How does Mr. Robot get his default Stat Weights?

[Revulva]

Stat weights can be confusing and misleading. Technically, static stat weights are only valid for one particular set of gear and stats. But, Mr. Robot has developed a method that we are going to call dynamic stat weights. Mr. Robot starts with base stat weights independent of any cap or diminishing returns, then he applies those mechanics to the "base" weights to get his "dynamic" stat weights. It is possible to use these dynamic stat weights to obtain optimal or near-optimal sets of gear over a broad range of stats. "Near-optimal" in Mr. Robot's opinion is within 1% DPS of the actual "optimal" set. (Optimal sets are up for debate, since there is no 100% accurate simulator for any class in existence, but, some are very good and we have to use the tools available to evaluate.)

Mr. Robot's site is not a simulator. It is also not a spreadsheet or program that tries to use formulas to predict the DPS of a certain spec. Mr. Robot uses his dynamic stat weight method to compare and optimize gear, but he's smart about it. He knows how to take into account soft and hard caps for stats. He knows that dodge and parry have diminishing returns. He knows... a lot. Remember that Mr. Robot's gear sets are optimal according to the stat weights in use. This is why his team has searched all over the web for information that will lead him to the best default stat weights possible. Feedback is welcome.

When using a simulator, such as SimC, or a "formulator" such as Shadowcraft, Mr. Robot tests multiple sets of gear to arrive at the stat weights he settles on for the defaults. A common mistake you will find on other theorycraft sites is that they will find the set of gear which "sims" out to the highest DPS they can find, and then calculate the stat weights from that set of gear. This is wrong. There are three main questions that are trying to be answered with stat weights:
1.) What set of stat weights should I use to optimize my current gear?
2.) What set of stat weights will tell me which item I should get next?
3.) What set of stat weights will results in the "Best in Slot" set of gear?
The answers to these questions are not necessarily the same. When you use a simulator such as SimC and sim the BiS set of gear for a spec, the stat weights SimC reports tell you the relative value of varying any one rating by 300, with respect to that set of gear. Those stat weights will not necessarily lead you TO that set of gear. In order to solve for question 3, you need to do multiple tests until you find the right set of stat weights. Our testing has shown that, more often than not, the stat weights which answer question 3 will also be very good for question 1 and 2. The reverse is not true. Most other sites are solving for question 2, which is insufficient.

Mr. Robot uses a very simple and effective method - and Mr. Robot is actually a necessary tool to solve this problem efficiently!
1.) Take a guess at a BiS set and sim it.
2.) Take the stat weights calculated by that simulation, and plug them into Mr. Robot to generate a set of gear.
3.) Simulate this new set of gear and note the stat weights.
4.) Repeat steps 2-3 at least one iteration past the set that provides the highest DPS.
Some judgement will be needed - it is sometimes the case that stat weights will swing back and forth across some inflection point, and you will need to try to guess what that point is - it could be a haste soft cap, for example. The goal is to find stat weights that lead us to the BiS set. Once you have done this, you should repeat the process at different gear levels. It is often the case that the same stat weights can be used at multiple or all gear levels, contrary to popular belief.

When we set up SimC, we do 10,000 iterations on a Patchwerk style fight of length 450 sec +/- 20%. We use the default ability rotation unless noted otherwise on one of the spec pages. 10,000 iterations is enough to get stat weights within +/- 0.04. If two weights are calculated so close that they fall within the margin of error, we make a judgement call and either set them equal or let one very slightly edge out the other, based on our knowledge of the spec.

Is this an exact science? No. Will it get you so close to optimal gearing that it is impossible for you to physically verify in-game? Yes. Mr. Robot believes this level of accuracy is more than sufficient for humans.

Keep in mind that you can always use the Stat Weight Editor to change the weights and create your own custom stat weight values!

We are in the middle of an extensive update of our stat weights sources! The stat weight source information will be moved to the class pages which are currently found here: http://www.askmrrobot.com/wow/specs. We will make all sorts of noise and posts about it when they are completed.

[mookei]

I guess what I don't understand is how to make sure using Mr. Robot that we are hit capped and expertise capped.  I am a NElf druid tank and the percentages are the same as anything else.  So how do I optomize the gear listed to make sure that I have 8% hit and 6.5% exp?  When I check the BiS for that class/spec, the hit is only showing like 2%. 

[Devloc]

Hit and Expertise are generally considered threat stats.  Thus far in Cata, threat is a non-issue.  Therefore, hit and expertise are almost always dropped and reforged away for avoidance (and sometimes mastery).

Except, of course, I understand bears have a hard time tanking multiple mobs compared to other tanks.  If this the situation you're in, you might want to adjust the default stat weights to favor hit/exp.  If you're just thinking single target, don't worry about it, it's correct.

[mookei]

Hit I can see as a threat stat, there are no repercussions to being below hit, outside the problem that you're not hitting the boss and thus not generating any threat.  Exp cap I would think is pretty important as if I get parried then the boss gets an extra attack on me and I take more damage.  If the avoidance negates that then I can see your point.  Maybe I'm just stuck in the old school tanking ways, heh.

[Devloc]

Yea, that's old school.  There is no parry haste anymore.  Exp does the same thing Hit does - make you hit the target more.

[mookei]

wow, I had no idea.  Thank you so very much for taking the time to answer my questions and in a timely fashion.  I'm going to give them a try and I'll report back.

Thanks again,

Mook

[Marzim]

From this post you say you get your stat weight from elitist jerks.

But when looking at the stat weight for 2h unholy dk mr robot show this for pre raid:

Attack Power   0.79
MH DPS      3.5
Strength      3.1
Mastery      0.9
Expertise      0.82
Physical Hit      0.82
Spell Hit      0.28
Crit         0.76
Haste         1.01

Where as elitist jerks show this:

[iLvl 359 - Normal Raids]
Stat  2H
Strength            2.96
Hit                    1.08
Haste                    0.86
Mastery            0.82
Expertise            0.72
Crit                    0.66
Attack Power    0.75

Is there a reason why the stats are so different?

If i change the stats to what elitist jerks show I get suggested other gear (eg the wep). I did remove the MH DPS stat (the wep i'm suggested instead does actually have 0.1 more dps) and the spell hit stat.

[yellowfive]

Stat weights posted on elitist jerks typically do not separate physical hit and spell hit.  There are a few exceptions, but in this case they do not.  Mr. Robot requires spell and physical hit to be weighted independently -- sure, you happen to get both from "hit rating" on gear, but they are completely independent stats, i.e. spell hit and physical hit never affect the same ability, cap at different values, etc.

If you look at our weight for physical hit and add our weight for spell hit to it, you get 1.1, which is the same as they have at EJ (we use the 372 ilvl weights).

Other than that, they are identical.  We add a weight for weapon DPS, because it is almost always true that a higher DPS weapon is better, regardless of the stats.  If you have no weight for weapon DPS, in rare cases a lower ilvl weapon may come out better than a higher ilvl weapon with really crappy stats, e.g. if you weight crit and haste really low, and the weapon only has crit and haste on it. 

[Baxpowa]

If you look at our weight for physical hit and add our weight for spell hit to it, you get 1.1, which is the same as they have at EJ (we use the 372 ilvl weights).

Interesting. So if we were to import your stat weights into Pawn, for example, we should add physical and spell hit raring together and use that as Pawn's hit rating?

[yellowfive]

Yes -- most other tools combine them into a single weight for hit, so adding them together would be a good estimate.

Note also that Mr. Robot accepts both a "base" weight and a "soft capped" weight.  I believe that tools like Pawn take your current stats into account, so e.g. if you're a dual wielder and you already have 12% physical hit chance, the weight you'd want to feed to Pawn would be a soft-capped hit weight.  In this case, you would want to add Mr. Robot's soft-capped physical hit to the spell hit value.

A good example of classes where this becomes a noticeable issue is elemental shamans and rogues.  With virulence now baseline, it's not a huge deal for DW DKs.

[Baxpowa]

Awesome, thanks for the info. You guys rock!

[Bandaj]

As far as I can tell, Mr. Robot doesn't take into account the human racial for 3% more spirit. For example, a human shadow priest offered two pieces that have the same stats except 127 spirit or 127 hit should always take the spirit piece, right?

For characters with multiple specs, it would be very interesting to have some ability to generate a "best of both worlds" gear list. For example, a discipline / shadow priest (or fire / frost mage, elemental / resto shaman, etc) would heavily favor a piece like Slippers of Moving Water (spirit) over Desert Walker Sandals (hit, equally rated for shadow, far down the list for Disc). I suspect it'd be a ton of work, so I'd hardly begrudge you ignoring it, but being able to track two specs side by side would be very handy.

Finally - thanks for some great work. This is the best site of this kind I've come across so far.

[Revulva]

Thanks much!

We sort of take into account the human racial... When given a choice between two items that are otherwise identical, Mr. Robot always picks the one with spirit for all hybrid casters. Do you have an example where this is not happening?

As far as a "hybrid optimized" set of gear... all you would really need to do is develop a set of stat weights. The tricky part is optimizing around the hit cap... do you do it in shadow mode and never let spirit go above the hit cap? Or, do you do it in healer mode and set a cap on spirit that you would need for healing? That is a personal preference. Obviously getting enough spirit to heal through a fight is going to waste a considerable amount of stats when switching to shadow. I personally like to keep around an extra weapon and cloak with heartsong/darkglow - that really helps with the mana.

Here is a possible optimization for a shadow/disc spec:
http://www.askmrrobot.com/wow/gear/priestdiscipline#v0-o0234Aq23-w0,2,1E8,7q7qNS7qVG,IAC643-s2,0,d4,AC-hVG,4-co,S
I capped spirit at 2000 - you'd have to adjust that number based on what you find you need.

[Bandaj]

Hmm, I hadn't noticed that it "broke" ties by defaulting to spirit. Good idea, that, across the board.

Still, let's take a less clear cut example:
Band of Secret Names and Twined Band of Flowers (both cost 1250 vp, so someone could easily have to choose between the two) -

Excluding Sta & Int as identical,

Secret Names:
127 Crit (* 44 = 5588)
127 Haste (* 57 = 7239)
Total: 12827

Flowers:
127 Spirit (* 58 = 7366)
127 Mastery (* 42 = 5334)
Total: 12700

Difference: 127

Actual robot difference is only 27, I assume because it already factors in reforging of crit to spirit and mastery to haste, respectively... though I'd think that'd only be a 50 pt difference. If it doesn't, well, tracking reforges in the point scale is a recommended new feature, and you may want to look into that loss of 100 points.

Anyhow, it currently recommends the Band of Secret Names as Best in (Second Ring) Slot. However, if I apply the human racial, The Twined Band of Flowers would gain about 221 points of value.

Not a big difference, but perhaps one still worth automagically calculating.

In the meantime, I'll just bump spirit by 3% in my valuation.

Thanks for the combined table. I'll play around with that!

[Revulva]

You are correct that the item list rankings do take into account the best possible reforge.

Mr. Robot does not necessarily use the items ranked 1 and 2, though. In the case of a shadow priest specifically, the item with haste on it will almost always come out a winner - assuming you can use reforging of mastery and crit to spirit/hit on the rest of your gear to achieve the hit cap.

I do see how the human racial could affect the item list ranking in this close of a case - and what you are doing is correct with the stat weights. Technically a human should weight spirit 3% higher than the other classes (it just doesn't really affect gear selection much). We actually do take the 3% spirit racial into account when we do the calculation for spell hit, btw. So - assuming you achieved ALL of your hit from spirit on gear... Mr. Robot will try to only put 1691 rating on your gear.