Sjera

My spreadsheet are based on calculating average damage to determine average dps.
As such what I'm interested in calculating here is the % chance that any give spell will be cast under the effects of clearcasting (to see if it has the +10% damage or not).

So:
The chance that this spell is cast under clearcasting % = the chance that either of the previous 2 casts crit.
If I recall my probability math correctly then
clearcasting % = 1 - ((1-Chance to crit) * (1-Chance to crit))

(i.e. in order to not clearcast - both previous casts must have been non-crits)

Originally I was just using our crit chance as the 'chance to crit' in the above - which is of course wrong.
I'm now using an overall average crit chance based on the ratio of FS,LvB,CL & LB and their relative crit chance.

Edit:
It would be interesting to see how this compares with your uptime method.

 

Gaguusi

I calculate average crit chance the same way. The results from our formulas are very close (46.39% crit chance = 0.725 uptime on mine, 0.713 on yours) though I'm not sure what causes the discrepancy.

 

• Binkenstein

Looking at an actual log, and I can see why the uptime is so messy.
Clearcasting log
I highlighted all the parts that should have been affected by clearcasting in red. The uptime according to StasisCL was 59%. That looks more like 80% (which it is, actually, as you can see here

I'm working on comparing the uptimes that WWS/StasisCL will see vs what should have occured. It all comes down to the combat log being a PITA for calculating stuff like this.
[sarcasm]Congratulations on coming up with a brilliant piece of theorycrafting there. We obviously didn't know how to do that before you posted it.... [/sarcasm]

Oh wait...

If you're not aware, the current discussion is on the uptime of Clearcasting. We jolly well know how to calculate the bonus once we've figured that out, thankyouverymuch.
The problem with a calculation like that is that it requires the same crit % for all hits.
Unfortunately, for a base crit chance C, we have C, C+5%, and 100%. This is where a simple formula will fall down.

Ideally, you need LB*C+5*C+5 + LvB*100 + FS*C*C+5, and divide that by LB+LvB+FS.

 

Gaguusi

Wow, way to be condescending and pompous. If you had actually read my post you'd have noticed that I asked for comments and possibly confirmation of whether this piece of TC is correct, not your approval. And if you'd read the post I've written before that, you'd have seen that there's a comment there about the uptime calculation.

Here it is:
(crit chance * total spells cast * cast time) / total spells cast

I'd still appreciate serious comments


edit: I don't want to start a fight, so I hope the non-TC part of the discussion can end here.

 

Ezareth

A quick question, I'd test it myself but I'm in the middle of rebuilding my computer.

Lava Burst's cooldown begins after completion of the cast or the start of the cast? I was under the impression is began after completing the cast.

 

Chant

Rather than working with an average critchance, I'd calculate the probability for CC to be active directly for any particular rotation. Just looking at LvB, 4xLB for sake of simplicity (and assuming FS is active), if we randomly select a spell, we have to distinguish two cases:

a) It is one of the two LBs following LvB - CC is guaranteed to be active
b) it is one of the remaining spells - the probability for CC to be active depends on the critchances of the previous two spells (that hit the target)

The probability for a) is 2/5, and 3/5 for b)

Consequently, the probability for CC to be active for any spell is 2/5 + 3/5 * (1 - (1-c)^2).
[Using an average critchance calculated as (4*c+1)/5 and inserting for p in (1-(1-p)^2) yields a different result.]

But should we even bother with calculating the probability for CC to be up for any spell, considering that the increase of spellpower has a different effect on total DPS depending on what spells it is active for?

 

Gaguusi

Correct, Ezareth, it starts when the cast is completed.

Chant, that's now the third formula we've got to calculate CC uptime. I don't suppose you have a way of telling which the correct one is

Seriously, though - I've used your formula (2/5 + 3/5 * (1 - (1-c)^2)) for each of the spells in the rotation, multiplied the result by the times that particular spell has been cast, summed up the results and divided them by the total amount of spells cast. The result (using my own crit chance of 31.41% for FS) was 0.7945, which is higher than either one I got from the other formulas used (see previous post).

 

Ezareth

Thanks. So this in effect means Lava Burst rotation is on 9.5(unhasted) cooldown instead of 8 then...with a minimum of 9 seconds.

Something else I've been wondering. With all of our problems with haste scaling....I wonder if 35 Spell Penetration to cloak is better than 23 haste.

I am seeing 4-5% DPS being "mitigated" in the combat logs by partial resists.

I've been out of the loop for a year or so on the raid changes, but people have told me that spell penetration is worthless in PVE.....yet how are we getting resists?

Has anyone done any testing on this front? I am starting to wonder if a high level of haste is preferable to other stats, if Spell Penetration is more useful than Haste I'd love to dump it.

 

Chant

Using an average critchance and applying that to 1-(1-c)^2 is incorrect because it assumes that the two events are statistically independent. But the probability for a random spell to crit given that the previous spell did not is higher than the average critchance, because it has an increased probability to be LvB, so to say.

Before calculating average uptime, you would have to modify the formula as to match your own rotation, of course.

But, as I said, I'm not sure we should use the average CC uptime (or the probability for a random spell to be affected by it) to calculate the damage increase. Instead, we might want to look at the individual probabilities for LvB, LB, etc. given particular rotations.

 

Gaguusi

Well, Bink was kind enough to make a graph showing the points where haste wasn't beneficial or could even reduce DPS - must be somewhere in these forums. Except for that haste should be pretty beneficial.

Despite that, I've also been wondering about whether the effectiveness of spell penetration has changed.

 

Chant

That boils down to crit chance * cast time, which means that Curse of Tongues (or a similar effect) would increase the uptime of CC (possibly beyond a 100% probability), which is nonsense.

It is a mere coincidence that your results roughly match what others have calculated.

 

Gaguusi

You're right, of course - could've figure that one out myself I guess. Appreciate the comment, though.

I'm still interested about everyone's thoughts about Sjera's formula (1 - ((1-Chance to crit) * (1-Chance to crit))) vs. Chant's (2/5 + 3/5 * (1 - (1-c)^2)). Also, if you're saying that the correct usage of yours would be with taking each individual spell's chance to crit into account, does that only leave a simulation or fight data as possible options?

 

Mmootimus

I think to get a very accurate model simulation will be required - thats not to say theorycraft can't get close enough to be very useful.

One issue I have thought of is spell ordering. I am not sure if the effect will be appreciable or not, but the issue of what to cast in the two guaranteed clearcasts after a Lava Burst may be important. Those two spells have a guaranteed 10% dmg bonus, whereas for the rest of our cycle its only around ~65%.

I have always been dead against spell pen, but have been wondering about this too. I have been too lazy to check whether its an artefact of WWS doing weird stuff or an actual boss mechanic (and if so, whether its one we can do something about).

 

Gaguusi

While I appreciate the accuracy of simulation, the usefulness of normalized TC results outweigh it for me.

I guess knowing that the 2 spells after a LvB was cast always get the 10% increased damage is useful, but I'm not really sure if it will affect the rotation in any significant way.

 

Chant

For a more comprehensive example, I currently use this rotation:

FS, LvB, CL, LB, LB, LB, LB, LvB, LB, LB, LB, LB

(other things being equal, moving the CL-cast to the slot right before the second LvB will benefit DPS)

There are three different types of spells when it comes to the probability of CC to be active:

a) those within 2 casts of LvB (CC is guaranteed to be active, so p_a = 1)
b) those preceded by 2 lightning spells (p_b = 1 - (1 - lightningcritchance)^2)
c) and the initial LvB, which is preceded by one LB and FS (p_c = 1 - (1 - lightningcritchance)*(1 - flamecritchance))

Accordingly, the probabilities for individual spells to be affected by CC are as follows:

LvB: 1/2 * p_b + 1/2 * p_c
LB: 3/8 * 1 + 5/8 * p_b
FS: p_b
CL: 1

I'm not taking into account the very first LvB here, for which the probability would be lower, of course. Further, I don't know how the FS-Dot-component interacts with CC. If we have to look at individual ticks and what casts precede them, things will get a little tedious, but it is certainly doable. Lastly, range may have an effect on what two spells previously hit the target, and the order would have to be modified accordingly.

EDIT: Spelling

 

Radek125

I've been following this thread closely in the shadows for a while now and I have a few questions.

I have 1951 spell power(with FT), 26% crit, 329 haste, 268 hit + [Snapper Extreme] = 308 hit rating and I am unable to get beyond 2300 dps on Patchwerk 25 man. I've done a few simulations, and they all say that I should be much higher dmg with my gear. Logic leads me to believe there is something wrong with my rotation but I am doing the standard rotation talked about here. FS LvB LB LB LB LB CL repeat. I still use CL because otherwise there is time lost in between casts. (I adjust this as needed based on haste and gdc timers)

Anyone have any idea whats wrong? Even a point in the right direction. I'm taking a look at my mods now to see if client lag may be causing problems but my fps is generally 30+.

My normal pve raid spec: WorldofWarcraft.com -> Info -> Classes -> Shaman -> Talent Calculator


Thanks

 

Manpile

[Thunder Capacitor] really needs to be looked at more closely I think. I had been using it along with Sundial of the Exiled for a very long time, but today I finally got Forge ember to drop and was very unimpressed with the results. Against a dummy in IF over 5 minutes per trinket (level 80, not boss, since I'm not going to ask a boomkin to keep FF up while i test) I was doing 2900-3000 sustained with Thunder Capacitor, and only 2700-2800 with Forge Ember in place of it. The normal dummies are close together so I was unable to use Chain Lightning in my rotation, but I was still able to squeeze in 5 LBs per LvB.

This is with 3 piece Valorous, the LvB totem (Patchwerk is a jerk), and exactly 2400 SP 18.91% haste, 35.54% crit before EO, with my totems down.

For anyone who doesn't know, the cooldown stated on the tooltip is incorrect; you can gain your 4 charges all at once, or within a second of each other, or whenever. The only internal cooldown is 2.5s once the LB proc is fired. At that time you are unable to gain a charge for 2.5s. The proc gains your crit chance (including those from talents), but doesn't generate lightning overload procs. I don't have skycall totem, totem of the elemental plane, or totem of hex so I don't know how it interacts with those.

It gains no benefit from your spell power, however it does gain extra damage from things like Earth and Moon. With that in mind, it is presumable that it will also scale with the modifier being added to elemental oath, which is quite nice. It may scale with the flat 4% boost from the new Glyph of Lightning Bolt too.

Fully raid buffed it is extremely common to get a proc every 10 seconds. It hits for 1200ish, crits for 2kish before earth and moon. I am a novice at theorycrafting but I would not be surprised if I was getting 150ish DPS from this trinket during raids, which I think is better than Forge Ember.

Sorry if this has already been discussed!

 

Radek125

http://elitistjerks.com/f79/t36782-s...1/#post1021578 - In short - yes it is better then our other trinkets and it should not be green. I expect the proc to be changed similar to how [The Lightning Capacitor] is.

 

• Binkenstein

It's more that you have posted a fairly simple piece of TC that hinges on something we're currently investigating. Additionally, if you ignore that uptime bit, we knew that already. It's like someone coming in and saying "hay guys, I worked out that the spell co-efficient of Lightning Bolt is 0.714!". While it may be correct, it's essentially useless.
I think I need to write this down again in big letters.

Fixed rotations are bad, mmkay?

If you use a priority rotation, like shadow priests use, haste scales normally (ok, so there are slight variations, but it's positive scaling all the way).

Now, while the formula posted previously will work fine for a fixed rotation, if you use a priority rotation you'll need to look at a longer timeframe to get your spell counts correct. I'm using 8.75 LB (ie: nature crit squared), 1 FS/LB (nature crit times fire crit), and 4.75 LvB (100% crit chance), and as haste goes up, that LB figure will increase.

 

Gaguusi

Which is why I asked for verification - I honestly don't see why such a condescending tone is necessary.

I've seen that you use a priority system with a time limit in your spreadsheet, but may I ask how you determine for each cast whether it's a crit? Or do you normalize damage according to each spell's crit chance?

 

• Binkenstein

Yes, you are right. Any uptime calculation is going to be proc chance times event that can cause proc divided by timespan, and any % bonus will be that uptime * bonus. Nevermind the fact that we've been using this basic calculation for a while.

I generated a list of random numbers, 0-1000, and put a list of 10,000 of them onto a separate page. I reference these (divided by 1000) to give a "roll". If the roll is below the crit chance, it's a crit, above and it's not. This way I can prevent massive recalculations via rand(), make the figures stable/more accurate, and also it means we can compare an increased crit chance more accurately (same roll events, different cut off points). As an additional variance reduction, I replicated everything 15 times (using different sections of the random number list) and averaged them.

 

Gaguusi

I think I'll end the prior conversation from my end now.

Thanks for the ingenious way to create a somewhat random yet static environment. However, shouldn't that be equal to or below?

 

• Binkenstein

It's a 0.001% difference, so it wouldn't matter in the long run anyway.

 

Chant

I don't see what's wrong with using a fixed rotation for a particular value of haste (in a static fight). Of course, upon analyzing how damage scales with haste, you might see awkward results unless the sequence of spells is altered dynamically, and a priority rotation does that for free.

When working with a priority system, in order to calculate the clearcast-probabilities, one must first determine the sequence of spells it generates and start from there, as you said.

EDIT: back to the drawing board, the logic was flawed ..

 

Mmootimus

I hope that the Elemental TTT entry is ready soon, because the amount of assumed knowledge required to post in this thread is getting rather daunting

It would be nice if newer posters here - assuming they aren't complete dribbling morons - didn't feel inclined to quit the boards as soon as they arrive.