Forums › English Language Forums › General › The Arsenal

Search

Just how special IS that damage?

11 replies [Last post]
Wed, 01/04/2012 - 22:49
Chronovore's picture
Chronovore

As a result of the conversation about split-special-damage weapons, I've been taking a closer look at how much of a bonus/penalty damage types get vs. the different monsters. My first look at it is at http://forums.spiralknights.com/en/node/34185?page=5

I have been using the weapon damage values from here: http://forums.spiralknights.com/en/node/7447

Since then, I've come up with a pretty good idea of how much damage weapons do to things they're strong/weak against in a general sense. Against resistant enemies, special damage seems to do the following, where A and B are parameters determined in some fashion by the depth and possibly the weapon's * rating:
Adjusted damage = (base - A) * B + A
So the first A damage is unaffected by the resistance, and the rest is reduced linearly by it.
For depth 28 and 5* weapons, it looks like A = 20, B = 0.15 gives good results.

For vulnerable enemies, the picture is a little more complicated. It looks like this: Below a certain threshold, the amount of bonus damage added looks like a parabola that opens downwards. Above that threshold, the amount of bonus damage added is constant. So with four parameters C, D, E, and F we can match the damage to the following function:
Adjusted damage = base + MAX(C * H(base - D), E - (base - D)^2 / F)
(where H is the unit step function)
C is the constant bonus added after the base damage reaches a certain value, D is the base damage at which the maximum damage bonus is added, E is the maximum damage bonus, and F is a conversion value. For depth 28 and 5* weapons, it looks like C = 81 (or 82), D = 125, E = 95, F = 95 gives good results.

The damage values used came from the Venom Veiler, Callahan, Polaris, Acheron, Barbarous Thorn Blade, Divine Avenger, Fang of Vog, Fearless Rigadoon, and Hail Driver. Every other 5* weapon either did not have a complete set of damage listings available at depth 28 or duplicated the numbers provided by one of those weapons (for example, Gran Faust duplicated the damage numbers of the Divine Avenger). The damage values for all available attacks were used (all swing damages for swords, charged attacks, expanded+non-expanded Polaris shots, etc.) but not for status damage. For split normal/special weapons, it was assumed that half the base damage was normal, and the normal and special damages were calculated independently of each other. That means that for a split normal/special weapon the adjusted damage is assumed to be calculated as follows:
Adjusted damage = base / 2 + f(base / 2)
where f() is the appropriate function from above.

Edit - I took these formulas and used them to graph the different combinations of damage types, to get an idea of how they compared. The general results: at very low base damages, split normal/special weapons are terrible, doing far less damage to vulnerable enemies, and split special/special weapons are worse, also doing less damage to neutral enemies. As base damage increases, this situation improves. You reach a threshold where split special/special weapons do better than base damage to neutral enemies (this threshold looks to be a little above the base damage of AP or Sentenza).

Split damage weapons also start to catch up to full-special weapons in damage to vulnerable enemies, and after a while you reach another threshold (somewhere around twice the base damage of the first one) where split damage weapons surpass full-special weapons in damage to vulnerable enemies. At this point, split special/special damage would be king, doing more damage to both vulnerable and resistant enemies than full-special weapons, and also doing more damage to neutral enemies than either full-special or normal/special weapons. There aren't any split special/special weapons in this base damage range, though. A lot of 5* sword attacks do fall into this range, however; the Divine Avenger's first swing, for example, would do slightly less damage to constructs/undead (at least at depth 28) if the DA became pure elemental.

As base damage increases further, these advantages fade. Next you reach a point where split special/special's damage to neutral enemies drops back below the base damage, leaving split normal/special at the top of the heap for damage output. Then comes the point where the damage vs. vulnerable enemies equalizes. After this, the only difference between full-special and normal/special weapons is that normal/special weapons do better damage to resistant enemies, and the only difference between normal/special and special/special is that special/special weapons do less damage to neutral enemies.

Thu, 01/05/2012 - 09:20
#1
Bopp's picture
Bopp
interesting

Chronovore, this is interesting and valuable; thank you.

In the part about vulnerable enemies, you are using a somewhat complicated function (piecewise linear/quadratic) to fit the data. When figuring out weapons for myself and advising others, I will have difficulty conveying such a function in a few pithy words. So I'm wondering whether such a complicated function is required. How closely do the data fall on this parabola? Is there a piecewise linear function that works almost as well?

Thu, 01/05/2012 - 10:23
#2
Fradow's picture
Fradow
Not sure if it can be applied

Not sure if it can be applied to optimize some things, but definitely interesting.

I catched a problem with your method : one of your assumptions is probably (not sure if it has been proven) false :
"For split normal/special weapons, it was assumed that half the base damage was normal"

While it's accepted that the first hit (and 2nd on 3 swing swords) is actually equally split, it's almost certain that the last hit is not split half/half, but rather have a higher special damage base. If i remember well (not sure at all here), the normal damage stay the same but the special damage increase.

For charges, I also think it's not half/half. Not really sure on that. If I'm not mistaken, it was Culture who did some research on this a while ago.

That may change significantly your result. I suggest you run again your tests using only data on first hit, to see if the result change.

Thu, 01/05/2012 - 13:54
#3
Chronovore's picture
Chronovore
Concerns

@Bopp
Here's the raw difference in observed damage vs. projected damage for the vulnerable enemies, judge the fit for yourself:
Venom Veiler Blast: -0.19
Callahan Shot: -0.62
Callahan Charge: 0
Polaris Shot: 0.41
Polaris Expanded: -0.22
Polaris Charge: 0
Acheron Hit 1/2: -0.19
Acheron Hit 3: -0.36
Acheron Charge: 0
Acheron Explosion: 0.55
Barbarous Thorn Blade Hit 1/2: 0
Barbarous Thorn Blade Hit 3: 0
Barbarous Thorn Blade Hit Charge: 0
Barbarous Thorn Blade Hit Projectile: -0.75
Divine Avenger Hit 1: -0.33
Divine Avenger Hit 2: 0
Divine Avenger Charge: 0
Divine Avenger Projectile: 1.27
Fang of Vog Hit 1/2: -0.19
Fang of Vog Hit 3: -1
Fang of Vog Charge: 0
Fearless Rigadoon Hit 1/2: 0.25
Fearless Rigadoon Hit 3: 0
Fearless Rigadoon Charge: -1
Fearless Rigadoon Final: 0
Hail Driver Shot: 0.41
Hail Driver Charge: -1
Note that in general, if the difference is exactly 0 or -1, then it's from the linear piece of the function. The fractional differences are from the quadratic piece of the function. I suspect the discrepancy in the linear part of the function comes from 5* weapons not reaching their full damage potential by depth 28, and when the damage values are adjusted downwards for depth, the base damage and increased damage sometimes round in the same direction, sometimes not.

@Fradow
I didn't start out by assuming anything of the sort. I started out looking at only pure-special weapons, then when I looked at split damage weapons to see how they came out differently, assuming half the base damage was normal made the damage bonuses for the other half fit neatly into the curve I got from the pure-special weapons (for all attacks, charge or normal). Therefore I concluded that the 50/50 damage split was correct, and the existing conclusion was based on the faulty assumption that the special damage bonuses are a simple percentage increase.

Thu, 01/05/2012 - 14:13
#4
Fradow's picture
Fradow
Hum, I am not sure I

Hum, I am not sure I understand what you said, but from what I understand, you took that into account, so I guess it's all good :)

This level of maths is a little too hard for me, I have to admit.

Thu, 01/05/2012 - 19:53
#5
Antistone's picture
Antistone
Wait

Wait, wait, wait. Not just the percentage of bonus damage, but the absolute flat quantity of bonus damage, begins to go DOWN after a certain point? So that the marginal utility of "good" damage actually becomes WORSE than the marginal utility of "neutral" damage?

Cripes. That's just bizarre.

I'd like to request that you upload those graphs (perhaps to the wiki) so we can all take a look at them. It would also be helpful to have a list showing the exact numeric values of all those thresholds you mention, and which range each of the weapons falls into, so you can see at a glance which qualitative behavior each should exhibit.

If there are outliers, maybe we should re-test them. It's always possible that some of this behavior has changed...maybe even while data was still being collected. (In particular, we might want to re-test the weapons at the top of the parabola.)

Also...I'm thinking it would be useful to know how this interacts with increased damage abilities (like the weapon trinkets, or skolver armor). If they affect base damage, they could potentially slide some weapons into a different category (which also might explain some of the noise people got when trying to figure out how much they added).

@Fradow: I believe the theories that charge attacks were not split 50/50 were based on the fact that they had a different damage ratio between neutral enemies and weak enemies--that is, the normal attack dealt X% more when the enemy was weak, while the charge attack dealt Y% more, so the ratio of types was assumed to be different. But if the bonus given to special damage isn't constant, then that reasoning no longer applies! So unless there's some other evidence that the ratio of damage types changes, I think it's covered.

Thu, 01/05/2012 - 21:09
#6
Chronovore's picture
Chronovore
Graphs!

Ok, you can see them (the ones I thought were interesting or useful; I made others that didn't turn out as well while I was trying to figure out how the bonus damage might be calculated) at http://wiki.spiralknights.com/User:Exasperation
Giving the exact values for those thresholds is tricky for a few reasons. First, since I don't have access to the SK code, I can only make estimates. Some of those estimates seem to have turned out pretty well, but I can't really guarantee their accuracy outside of the ranges we have measured weapon damage for. Second, the location of those thresholds depends on those equation parameters, which in turn depend on the depth of the floor and star rating of the weapon, and there's no guarantee that those are in sync with the damage penalties for weapons based on depth (in fact, we have some compelling evidence that they don't, in the reports of weapons that go from neutral damage to good damage against the same type of enemy as you go deeper; that implies that as you descended into the clockworks, the weapon went from being on one side of one of those damage thresholds to being on the other side). I have only come up with (estimated!) values for those parameters for 5* weapons at depth 28, and I haven't taken the time to solve those equations for the exact point where the lines meet.

Thu, 01/05/2012 - 22:58
#7
Antistone's picture
Antistone
Lists!

Sure, sure, this is all estimates and conjecture. That doesn't mean we shouldn't display those estimates and conjectures in a format that's easy to understand. Here:

(At low damage, pure effective > half-effective, pure neutral > mixed split)

62 - Blitz Needle (per bullet)
62 - Dread Venom Striker (second part)
(estimated ~73) - Argent Peacemaker / Sentenza

(76 base damage: mixed split becomes better than pure neutral)

88 - Polaris (small)
108 - Dread Venom Striker (first part)
112 - Polaris (large)
117 - 5* Brandish (explosion from charge attack)
121 - Blitz Needle (charge attack, per bullet)
131 - Callahan
153 - Status Bomb (initial explosion)

(180 base damage: half-effective becomes better than pure effective)

183 - Fang of Vog (first two swings)
203 - Nitronome
203 - Dark Briar Barrage
203 - 5* Brandish (first two swings)
203 - Leviathan Blade (first two swings)
203 - BTB = Final Flourish (first two swings)
234 - Divine Avenger (first swing)
251 - Fang of Vog (combo finisher)
258 - Leviathan Blade (combo finisher)
258 - BTB = Final Flourish (combo finisher)
258 - Big Angry Bomb

(262.5 base damage: pure neutral becomes better than mixed split again)

266 - Polaris (charge attack)
292 - 5* Brandish (combo finisher)
296 - Callahan (charge attack)

(320 base damage: pure effective == half-effective from here on out)

363 - Divine Avenger (second swing)
368 - Leviathan Blade (charge attack)
386 - BTB = Final Flourish (charge attack)
483 - Fang of Vog (charge attack)
531 - Acheron (charge swing)
600 - Divine Avenger (charge swing)

In Summary
As a general rule, it looks like guns are in the range where pure special damage is better, while swords and damage-dealing bombs are in the range were split normal damage is better. That would mean that piercing is actually swords' WEAK damage type, not it's strong damage type as I previously assumed. Also, nearly all weapons occupy the damage range where mixed special damage is better than pure neutral damage--except the Antigua line!

Once again, this is all for 5* weapons at depth 28, and assuming that Chronovore's formulas are accurate. Take it all with a sizable grain of salt.

Fri, 01/06/2012 - 19:51
#8
Chronovore's picture
Chronovore
Progress Update

I spent some time working with both the depth 28 data and data for the same weapons at depth 27. This reduced the number of data points available, since some of the weapons didn't have data for depth 27, but it allowed me to refine things quite a bit.

The base damage reduction based on depth appears to work something like the following: there is a value K for each depth. The adjusted base damage B of a weapon at a given depth is calculated from the true base damage T, based on K. My current approximation to the formula for doing this is:
B = T * 0.4 * K + T * 0.6
This adjusted value is then used to find the bonus/penalty for the weapon. My current iteration of these formulas is:
Penalty = (B - 20) * 0.85
* (this is just a restating of the formula I had above, I mention it here because it appears to hold for depth 27 as well, although there are more outliers in the depth 27 data)
Bonus = K * 100 - (MIN(B, 160) - 125)^2 / 100
* (MIN(B, 160) is another way of putting in the linear piece of the earlier version of the function; 160 is the value at which the linear and quadratic parts of the function meet)

K appears to have values of ~0.94 at depth 28 and ~0.9 at depth 27.

So how close is the fit? Well, I chose values of T for each data point so that at depth 28, the calculated value of B at depth 28 was less than 0.5 from the observed damage to neutral enemies (that is, it would round to the observed value). Then, using that same value of T, I calculated the bonuses at depth 27 and 28. At depth 28, the fit is amazing. Out of 17 data points available, 9 had differences between observed and calculated values of less than 0.5, and the largest difference out of the 17 was 1.06.
At depth 27, the fit is not as good, but still pretty good. The discrepancies in observed vs. calculated base values run from -12.96 to 8.24, but there isn't any obvious correlation between the differences and the base values; also, out of the 17 available data points, 6 of the differences were less than 1, while another 5 were between 1 and 2. For the bonuses, the range was from -2.82 to 5.57, with 10 less than 1 and another three between 1 and 2. Interestingly, in one case the calculated base value was 5.28 less than observed, while the calculated bonus was 5.38 more than observed; that means that the observed high damage value was only 0.1 less than the calculated value, even though the base values were 5 apart.

Qualitatively, this doesn't change anything from what Antistone posted above; the damage functions are still the same shape, the estimated base damage thresholds are still around the same values, and the stuff in the summary still holds just as well.

Mon, 02/13/2012 - 17:13
#9
Arquebus's picture
Arquebus
Some new info

I realize this thread is now old, but I've been gathering damage data for guns for the wiki, and I think you may find these numbers for AP and Sentenza very interesting...

https://docs.google.com/spreadsheet/ccc?key=0AjvYpSStSGjLdFZYd0F3dEZzbXk...

I double-checked any numbers that seemed odd, and made sure everything was correct... the damage for the charge does in fact go DOWN in the first stratum!

Mon, 02/13/2012 - 20:54
#10
Chronovore's picture
Chronovore
That is interesting. I

That is interesting.
I noticed something similar while getting rid of the 200 cr fee to get to tier 2, where against beasts/gremlins the damage for the Blackhawk's basic attacks would dip down on floor 2 and go back up on floor 3.

Tue, 02/14/2012 - 16:06
#11
Otaia's picture
Otaia
Damage with all weapons dips

Damage with all weapons dips at Depth 2 IIRC. There's probably some kind of damage adjustment parabola with a minimum between 1 and 3.

Powered by Drupal, an open source content management system