Accuracy bug layer

Discussions about AI types, units, tactics & strategy.
User avatar
Iluvalar
Regular
Regular
Posts: 1828
Joined: 02 Oct 2010, 18:44

Re: Accuracy bug layer

Post by Iluvalar »

Stratadrake wrote: At a rated accuracy of 75%, Mantis and Python took damage 100% of the time (a.k.a. 100% of misses), again no difference between them.
Thank you for the test, I have a full SVG showing all this... waiting for the devs to allow svg as attachement... It's here now.
https://docs.google.com/open?id=0B9FqMM ... kpXUF9PdUU

That tells me how the rouding error affect my chart... However I still need the same test with tiger to confirm that since it is 2px smaller than mantis, the accuracy stay at 75%.

I believe you are overevaluating a little the effect "miss behing" that should gives ~1/4 of the miss as hit. It would mean 62.5% when 50% of accuracy. And 82% when announcing 75%. Otherwise your experiment exactly confirm the source code :lol2: .
Heretic 2.3 improver and proud of it.
User avatar
Stratadrake
Trained
Trained
Posts: 197
Joined: 07 Sep 2008, 09:43
Location: Pacific NW
Contact:

Re: Accuracy bug layer

Post by Stratadrake »

I believe you are overevaluating a little the effect "miss behing" that should gives ~1/4 of the miss as hit. It would mean 62.5% when 50% of accuracy. And 82% when announcing 75%. Otherwise your experiment exactly confirm the source code :lol2:
It was noted that missing behind the target has little to no effect on artillery or VTOL weapons, only ground-based direct-fire ones.

Okay, I've tweaked my testing method and achieved the following results:
Given: 50% base accuracy, Wheeled Truck vehicles, 400 attacks
-----
Bug: 75%
Viper: 64%
Leopard: 74%
Retaliation: 78%
-----
Scorpion: 77%
Cobra: 75%
Panther: 72%
Retribution: 80%
-----
Mantis: 83%
Python: 88%
Tiger: 82%
Vengeance: 96%
Dragon: 98%
-----
Note that it seems virtually impossible to miss a Vengeance or Dragon even with weapons as low as 50% accuracy. :(
Strata @dA, @FAC
User avatar
Iluvalar
Regular
Regular
Posts: 1828
Joined: 02 Oct 2010, 18:44

Re: Accuracy bug layer

Post by Iluvalar »

What was the distance of the target ? It seem damn close here... Or else, the "shoot behind" part of the bug have more effect than I thought.
Heretic 2.3 improver and proud of it.
User avatar
Stratadrake
Trained
Trained
Posts: 197
Joined: 07 Sep 2008, 09:43
Location: Pacific NW
Contact:

Re: Accuracy bug layer

Post by Stratadrake »

About six squares or so. But doesn't that have zero effect on shot scatter? Range determines when you can attack; scatter is calculated when said attack is actually executed.
Strata @dA, @FAC
User avatar
Iluvalar
Regular
Regular
Posts: 1828
Joined: 02 Oct 2010, 18:44

Re: Accuracy bug layer

Post by Iluvalar »

No, i fixed the miss behind angle to 90° in the fixes for simplicity, but it's actually affected by the distance a little.

Nothing In the bugs I found could explain a ratio of false misses higher than 50% (50%=>75%). If your data are true, it's another problem...
Attachments
Eclipse angle behind the target (green) reducing with distance
Eclipse angle behind the target (green) reducing with distance
angBehind.png (992 Bytes) Viewed 5359 times
Heretic 2.3 improver and proud of it.
User avatar
Stratadrake
Trained
Trained
Posts: 197
Joined: 07 Sep 2008, 09:43
Location: Pacific NW
Contact:

Re: Accuracy bug layer

Post by Stratadrake »

It is impossible to tell whether a "false miss" was targetted "inside" or "behind" the target hitbox because the projectile explodes the moment the game detects a hitbox collision - zero projectiles ever make it to the behind of the target.

It's mathematically possible to identify how many of these misses landed behind the target's hitbox (based on known size, scatter, and range to target), but the steps get a little complicated....
Strata @dA, @FAC
User avatar
Iluvalar
Regular
Regular
Posts: 1828
Joined: 02 Oct 2010, 18:44

Re: Accuracy bug layer

Post by Iluvalar »

I know, what I say it's that it's impossible to explain the data you got only by the miss behind bug. There is another undocumented bug of the accuracy system yet to find...
Heretic 2.3 improver and proud of it.
User avatar
Stratadrake
Trained
Trained
Posts: 197
Joined: 07 Sep 2008, 09:43
Location: Pacific NW
Contact:

Re: Accuracy bug layer

Post by Stratadrake »

No, I don't think so. If I take the figures you gave in previous posts for hitbox and scatter radius determination, then let's break down the results:

1 - Python took 88% of hits from a 50% base accuracy weapon. So 38% of all shots fired (a.k.a. 76% of misses) were false hits.
2 - If Python has a hit radius of 54 and the weapon has a scatter radius of 100, then (because actual scatter is determined by a radius/angle offset from target position) we can expect 54% of misses to land inside the hitbox radius (regardless of angle) and score a false hit.
3 - Misses targetted "behind" the hitbox will impact the target en route and also score a false hit.
4 - If only 54% of misses should have hit, but 76% actually did, this is a +22% difference and it represents the number of misses that were aimed behind the hitbox.

So in total, of 400 shots fired at the Python:
- 200 (50%) were expected to hit the target (and did)
- 108 (27%) were expected to miss the target, but landed inside the hitbox and scored a hit
- 44 (11%) were expected to miss the target, but landed behind the hitbox and scored a hit en-route
- 48 (12%) actually did miss the target.

PS: Here's some trig for you to crunch on. Given a range-to-target (x), hitbox radius (r) and maximum scatter radius (s), then:

- Target radius expressed as an angle in the attacker's FOV: = arcsin(r/x) (total target diameter relative to attacker's FOV is twice this)
- Angular radius that represents an area always "behind" the target, regardless of scatter distance: = (90 - arccos(r/s) - arcsin(r/x) )
- Any scatter angle that exceeds (90 + arcsin(r/x)) is either aside or in front of the target, regardless of scatter distance
- Any scatter angle between the above two values may or may not land behind the target, depending on scatter distance

I can illustrate it in a diagram if you need it.
Strata @dA, @FAC
User avatar
Iluvalar
Regular
Regular
Posts: 1828
Joined: 02 Oct 2010, 18:44

Re: Accuracy bug layer

Post by Iluvalar »

Well then, I guess you are right and I was wrong... There is a scattering effect. Even if I dont know where it come from...

And that scattering effect just make even more misses hit... XD
Heretic 2.3 improver and proud of it.
Post Reply