on an individual bases (individual being the key word here), not "50% of all shots will make a hit". That's just how practically everyone assumes "chance" and "odds" work. Arithmetic is a heartless ***** like that.

Of course, computer driven RNG sucks as is anyways, so don't ever rely on it like that. It'll save you some headaches...and a few Aleve tablets.


If you believe that what you write is the explanation for the OP getting 75% miss rate after hours of continuous data gathering, then I suggest that you reevaluate.

There's no reason that the OP shouldn't get 50% miss rates when gathering data over several hours. Either it's a bug or the OP misunderstood something or set up a faulty test bench.
Torgeir Hekard
I MYSELF AND ME
#9 - 2015-04-13 16:49:02 UTC  |  Edited by: Torgeir Hekard
Eremitae wrote:
Torgeir Hekard, i just fired at jita monument from optimal+0.9falloff range for several hours and counted hits and misses in log after that. No moving, to make sure that nothing will affect the results except falloff accuracy.

Hmmm. I will probably try tomorrow on SiSi. Only with ships, not monuments. The thing is, I'm not sure that distance to structures is calculated the same way as distance to ships for the purpose of range to target calculation.

We already established that the distance for range is calculated differently than the distance for tracking even for ships, so maybe the problem is not your lasers, but that you are shooting a structure.
Logan Revelore
Symbiotic Systems
#10 - 2015-04-13 17:32:52 UTC  |  Edited by: Logan Revelore
Sobaan Tali wrote:
Don't know what to tell you then. Anyone in exploration running anoms and sigs will tell you the same thing, that you just can't look at the concept of chance in a game like that and expect it to work that way. Fact of the matter is that the chance of 50% in this -- and any similar -- encounter is handled on a case-by-case system, where each shot is neither affected by any previous instance nor has any affects on any proceeding shots; just because the first five out of ten shots could miss doesn't mean all or even most of the remaining shots land a hit should that scenario occur. Each shot is its own event strictly handled and calculated independantly of all other shots.

You're on the right track, but you aren't thinking about it like a computer using RNG to generate simulated odds does.


I'm impressed. You manage to address and, seemingly, understand one of the usual misconceptions of probability having memory, yet still you show that you really don't understand the probability laws governing this test.

With a 50% miss rate, yes sure you could get 75% miss rate if the sample size is small, but if the sample size is 13K shots, then the chance of getting a miss rate of 75% is actually:

Matlab gave up here, couldn't handle these sizes of numbers in the standard library.

So let me try again, if we assume 100 shots, the probability of getting a 25 % hit rate with a 50 % hit rate is:

1.9131*10^(-5) % <--- VERY small. The probability for the 13000 sample pool was astronomically small.


Formula used is: ( (n!) / (k!*(n-k)!) ) / (2^n), where n=13000, k=n*0.25
Logan Revelore
Symbiotic Systems
#11 - 2015-04-13 17:37:31 UTC
Sobaan Tali wrote:
...That's either a methodology error, or the guy just won all the lotteries in the Universe.


My thoughts exactly.
Sobaan Tali
Caldari Quick Reaction Force
#12 - 2015-04-13 23:18:41 UTC
Wait. 13000? Hm, I didn't catch that part. I had assumed you had committed a smaller sample size of test shots. In that case, that does seem awfully weird for the results to deviate that much. I don't know, math was never my strong side...I could never get the muscle-memory of the various rules, functions, and equations down long enough to remember them off the top of my head.

If after something like 13000 shots using the same scenario without adverse interruptions, the results come up with such a consistently off track number of failures, then something else is afoot.

I stand corrected then. This is strange indeed. As someone already pointed out, either something may be contaminating the test itself and altering the rules here, or you need to grab a scratch-off from somewhere.

"Tomahawks?"

"----in' A, right?"

"Trouble is, those things cost like a million and a half each."

"----, you pay me half that and I'll hump in some c4 and blow the ---- out of it my own damn self."
Sobaan Tali
Caldari Quick Reaction Force
#13 - 2015-04-13 23:32:54 UTC
I'm curious about something. So I haven't read anything wrong, you said you're seeing about 75% misses at Optimal+Falloff, right? And that should be, assuming I understand how optimal and falloff works in Eve, 50% instead, right?

Just for ***** and giggles, what if you alter the range so that you SHOULD be getting what you're getting now? In other words, if you tweak the rules of the experiment and observe how the results change, maybe that might hint at what is causing this.

"Tomahawks?"

"----in' A, right?"

"Trouble is, those things cost like a million and a half each."

"----, you pay me half that and I'll hump in some c4 and blow the ---- out of it my own damn self."
Torgeir Hekard
I MYSELF AND ME
#14 - 2015-04-14 06:16:54 UTC
Okay, I've tested a little, and can't see any problems.

I used a punisher with a gatling pulse laser and MF versus a phoenix. Falloff 4350 (at this point overview still shows distance in meters, not kms). Range between ships was 4348.
The sample size is relatively small at this point, but the result does not deviate that much from the expected outcome (259 hits 275 misses at the time of writing this).

There's one thing of note though. If you put several guns into one group, you get much better hit chance but worse hit quality (apparently representing the fact that not all guns in the group hit, but at leas some of them do). So the testing must be done with a single gun.
Eremitae
Ministry of War
Amarr Empire
#15 - 2015-04-14 07:01:47 UTC  |  Edited by: Eremitae
You need to test at optimal+falloff, not just falloff. And btw since you were at about optimal + 0.25*falloff (according to at least my skills) you should have much higher hit chance than 50%.
Torgeir Hekard
I MYSELF AND ME
#16 - 2015-04-14 20:47:23 UTC
By falloff I mean, of course, optimal+falloff. It's just called falloff on the tooltip. That's 3150 optimal + 1200 falloff or something like that.
granny Lina
Viziam
Amarr Empire
#17 - 2015-04-23 12:15:00 UTC  |  Edited by: granny Lina
im intrigued to test this myself if you tell me how to collect the data.
my experience and knowledge about RNG tells me that expected result can deviate quite a lot even in large sample

edit:
im no longer intrigued.