Warlock Stat Preference: Simulation vs Empirical Data

[mtp1032 0]

How do the results produced by the simulator match with results from empirical testing? 

 

I'm not questioning the value of simulation (I used to write C++ event-based simulations as a professional S/W engineer and I truly appreciate their value), but I'm trying to better understand how to get the best performance out of my characters. For example, when I conduct empirical testing using my 98 affliction lock I get the following results (% Increase Total Damage over the control):

 

Versatility: 6.11%

Crit: 5.79%

Mastery: 3.73%

Multi: 3.64%

Haste: 1.88%

 

Since these data run counter to the advice in the Classes section of the web site, I'm wondering if I completely understand how these stats work.

 

For example, with +75 Versatility (Skulker Chowder), the total damage with the buff was 619374 over the mean control value of 581553 (= 6.11%). By contrast, the mastery buff from Fat Sleeper Cakes yielded only 3.73% improvement over the control. Similarly, Haste was the least beneficial. Since the advice from the sim folks suggest Mastery and Haste are the most beneficial, the liklihood that I'm missing something (e.g., contribution from armor at level 100) may be pretty high.

 

FYI, the control samples (total damage via recount) were run with no armor and no buffs casting at a level 85 dummy. The values of the independent variables were generated using WoD food buffs (+75 of the relevant stat), no armor, casting at the same dummy.

 

Each result was generated as the average of 5 separate casts of a single Soul burn followed by a Soul Swap (i.e., so as to apply UA, Agony, and Corruption simultaneously). Each cast was allowed to decay completely.

 

The standard deviation was less than 1% (i.e. 0.95%) for all experiments.

 

Thanks,

Edited December 29, 2014 by mtp1032

[Zagam 1,982]

Buffs mean a lot.  There's no point simulating DPS without full buffs since that's how they are needed to be evaluated for practicality.

[mtp1032 0]

Buffs mean a lot.  There's no point simulating DPS without full buffs since that's how they are needed to be evaluated for practicality.

Do not take this reply as a disagreement, I am just trying to better understand what's going on.

 

First, I'm not asking about simulating DPS. I'm really asking about the assumptions necessary to conduct empirical tests of the effect of the various stats on DPS.

 

So, here's where I'm not understanding what's going on: it seems to me that for your statement to be true (and I do not dispute it), the value the mastery buff, for example, must change in the presence of other buffs. Put another way, perhaps you're arguing that simplistic experiments such as mine cannot be valid since buffs interact in ways not [readily] testable.

 

Is this correct? 

 

Thanks for the quick reply,

[Virzaol 0]

That's exactly what he's saying, yes. Stat values fluctuate with even the slightest change to one. Increasing your base spell power will make mastery more powerful, because it adds a % damage modifier. Increasing haste will decrease its overall value as you approach the GCD cap.

If you're not trying to understand the actual damage output of those spells in the context of a buffed raid character against a debuffed boss, you're not getting the whole story. That's like asking how fast a car can go in a straight line without ever putting it on a road with turns, other cars, etc.

Edited December 29, 2014 by Virzaol

[mtp1032 0]

Virzaol,

 

OK, that makes perfect sense. Do you (or someone) know of additional information documenting the damage multipliers and how they are applied that I could read? For example, where to the designers of the simulator get these values?

 

Thanks, again

Edited December 29, 2014 by mtp1032

[Karigg 0]

 

 

Each result was generated as the average of 5 separate casts of a single Soul burn followed by a Soul Swap (i.e., so as to apply UA, Agony, and Corruption simultaneously). Each cast was allowed to decay completely.

 

 

Using that method I would have expected all the dps stats to be roughly equal with perhaps mastery slightly ahead, so it's surprising. Though, with a small sample size, the results could be a fair way off. Simulations generally go through 10,000 repetitions of a 6-9 minute fight. 

 

One big thing I would consider is the time taken to cast the spells, and shard generation which are both effected by haste. 

 

Drain soul on top of the dot damage would act similar to a flat damage buff, as well as Agony staying at 10 stacks rather than building up to 10 after each cast. 

 

 

It used to be pretty easy to look at your damage breakdown and just scale parts of it as if you had added 100 rating of a particular stat. But now with Multistrike, and the way haste scales here, it's not so easy. Might be fun to have a go though!