Rocket scatter slightly weird(?)

[Blackadder]

So, this may be just me being thick...

Based on JoINrbs' rocket scatter calculator:
https://docs.google.com/spreadsheets/d/ ... edit#gid=0

the chance of a rocket scattering to any individual square can be higher than the chance of a direct hit (or any individual tile with a scatter of a smaller distance) in some (most) cases. I understand this is a direct result of the rolling of the scatter dice.

Now, I'm not suggesting this be changed but...for the sake of starting a thought experiment:

If I'm throwing a dart at a dartboard and aiming for the bullseye then, in the absence of biased throwing, the bullseye is the most likely spot for my dart to land of all the individual spots it could land (even if I'm lousy at darts). This isn't to say that it is more likely I get no scatter than some scatter. Indeed, I find it very reasonable that the amount of scatter would have a peak in it's probability distribution at some non-zero number. But this only says that it is more likely that the dart lands at a non-zero distance from the bullseye, not that it is more likely that the dart lands at that distance and in a specified direction from the bullseye.

I realise that changing this to something more akin to a 2D Gaussian distribution (or other uni-modal distribution) may mean rockets are then too accurate and so balance may dictate either further changes or...none at all.

Just thought this may be an interesting discussion (and I may be totally wrong in which case...let me know).

Cheers

[Alketi]

As nice as the HUD tooltip overlay is, where it tells you a +/- value, it's unclear what that value really means.

When you see something that's +/-, everywhere else in life that implies a RANGE, i.e. MAX/MIN.

Except, in the case of a LW2 rocket.

And so, when someone has a +/- 2.36 scatter and they center the diameter 8 rocket on a target and then they MISS the target entirely, they don't have a very good time of things.

So, what does the +/- value represent? 1 sigma? 50% chance? 68% chance? How would anyone who's only looking at the HUD know? This UI nicety unfortunately requires yet another secret decoder ring.

I happened to be watching Mal's recent mission and he ran into this exact scenario.

Video (a blind user test of the UI):
https://youtu.be/FgpnvQNKd9Y?t=7m6s

[fowlJ]

As nice as the HUD tooltip overlay is, where it tells you a +/- value, it's unclear what that value really means.

It's the average scatter. The rocket rolls one aim check for each tile of possible scatter to see how far from the target it lands, so the scatter display gives you [chance of missing] * [maximum scatter]. I'm not sure why they represent it with the +/- (though off the top of my head I can't think of a better way to do it that doesn't involve tossing a bunch of ugly text into the middle of the screen), but the idea is for it to mean 'the rocket will land more-or-less this far away, most of the time'.

[Dlareh]

Heh, Mal's video perfectly illustrates how misleading LW2's rocket interface is.

That's one side of the coin of why people new to LW2 quickly come to hate technicals and are convinced they're weak...

(the other side is they don't know how to use the flamer effectively and don't immediately realize that burning incapacitates enemies)

[JulianSkies]

If I'm throwing a dart at a dartboard and aiming for the bullseye then, in the absence of biased throwing, the bullseye is the most likely spot for my dart to land of all the individual spots it could land (even if I'm lousy at darts). This isn't to say that it is more likely I get no scatter than some scatter. Indeed, I find it very reasonable that the amount of scatter would have a peak in it's probability distribution at some non-zero number. But this only says that it is more likely that the dart lands at a non-zero distance from the bullseye, not that it is more likely that the dart lands at that distance and in a specified direction from the bullseye.

See, that is the point, the bullseye is not in fact going to be more likely for you to hit than any other particular point, in fact i'd say it is the least likely point.
Why? Because there is in fact no unbiased aiming, everyone has their own particular deviation in how they target, be it because of the way their arm works, their depth perception, their hand-eye coordination. If you have a very steady hand it's quite likely you will always deviate the save value in the same direction, making the deviation more predictable (Fire in the Hole), however if you have the right skill to know that and aim at a different place than the bullseye so that your deviated shot goes in the center then and only then you will have a higher chance to hit it (high Aim).

Also Alketi, i'm not sure what "More or less two point twenty three tiles" is unclear? Like that sounds incredibly clear to me, and that rocket sadly deviated, in fact, all way to the lower-right all those two point three tiles in the exact way to miss entirely, very low chance- But a chance nontheless and exactly as advertised in the UI. That's what we call "That's XCOM, baby" I guess?

[DonCrabio]

Rocket scatter completely weird, up to the point where its use is absolutely unreliable. I even tried to take first skill instead of Roust or Suression, but it make no difference. This is completely block rocketeer build for me.

And I can't see a reason for such scatter on not-so-great single use ability.

[Blackadder]

If I'm throwing a dart at a dartboard and aiming for the bullseye then, in the absence of biased throwing, the bullseye is the most likely spot for my dart to land of all the individual spots it could land (even if I'm lousy at darts). This isn't to say that it is more likely I get no scatter than some scatter. Indeed, I find it very reasonable that the amount of scatter would have a peak in it's probability distribution at some non-zero number. But this only says that it is more likely that the dart lands at a non-zero distance from the bullseye, not that it is more likely that the dart lands at that distance and in a specified direction from the bullseye.

See, that is the point, the bullseye is not in fact going to be more likely for you to hit than any other particular point, in fact i'd say it is the least likely point.
Why? Because there is in fact no unbiased aiming, everyone has their own particular deviation in how they target, be it because of the way their arm works, their depth perception, their hand-eye coordination. If you have a very steady hand it's quite likely you will always deviate the save value in the same direction, making the deviation more predictable (Fire in the Hole), however if you have the right skill to know that and aim at a different place than the bullseye so that your deviated shot goes in the center then and only then you will have a higher chance to hit it (high Aim).

OK, but if I have a strict bias it would mean it would scatter asymmetrically which is not what rocket scatter is modelling I think. Also, training with a weapon I would imagine means you compensate for your bias.

Please note before this gets sidetracked, I am NOT saying that rocket scatter is a bad design...it's like a regular shot - a rocket can miss. What I'm querying is that the chance of scattering in any particular direction (to the left for example) and distance can have a higher probability than a direct hit.

[JulianSkies]

If I'm throwing a dart at a dartboard and aiming for the bullseye then, in the absence of biased throwing, the bullseye is the most likely spot for my dart to land of all the individual spots it could land (even if I'm lousy at darts). This isn't to say that it is more likely I get no scatter than some scatter. Indeed, I find it very reasonable that the amount of scatter would have a peak in it's probability distribution at some non-zero number. But this only says that it is more likely that the dart lands at a non-zero distance from the bullseye, not that it is more likely that the dart lands at that distance and in a specified direction from the bullseye.

See, that is the point, the bullseye is not in fact going to be more likely for you to hit than any other particular point, in fact i'd say it is the least likely point.
Why? Because there is in fact no unbiased aiming, everyone has their own particular deviation in how they target, be it because of the way their arm works, their depth perception, their hand-eye coordination. If you have a very steady hand it's quite likely you will always deviate the save value in the same direction, making the deviation more predictable (Fire in the Hole), however if you have the right skill to know that and aim at a different place than the bullseye so that your deviated shot goes in the center then and only then you will have a higher chance to hit it (high Aim).

OK, but if I have a strict bias it would mean it would scatter asymmetrically which is not what rocket scatter is modelling I think. Also, training with a weapon I would imagine means you compensate for your bias.

Please note before this gets sidetracked, I am NOT saying that rocket scatter is a bad design...it's like a regular shot - a rocket can miss. What I'm querying is that the chance of scattering in any particular direction (to the left for example) and distance can have a higher probability than a direct hit.

I don't think that is strictly true on a good rocket hit (Fire in the Hole, good aim, pre-move), those who train specifically for rockets can definitely have less than 1 tile of scatter, and rockets are immense as it is. People often seem to think taking hail mary post-move shots are a good idea when those are really last resort moves.

[Tuhalu]

The ideal Rocket Technical has 75 aim at rookie and 91 aim at master sergeant. Stack a Superior Perception PCS on top of that (with the Integrated Warfare GTS training) for +9 or +10 aim and you now have 100 aim. This would make your Rockets never miss, with or without Fire In the Hole and with or without movement.

The main idea here is that you can't expect a Rocket Technical to be able to hit a perfect shot without investing in their aim. For most of their levelling time, you'll need to rely on Fire in the Hole and firing with your technicals first action to produce semi-reliable results (producing 2 tile max scatter). You might even have to center your rocket on the one thing you absolutely must hit.

[Alketi]

Also Alketi, i'm not sure what "More or less two point twenty three tiles" is unclear? Like that sounds incredibly clear to me, and that rocket sadly deviated, in fact, all way to the lower-right all those two point three tiles in the exact way to miss entirely, very low chance- But a chance nontheless and exactly as advertised in the UI. That's what we call "That's XCOM, baby" I guess?

If the +/- 2.36 tiles in that example truly was the MAX deviation for the given shot based on the actual center pixel of the blast, then I agree it's perfectly clear.

Here are the screenshots from Mal's rocket. http://imgur.com/a/ejAsY

Did he miss 2.36 tiles on the diagonal which was just enough to miss the group entirely? I agree that it's possible, as we can't perfectly see all the tiles.

However, fowlJ said upthread that the number actually represents the AVG scatter, not the MAX.

chrisb in another thread said it's 70%.

So, is it AVG or MAX or some SIGMA calculation?

+/- implies MAX.

If it's MAX, then it's good by me. The question then becomes, is it calculated correctly?

[DonCrabio]

75 AIM from the start for Technical, seriously? You can have low aim Grenadier from the very start of the game, having perfect accuracy and 3 charges.

Maybe some day I will have more good AIM soldiers than I need for Rangers, Sharpshooters, Gunners and Shinobies and spare Superior Perception PCS, so I will try rocket spec just for entertainment.

[Alketi]

So, using joinrbs calculator I see that the HTML text can't be MAX. In fact, the code refers to the overlay as "AverageScatterText", which matches fowlJ's comments.

IMO, the +/- in the overlay should be removed and it should be labeled as "AVG: 2.36 Tiles", per the example above. Then the player would at least not blame the game for lying to them.

Code: Select all

static function float GetExpectedScatter(XComGameState_Unit Unit, vector TargetLoc)
{
	local float ExpectedScatter, EffectiveOffense;
	local int TileDistance;

	TileDistance = TileDistanceBetween(Unit, TargetLoc);
	EffectiveOffense = GetEffectiveOffense(Unit, TileDistance);
	ExpectedScatter = (100.0 - GetEffectiveOffense(Unit, TileDistance))/100.0 * float(GetNumAimRolls(Unit));
	`LWTRACE("ExpectedScatter=" $ ExpectedScatter $ ", EffectiveOffense=" $ EffectiveOffense $ ", TileDistance=" $ TileDistance);
	return ExpectedScatter;
}

[JulianSkies]

75 AIM from the start for Technical, seriously? You can have low aim Grenadier from the very start of the game, having perfect accuracy and 3 charges.

Maybe some day I will have more good AIM soldiers than I need for Rangers, Sharpshooters, Gunners and Shinobies and spare Superior Perception PCS, so I will try rocket spec just for entertainment.

... What do you need aim in shinobi for???
That aside, even the absolute average rookie, which is 65 aim, will have 81 aim at end of career. With FITH at range 18 and pre move, about rifle engagement range (no penalty range is 19), that's still 78% chance to deviate max 1, very accurate I'd say for that much terrain leveling.

Also, interestingly, technicals are really, really tanky. I tend to use them as secondary riflemen.

[JulianSkies]

So, using joinrbs calculator I see that the HTML text can't be MAX. In fact, the code refers to the overlay as "AverageScatterText", which matches fowlJ's comments.

IMO, the +/- in the overlay should be removed and it should be labeled as "AVG: 2.36 Tiles", per the example above. Then the player would at least not blame the game for lying to them.

Code: Select all

static function float GetExpectedScatter(XComGameState_Unit Unit, vector TargetLoc)
{
	local float ExpectedScatter, EffectiveOffense;
	local int TileDistance;

	TileDistance = TileDistanceBetween(Unit, TargetLoc);
	EffectiveOffense = GetEffectiveOffense(Unit, TileDistance);
	ExpectedScatter = (100.0 - GetEffectiveOffense(Unit, TileDistance))/100.0 * float(GetNumAimRolls(Unit));
	`LWTRACE("ExpectedScatter=" $ ExpectedScatter $ ", EffectiveOffense=" $ EffectiveOffense $ ", TileDistance=" $ TileDistance);
	return ExpectedScatter;
}

To be honest the game doesn't lies about max scatter, says right in the ability what it's is. In fact, in that linked screenshot earlier you can see it says "Max scatter: 4" at the bottom of the screen.

[DonCrabio]

... What do you need aim in shinobi for???
That aside, even the absolute average rookie, which is 65 aim, will have 81 aim at end of career. With FITH at range 18 and pre move, about rifle engagement range (no penalty range is 19), that's still 78% chance to deviate max 1, very accurate I'd say for that much terrain leveling.

Also, interestingly, technicals are really, really tanky. I tend to use them as secondary riflemen.

I stopped playing stealth missions to somewhat "emulate" 1.3, so I use only sword Shinobies, misses and grazes very punishing for them, so they need a good aim just to stay alive and be effective.

Average AIM Technical will be unreliable most of his career. And at MSGT level he will be as good as MSGT Grenadier.

I don't say it can't be good. If some day I will have high AIM rookie with decent mobility not needed for more obvious duties, I will give it a shot.

[LordYanaek]

IMO, the +/- in the overlay should be removed and it should be labeled as "AVG: 2.36 Tiles", per the example above. Then the player would at least not blame the game for lying to them.

I won't argue against a change that makes things even more obvious than they already are but the first time i saw that +/- i immediately read it more or less x tiles and translated that into approximately x tiles which is the exact same as AVG.

I've never heard of +/- being used to tell anything other than "more or less" so this entire discussion looks like complaining for the sake of complaining but at least the change you are suggesting is sure to cut off any useless complain so it looks like a good idea.

[RantingRodent]

IMO, the +/- in the overlay should be removed and it should be labeled as "AVG: 2.36 Tiles", per the example above. Then the player would at least not blame the game for lying to them.

The scatter on each axis is calculated separately, so I think this is what this is attempting to indicate. It's the average scatter on each axis, not the average scatter overall.

[Blackadder]

So, this seems to have moved topic to whether the indicator is a good one or not. Just on the original point, I am not complaining about the mechanic as is, merely pointing out that as implemented it causes a slightly odd quirk* which may be either insignificant, an unintended byproduct of a concise and easy to code mechanic, or intended for a reason I can not see.

Put another way, I'm a maths geek and was interested in this from a modelling approach, NOT a gameplay/balance/"this is unfair!" standpoint.

* the quirk being that, sometimes, some tiles away from the centre are each more likely to be hit than the one aimed for.

[Tuhalu]

IMO, the +/- in the overlay should be removed and it should be labeled as "AVG: 2.36 Tiles", per the example above. Then the player would at least not blame the game for lying to them.

The scatter on each axis is calculated separately, so I think this is what this is attempting to indicate. It's the average scatter on each axis, not the average scatter overall.

This is not true. The game determines the distance scattered by making a number of accuracy rolls equal to the max scatter and then randomly determines the direction of the scatter. Finally, it translates that distance and direction into a tile location.

The "average scatter" indicator is simply the probability you'll fail the modified accuracy roll multiplied by the maximum number of tiles a rocket can scatter. This is basically the peak of the normal distribution that describes the full set of outcomes.

In any case, because the direction of scatter is random, the vector created by the combination of direction and distance will almost never point exactly at the center of a tile. This means the game has to find the nearest tile center and place the rocket there. This can cause the actual distance between the point targetted and point actually hit to be a non-integer value.

That aside, I noticed something else that nobody is really talking about. The aim modifiers for movement and range. If you have 1 or fewer action points when firing, you get a movement penalty of +2 on your max scatter AND -30 to all your accuracy rolls. You also have range penalties to accuracy past 9 tiles of -3 per tile (capped at -50 for 26 tiles, which is max distance for rockets with Javelin Rockets perk).

[fowlJ]

* the quirk being that, sometimes, some tiles away from the centre are each more likely to be hit than the one aimed for.

I don't think this is really that strange, ultimately. If you take, say, a 30% shot with a soldier's gun, you're going to miss more than twice as often as you hit, so why not with a low chance rocket?

[Jadiel]

* the quirk being that, sometimes, some tiles away from the centre are each more likely to be hit than the one aimed for.

I don't think this is really that strange, ultimately. If you take, say, a 30% shot with a soldier's gun, you're going to miss more than twice as often as you hit, so why not with a low chance rocket?

True, but if I found that the best way to maximise the probability of hitting a head shot was to point at the targets feet, I'd ask for a different gun...

[Zyxpsilon]

Once you start forgetting about the inherent Maths or various implicit code mechanics that enforce any given range of scattering effects.. all that remains is the generic impression of a modern weapon that jitters constantly for every good or bad reasons you can think of.

Gameplay wise though -- it IS extremely frustrating, for me!

Who the hell would start relying on highly (or not, btw) ranked soldiers with a SINGLE shot action device like that new LW2-Gauntlet specially tied with the Technical Class which misses more often than not??

In the indirect context of our dear Sci-Fi warfare.. it just doesn't make any friggin' sense. Well sure -- RNG is supposed to be a relative factor.. but frankly, it shouldn't apply to the Rockets at all. It's perfectly logical that a very slight accuracy check could be thrown in for rational HUD conditions with respect for the usual % odds of any other weapons. Yet, in the end --and after a bunch of persistent misses-- you suddenly must admit a Technical is a waste of combat advantages that really should be reliable enough.

Thus (as with LW1).. i dug deep & wide into the code to fix that messy absurdity (IMHO). Movement or distances included, the fact is simple == ONE rocket HIT ---can--- mean much less injuries to nearby squadmates once the Aliens begin spreading around as perfectly fine reactions to the minimally accurate blast (some of which would take immediate shots too - remember?).

Sooo here's my custom editing tricks to the XComLW_SoldierSkills.INI file for lines #118+119+120.

Code: Select all

MOVEMENT_SCATTER_AIM_MODIFIER=0		;-30
MOVEMENT_SCATTER_TILE_MODIFIER=0	;2
NUM_AIM_SCATTER_ROLLS=1		;4

Stop yelling - start playing.

[LordYanaek]

True, but if I found that the best way to maximise the probability of hitting a head shot was to point at the targets feet, I'd ask for a different gun...

That comparison is overlooking two key facts about rockets

• They scatter in a random direction
• They have an AoE

The result is that even when it's more likely your shot will end up two tiles away from the targeted tile (bad aim, no FitH), you're still better targeting directly the tile you want to hit because it maximizes the chance that tile will still be in the AoE. If you were to target a tile that's 2 tiles to the left of your target because the best probability is to hit two tiles away, you are as likely to have your rocket move further left and finally hit a tile that's 4 tiles away from the one you wanted to hit.

To use your comparison, maybe you have a slightly better chance of hitting a head shot if you target the feet, but you are also much more likely to totally miss your target (or even sometimes to hit your friend), whereas if you were targeting the head you might hit the chest or an arm instead (which is still better than not hitting at all).

In the indirect context of our dear Sci-Fi warfare.. it just doesn't make any friggin' sense.

Who said the Gauntlet was a perfectly controlled piece of tech (BTW which perfectly controlled piece of tech never have any failure?). It looks like some quickly assembled odd job to me

Realism apart (i always try to adapt my view of realism to the game) your changes will definitely change the balance of the class and leave them with a totally useless perk (which was about the only useful one on that rank!)
When i first played with techs i was like you and only used the flamer but now that i tried to play with them more, i'm getting better at using rockets efficiently.
The first trick is to mentally reduce the displayed AoE so you don't consider you'll hit everything. If you do, great, it was a lucky shot but i don't rely on it more than i would have relied on a direct kill from a vanilla repeater. Once you do this you'll learn to appreciate how much cover you can reliably blow and plan on this rather than try to include as many enemies as possible in the AoE.

You should probably reduce the AoE of the rockets with your accuracy changes if you want to keep the balance.

[Zyxpsilon]

OH -- but my objective wasn't to alter the generic balance involved in using Rockets.

The only "principle" my edit touches is the penalties for the post-movement aspect & the number of Rolls that add too much unpredictable results. Perks still apply under such conditions -- only better.

[Blackadder]

* the quirk being that, sometimes, some tiles away from the centre are each more likely to be hit than the one aimed for.

I don't think this is really that strange, ultimately. If you take, say, a 30% shot with a soldier's gun, you're going to miss more than twice as often as you hit, so why not with a low chance rocket?

Not my point, it's more subtle:

1) Yes, it is more likely and should be more likely that the rocket should miss than hit the tile aimed for.
2) Further it is reasonable that the total chance of scattering any given distance (added up(/integrated) over every different possible scatter direction) can be larger than a zero scatter.
3) What seems odd is that the chance of scattering a non-zero distance in an individual direction can be higher than the chance of zero scatter.

(3) implies that the soldier has a very odd sort of 'bad aim' where arbitrary tiles which aren't the one aimed for will each be more likely than the one aimed for. Bad aim would be more intuitively modeled instead (I believe at least) where the variance of the shot increases while the profile of the shot distribution remains qualitatively similar (i.e. if it were a normal distribution for high aim then it is a normal distribution but with higher variance for low aim - not a different distribution entirely which is what we appear to generate here).

(and again not railing here for it to change as ultimately it achieves the most important objective of implementing rocket scatter and thus not just a 'I click this button and everything dies' weapon)