View Single Post

MGNMTTRN's Avatar


MGNMTTRN
05.09.2013 , 12:44 PM | #18
So I attempted to tackle crit vs power. It turned out to be hard.

I assumed that for the ith attack in your rotation, the attack's average non-crit damage is well-modeled as bi + ci*bonus damage, where bonus damage is ranged/melee/tech/force damage depending upon what the attack type of the ith attack is. Since our original optimization equation was E(dam) = (1-acc)*dam + dam*acc*(1-crit + crit*surge) we can now replace these dam values by b+c*bonus dam. This makes E(dam) a function of both bonus damage and acc/surge. Ranged/melee Bonus damage is a composite of main stat and power, while tech/force bonus damage is a composite of mainstat, offstat (cunning for troopers, willpower for warriors, etc), tech/force power from mainhand and offhand, and power. Of these, I assumed we only really wanted to vary power.

For example, here is the data that led me to decide that incendiary round's damage is well-modeled as 662 + 2.44*tech bonus damage:
Code:
incendiary dam	tech bonus dam				
3716.50			1250.90				
3597.00			1201.70				
3471.00			1150.40				
3378.00			1112.30				
3260.00			1063.80				
					
presumed base					
662.00					
original dam	Dam  base	c	average C		
3716.50		3054.50	2.44	2.44		
3597.00		2935.00	2.44	2.44		
3471.00		2809.00	2.44	2.44		
3378.00		2716.00	2.44	2.44		
3260.00		2598.00	2.44	2.44		
					
projected base	tech bonus dam		reconstituted	actual	diff
662.00			1250.90				3716.69	3716.50		0.03
662.00			1201.70				3596.54	3597.00		0.21
662.00			1150.40				3471.27	3471.00		0.07
662.00			1112.30				3378.23	3378.00		0.05
662.00			1063.80				3259.79	3260.00		0.04
					
														total diff
														0.41
So for bonus damage values between 1250 and 1070, this was a reasonable model. The sum of squared errors (predicted - observed damage) was only 0.41, so I felt comfortable using this model in my simulations of SWTOR damage. I repeated this exercise to get b and c values for ion pulse, plasma cell, incendiary round, assault plastique's kinetic damage, assault plastique's dot damage, and high-impact bolt. The incendiary, pcell, and ion pulse sum of squared errors were really low (below 0.5), but the other values had high errors (e.g. 150 error for ap, 5500 for HIB). This b + c*bonus damage calculation is absolutely not how the game actually calculates the potential damages of your attacks, but in my opinion it is sufficient to do some estimations for values near 1250 and 1070 bonus damage.

My simulations assumed stat pools of 1239 power/crit and 720 acc/surge, and assumed that power and crit would trade off and surge and acc would trade off. We just want the highest value in a two-dimensional, 120-entry space.

In these images, the important values are to the lower right of the sheet. Crit value increases toward the right, and surge increases toward the bottom. I opened the cell of whatever value was maximized for that combination of b, c, acc, and crit.

http://i.imgur.com/w4wYnAU.png Assault plastique kinetic damage
http://i.imgur.com/nWJg2lg.png Incendiary total damage
http://i.imgur.com/5Hh6l3r.png Ion pulse total damage
http://i.imgur.com/SV5tyE6.png one tick of plasma cell

For all of these attacks except Ion Pulse, the maximum damage was gotten by getting accuracy to 100% (which we knew from the original post) and getting about 52 points in crit. Ion Pulse was the only attack which I bothered to model which needed a whole 104 points in crit to maximize its damage.

Notes:
  • I hoped to take a weighted sum of all my attacks, to simulate a rotation. If I migrated this modelling to MATLAB, it would become extremely easy to optimize the rotation instead of maximizing for a single attack at a time. But I'm not sure doing additional work on it is necessary, since the results all corroborate with each other and with "tests" conducted on the PTS; keep your crit low. Also, do note that the results indicate that a small amount of crit is better than 0 crit.
  • Some general advice: I've seen people passing on gear with 173 mainstat, 84 crit, and 73 acc because they have a gear piece with 123 mainstat and 83 power. They said they were doing a "zero crit" build. But here's the problem: power is just slightly better than crit. At 52 crit and 100% acc our maximum incendiary round damage was 4952, and with 416 points in crit it dropped to 4906 damage. This is a change of less than 1%. It's probably in your interest to wear gear with crit in it if it provides large upgrades to your other stats.
  • 52 crit points is not the magical number at which everyone wants to start dumping power. It may be that it's actually 60 or 70 or maybe even 100 points that optimizes the distribution for your class and spec. However, my Vanguard does get +30% surge on most everything, which would have encouraged a higher crit value.