Introduction

Juggernaut

Powertech

Assasin

Notes

Credits

Updates

Stats tested as full 198 Juggernaut Tank w/o MH and found to be valid.

What does this mean exactly?

#1 Malaphar : M/R+K/E: 52,6%

F/T+K/E: 0%

F/T+I/E: 47.4%

No reason for stacking absorb as there is only M/R+K/E dmg.

Your stats :

KBN 3800 0749/1521/1530

Dipstick 1919/1200/0722

What does this mean exactly?

http://parsely.io/parser/view/20392

Temple Sword Squad + some Revan

Squad wasn''t perfect as i got hit by rapid fire which made 115 dtps, so in case you play perfect you will be at 2600dtps.

Revan was close to worst tries so they are actually not counting.

Dtps seemed good + Healers found it easy to heal. Further testing it tommorrow in Ravagers.

#1 Malaphar : M/R+K/E: 52,6%

F/T+K/E: 0%

F/T+I/E: 47.4%

 

No reason for stacking absorb as there is only M/R+K/E dmg.

 

Your stats :

 

KBN 3800 0749/1521/1530

Dipstick 1919/1200/0722

This is just a function of the fact that Malaphar is lumped in with other low-M/R bosses, most of which have lower I/E and higher F/T+K/E. Honestly, I think I really should have done the groupings by M/R vs F/T+K/E ratio, rather than just blindly looking at unadjusted M/R+K/E.

Or you could've treated Malaphar as an unique boss?

So basically Malaphar is also a high M/R+K/E boss for you?

B-Mods Stat Budget = 3841

Your Averages:

Juggernaut 1098/1505/1219 = 3822

Powertech 800/1747/1250 = 3797

Assassin 649/1458/1690 = 3797

Closest Actual Attainable Stats (Matched Defense, then Shield, and rest in Absorb)

Juggernaut

Def - 1094 = 25.08%

Shield - 1512 = 47.97%

Abs - 1235 = 46.63%

Pwertech

Def - 807 = 18.48%

Shield - 1720 = 51.20%

Abs - 1314 = 51.69%

Assassin

Def - 653 = 23.94%

Shield - 1460 = 58.38%

Abs - 1728 = 53.57%

Happy Tanking!

Or you could've treated Malaphar as an unique boss?

 

So basically Malaphar is also a high M/R+K/E boss for you?

No, Malaphar is counted as a low MR/KE boss exclusively, which is a problem. My criterion for separating low, mid and high-MR is simply to look at the absolute MR level, rather than normalizing to just MR+KE vs FT+KE, which would shift Malaphar into high-MR (since the normalized ratio is 100%).

This was never a problem in previous tiers since the I/E levels were relatively low on basically every boss (and also fairly constant), but the error increases with the percentage of damage that is I/E. Malaphar is of course the poster-child for that kind of boss, so it is the one most noticeably mislabeled.

nice to see so many people participating in tank theorycrafting. i just wanted to point out that my numbers should minimize the dtps, and i do not attempt to minimize spikeiness at all (i would be open to a discussion of how to do so im my thread). i have yet to dig into kbn's mathematica script to find all the deviations of our assumptions from one another. the biggest is the process by which numbers are derived. kbn uses mathematica to generate his numbers, and i believe he has a method of testing the neighborhoods of the points suggested by mathematica (or some algorithm to avoid local minima). I use excel solver and if i see the numbers lobesided, it is typically due to a bad seed (initial guess) that pushes the numbers into a particular local minima of dtps. i will then change the seed to something like 1300 of each stat (if you look at my sheets you will see cells next to defense, shield and absorb that are there just to copy and paste to set the seed). i used to use a maple (another math program, for symbolic stufts mostly) method of lagrange multipliers to find optima, but, like kbn's process, it would take around 5 minutes to get d,s,a ratings for a single set of lagrange constants, for a single type of tank (and that was before kinetic bulwark added around 100 operations per damage/stat setup). even when we were both using numerical analytical methods, we still differed in that he used ratios of stats (like x+y+z=1) where as i set up lagrange constants like defense=stat_pool-shield-absorb. i think kbn still uses these ratios in his script.

the more meaningful difference between kbn and i (outside the realm of numerical analysis considerations) would be the assumptions behind things like heat blast/screen and kinetic bulwark (kbn does a much better job of kinetic bulwark calc since he doesnt rely on a regression like i do, but im not sure if he has a modified script for the new set bonus). i am also still not sure how the game is handeling the math after 3.0, in terms of which abilities get a multiplicitive damage reduction, and where they pop up for things that are very specific (3.0 now has damage reduction talents for every type of damage type and attack type).

anyway, happy to see more people looking into this stuff.

Stats will change as soon as i update Number from KBN to be 3841 insted of 3800 which ofc had an influence of the total stat budget as Jethsidi wrote. In addition i have to update Malaphar numbers, so also the average for each tankclass will change as soon as KBN will update his.

To dipstick :

Your term squishiness in your calculation does refer to spikyness, so i thought you took this also into consideration and that was the reason for your numbers to differ from KBN's - seems i was wrong. Will change my introduction soon with your stated differences between you and KBN.

anyway, happy to see more people looking into this stuff.

Actually i am not looking into this stuff, but looking into your ideal tankstats and those of KBN and Jethsidi So basically it is

(Dipstick <def> + KBN <def> + Jethsidi <def>) / 3

i have yet to dig into kbn's mathematica script to find all the deviations of our assumptions from one another. the biggest is the process by which numbers are derived. kbn uses mathematica to generate his numbers, and i believe he has a method of testing the neighborhoods of the points suggested by mathematica (or some algorithm to avoid local minima). I use excel solver and if i see the numbers lobesided, it is typically due to a bad seed (initial guess) that pushes the numbers into a particular local minima of dtps. i will then change the seed to something like 1300 of each stat (if you look at my sheets you will see cells next to defense, shield and absorb that are there just to copy and paste to set the seed). i used to use a maple (another math program, for symbolic stufts mostly) method of lagrange multipliers to find optima, but, like kbn's process, it would take around 5 minutes to get d,s,a ratings for a single set of lagrange constants, for a single type of tank (and that was before kinetic bulwark added around 100 operations per damage/stat setup). even when we were both using numerical analytical methods, we still differed in that he used ratios of stats (like x+y+z=1) where as i set up lagrange constants like defense=stat_pool-shield-absorb. i think kbn still uses these ratios in his script.

I calculated a new set of ratios based on 3.0 gear (namely the sturdiness/immunity enhancements and new relics). These ratios are a first-order approximation in the purest sense of the phrase, so there's always room for improvement. Should be pretty accurate for full Revanite gear though.

Mathematica actually works its way out of local optima on its own, which is kind of fun (and also part of why it takes such an absurd amount of time). I used to have to give it an initial seed set and prune off the computation manually, but by doing better on the min/max values of shield given a certain stat pool, I can give Mathematica enough information to bound its differential solution space. It does have certain points where it numerically prunes off the approximation, and this can result in "jumps" in the continuous graphs (you can see this in my post). In general though, I'm just climbing the hypersurface. The most complicated element of the optimization is, as you referenced, Kinetic Bulwark, simply due to the fact that defense levels affect frequency of shielding which affects KB value which affects absorb (and shield) value which affects defense and shield value which loops back to the beginning, so that optimization is quite complex. I have updated for the new set bonus, as well as the new swing timers (new content has a MUCH higher swing timer than old content).

A lot of the really ugly bits of my worksheet are devoted to memoization (without which even computing a single data point could take weeks of CPU time) and parallelization. The actual math bits are quite straightforward.

many thanks

spike is a function of squish... i wonder if i can solve for squish as a function of spike and use that for something...

i still think using boss abilities and frequencies to generate death probabilities as a function of fight length, and setting up a model where endurance can trade with mitigation to find a balance between required heals and ... required healing actions per minute? last i checked kbn wanted to do a signal process type thing. probability of a boss ability, probability of being at a certain amount of health should give probability of dying to a certain hit. we want to decrease the probability of dying.

there might be a quick way of using spike as an exponent and using time to kill as a type of carrying capacity or something... logistic model?

I calculated a new set of ratios based on 3.0 gear (namely the sturdiness/immunity enhancements and new relics). These ratios are a first-order approximation in the purest sense of the phrase, so there's always room for improvement. Should be pretty accurate for full Revanite gear though.

 

Mathematica actually works its way out of local optima on its own, which is kind of fun (and also part of why it takes such an absurd amount of time). I used to have to give it an initial seed set and prune off the computation manually, but by doing better on the min/max values of shield given a certain stat pool, I can give Mathematica enough information to bound its differential solution space. It does have certain points where it numerically prunes off the approximation, and this can result in "jumps" in the continuous graphs (you can see this in my post). In general though, I'm just climbing the hypersurface. The most complicated element of the optimization is, as you referenced, Kinetic Bulwark, simply due to the fact that defense levels affect frequency of shielding which affects KB value which affects absorb (and shield) value which affects defense and shield value which loops back to the beginning, so that optimization is quite complex. I have updated for the new set bonus, as well as the new swing timers (new content has a MUCH higher swing timer than old content).

 

A lot of the really ugly bits of my worksheet are devoted to memoization (without which even computing a single data point could take weeks of CPU time) and parallelization. The actual math bits are quite straightforward.

pray tell, what is the new swing timer? since im doing all per boss i guess i could use per boss swing timer info... if anyone has those...

B-Mods Stat Budget = 3841

Your Averages:

Juggernaut 1098/1505/1219 = 3822

Powertech 800/1747/1250 = 3797

Assassin 649/1458/1690 = 3797

 

Closest Actual Attainable Stats (Matched Defense, then Shield, and rest in Absorb)

Juggernaut

Def - 1094 = 25.08%

Shield - 1512 = 47.97%

Abs - 1235 = 46.63%

 

Pwertech

Def - 807 = 18.48%

Shield - 1720 = 51.20%

Abs - 1314 = 51.69%

 

Assassin

Def - 653 = 23.94%

Shield - 1460 = 58.38%

Abs - 1728 = 53.57%

 

Happy Tanking!

I've added these in as Methoxa 3841 Avg. I'll wait for KBN to check his boss categorisations and Methoxa's recalculations before I add in the per boss ones.

Stats have been updated with KBN's 3841 numbers.

Malaphar stats updted with KBN's new numbers

 

Dtps seemed good + Healers found it easy to heal.

Not to be difficult, but I'm pretty sure with the same quality of execution your healers wouldn't be able to tell the difference between any of these 4 gear stat distributions (the three you're averaging and your average). The variation of incoming damage is far more dependent on executing mechanics and RNG than these slight differences in stat weights. Frankly, you could probably swap 500 points around in your revan gearing distribution and your total dtps difference would be totally wiped out by timing saber reflect a tiny bit better (or the attack timings being friendlier) and catching a single extra cleave with it.

I certainly see the value in academic comparisons of incoming damage based on gear, but healer feedback/logged dtps is a really unreliable way to judge the relative merit of stat distributions.

For perspective, if you see a 1% change in damage taken on a fight where you're taking 1500 dtps (which as far as I can tell is a larger change than the difference in any of these stat weights), it's a 15dtps difference. That's 9000 damage total in a 10 minute fight (like revan or coratanni).

Not to be difficult, but I'm pretty sure with the same quality of execution your healers wouldn't be able to tell the difference between any of these 4 gear stat distributions (the three you're averaging and your average)

No one would, thats why we use parsers but still ask for feedback from our healers because parsers don't say much about spikyness which is highest at sword squadrons. Thats why i run the tests. Furthermore my dtps are in fact a little lower than with KBN stats however, that might be coincidence as you would need to run 10000 tests to proof something like that. It however tells me that my numbers are not totally bad.

KBN 3841 : 0752/1541/1548

Methoxa 3841 : 734/1707/1400

The difference is the shield and absorb. While trying to minimize dtps you will want to try to keep absorb about the same than your shieldrating. My numbers however prefer a higher shield. Shield means less spikes. While my numbers might increase dtps they will reduce the amount of spikes. KBN's number however might not decrease the frequency of spikes, but the damage of those spikes. I prefer receiving less spikes, because as a juggernaut i have other possibilities of llower the dmg of a spike.

But as i said before, it might also be coincidence that my numbers let my healers feel that i was less spiky. Of course they might be wrong, maybe some other healer would think that i was far more spiky. But for me it is not possible to calculate spikiness and inplement it into a tanking formula. Thats why i trust the 3 people that might be able to do that one day. KBN for his dtps minimizing formulas, as well as dipstick and Jethsidi for his insight about DR mitgation. Putting all 3 together might be closer to perfection your equipment than just sticking with one of them.

And as you said, perfect execution is more important than tankstats, however even if you do a boss for the thousands time something might go wrong, or someone might not play perfect and thats where stats come into play.

you could probably swap 500 points around in your revan gearing distribution and your total dtps difference would be totally wiped out by timing saber reflect a tiny bit better

Well we are still learning this fight so excuse the non perfect execution. However in the few tries we had i observated that Revan sometimes delays his Overcharged Dual saber attack after a heave so it wont fit into the Saber reflect window because it is delayed by nearly 6 seconds.

I very well understood what you wanted to say, and i thought i answered to it with this sentence :

And as you said, perfect execution is more important than tankstats

If you had 2 tanklogs, both perfectly executed but one of the tanks without offhand you can tell the difference. The difference between a tank without min-max gear and a tank with min-max gear might be slighter, but is this really what you essentially want to say? Min - maxing is not worth it? I thought Zorz was min-maxing kings.

Sometimes even 5 dtps less can make the difference. Thats why i try both perfect exection and perfect equipment.

Of course a perfect execution is more important but some raids <mine recently and in the past> lack in consistancy regarding the players, who then have to learn the fight and can not play perfectly,yet.

I very well understood what you wanted to say, and i thought i answered to it with this sentence :

 

 

 

If you had 2 tanklogs, both perfectly executed but one of the tanks without offhand you can tell the difference. The difference between a tank without min-max gear and a tank with min-max gear might be slighter, but is this really what you essentially want to say? Min - maxing is not worth it? I thought Zorz was min-maxing kings.

 

Sometimes even 5 dtps less can make the difference. Thats why i try both perfect exection and perfect equipment.

I'm saying that you can't min-max by looking at logs because you'll never have perfect execution. That's why I said

"I certainly see the value in academic comparisons of incoming damage based on gear, but healer feedback/logged dtps is a really unreliable way to judge the relative merit of stat distributions." The 10dtps difference is going to be totally lost in fight-to-fight changes.

Mine combine the assumptions of all three, that what i think is the best for me. And i wanted to give other people the oppurtinity of trying for themselves. If math was the only way Jethsidi's stats wouldn't be viable because he doesn't offer the math KBN does nor does any formula proof that KBN's stats of minimizing dtps is the right way. Which lead me to the conclusion that the middle-way would be near the truth.

By viable i meant not that my stats are the best but that the outcome is nearly as good as with the other three. Just wanted to see if there was a bossfight where i would take insanely high dmg with my stats, which would proof them false if execution wasn't the problem. That did not occur, thats why i find these stats to be viable.

Jethsidi B-mod tankstats : 1094/1720/1027 Juggernaut

Go with these stats to underlurker and you'll see dtps increase even more than which could be explained with variance than with

KBN 3841 stats : 1618/1142/1081 Juggernaut

The difference will be 100-150 dtps. So Jethsidi proven wrong? No, because his tankstats do not only strife for dtps minimization but take DR into account as well as spikyness. Spikyness is a term which has not been mathematically calculated or somehow been implemented into Tankstats before. Mathematically KBN is apparently right. However i've never seen all his formulas for calculating all the things nor do i have the time and knowledge to validate those. I guess that most ppl here can't do that. So in other words, KBN could still be wrong when one of his thousand formulas is not correct.

There is no math that would implement all factors in tanking and give out the best tankstats. Not even KBN's.

Btw. the best tankstats for me would be not tankstats that would minimize my dtps but would make my dmg income controlable and static so that healers would never need to react but simply know ahead of time when they have to heal me. So basically they know Methoxa needs exactly 20k heal at 0:32 infight, 0:45 infight an so on. Meaning no variance in dtps profile. One factor that influences that is spikyness. But how do you mathematically prove certain tankstats to be better against spikyness than other stats? I can't find a formula and because of my lack in mathskills i have no option but use parsers and feedback of others to determine my spikeness.

Thanks for your Feedback though Smugglin'

wanted to pop in and say i do show the equation i use for spike:

spike= sum(probability[ij]*(squish[ij] - squish)^2)

where i runs over kw_named, kw_bland, kf and ef (i thought there was ew damage, but i guess not)

and j runs over defended, resisted, shielded and full hit. (see my thread for variables)

basically, it is the weighted variances of all the types of hits you can take, comparing the squish of each hit type to the overall squish.

minimizing spike typically leads to low defense values, as you can imagine. here are some results for corttani, for a shadow, minimizing X*squish+Y*spike, where X+Y=1. if X=1 and Y=0, then you are only optimizing for squish and ignoring spike. likewise, when y=1 and x=0 you are gearing completely for spike and ignoring squish. Each tank will have a different opinion on how to weight these, so having access to my spreadsheet and being able to use excel solver would allow them to gear for whatever kind of squish-spike weighting they want.

m/r k/e 0.704

m/r i/e 0

f/t k/e 0.181

f/t i/e 0.115

X=0

d 238

s 1928

a 1837

base 0.2330

D 0.2277

A 0.2291

S 0.2291

spike 0.02155

X=.1

d 437

s 1928

a 1638

base 0.2323

D 0.2273

A 0.2280

S 0.2283

spike 0.02158

X=.2

d 568

s 1838

a 1596

base 0.2317

D 0.2270

A 0.2274

S 0.2277

spike 0.02170 0.8

X=.3

d 704

s 1712

a 1587

base 0.2312

D 0.2267

A 0.2270

S 0.2272

spike 0.02186

X=.4

d 785

s 1639

a 1580

base 0.2310

D 0.2266

A 0.2269

S 0.2270

spike 0.02197

X=.5

d 838

s 1591

a 1574

base 0.2309

D 0.2266

A 0.2269

S 0.2269

spike 0.02204

X=.6

d 877

s 1556

a 1570

base 0.2308

D 0.2266

A 0.2269

S 0.2268

spike 0.02210

X=.7

d 906

s 1530

a 1566

base 0.2308

D 0.2266

A 0.2269

S 0.2268

spike 0.02214

X=.8

d 929

s 1510

a 1564

base 0.2308

D 0.2267

A 0.2269

S 0.2268

spike 0.02217

X=.9

d 948

s 1494

a 1561

base 0.2308

D 0.2267

A 0.2269

S 0.2268

spike 0.02220

X=1

d 963

s 1481

a 1559

base 0.2308

D 0.2267

A 0.2269

S 0.2268

spike 0.02222 0

I would be interested in how much does lowering shieldating has an effect on spikyness and how much does lowering absorb has an effect on spikyness.

In your formulas you lower both except for the first and second formula. Here you are lowering absorb only. My guess is that lowering shield increases spikyness more than lowering absorb.

Another question does your spikyness only represent the frequency of spiky hits or does it also represent the strength of a spiky hit?

If first is true then its only natural that shield has more impact on spikyness than absorb.

If second is true then we might have overvalued absorb too much and its effect is nearly neglectable when it come to avoiding spikes. As an dtps minimizing value it sure still has a value.

It would be interesting to see how much decrease in spikyness 1point of shieldrating had and how much decrease in spikyness 1 point of absorb rating had, where the DR becomes too strong so that at a certain shieldrating absorb will decrease spikyness more than shieldrating and so on.