[Elemental] Cataclysm Discussion - Patch 4.3 - Page 11

Aavarn · 02/22/11, 11:12 AM

If you are looking for a nice addon to manage your cooldowns/DoT timers as an alternative to the ones you listed, I can thoroughly recommend EventHorizon. It even includes trinket procs, etc which can be very useful for maximising dot damage. Fully configurable, although only through text file editing (but its really not that hard)

I used it alot with my Shadow Priest and wouldn't know how to play without it now.

Moshne · 02/22/11, 1:19 PM

Originally Posted by revulva

I would really like to see it so I can make sure Mr. Robot - World of Warcraft is up to date for elemental shaman. For now we are using the stat weights in my analysis above, because they seem to provide the best gearing results.

You aren't seeing it because your methodology is flawed. The reason haste and mastery have changing values is the nature of the stats. Haste effects how many ticks of Flame Shock we get between recasts, it also affects how many (on average) Fulminations we get between Flame Shock refreshes (due to the number of casts.) Mastery obviously also impacts that function. Haste is also altering the number of LBs that we can cast during the deadtime of a shock CD, which is relevant when using an Earth Shock near the end of Flame Shock's duration and raises the chance of wasted LS charges.

This isn't an issue of how they scale with Spell power, which is essentially what your method is showing, absolute gear levels, it will vary in much smaller increments than that. You can observe this in SimC by doing a plot of both stats. You'll notice the changing values as you increase either haste or mastery, they aren't linear.

The other problem that in the methodology is the context in which you are seeing it. As I'm one of the people who've been using the language you are talking about, I can at least give my perspective on it. This is new territory for Elementals. In the past we've not had close stats, so we could usually just pick our best and stack it and spell power more or less to the exclusion of everything else. When the stats are close, you don't just go and reforge everything to the one that is .03 higher than the others. Stat weights don't happen in a vacuum and when you reforge 90 haste into mastery it affects the stat weights that gave you the formula you are using. This is a problem that melee DPS (namely our Enh brothers) have encountered for years. Players would see that StatA would overtake StatB, then go and re-enchant and re-gem for that stat and be astounded when their performance and the sims would now tell them it was a DPS loss.

I've attached a graph from my own character, as well as the BiS profile included in SimC. For the sake of context, SimC is giving me normalized stat weights of 1.0 for Int, 0.45 for Haste, 0.42 for Mastery at the current gear levels. You'll also noticed looking at my gear, that I don't necessarily practice what I'm preaching in the stats, I do tend to reforge to Haste over Mastery, and gem exclusively Haste orange gems/Haste enchants. This isn't because of ignoring the stats that the model is giving, but because I'm using a different model, one that includes movement.

It is important that we get a good model of fights without movement, despite the fact it doesn't model any current content. Until we have a functioning model of no complications, we can't begin to model anything else, but you need to be careful not to abuse the information you get from it. An output is only good for the inputs you give it.

♦ masanbol · 02/22/11, 2:35 PM

Originally Posted by Moshne

The other problem that in the methodology is the context in which you are seeing it. As I'm one of the people who've been using the language you are talking about, I can at least give my perspective on it. This is new territory for Elementals. In the past we've not had close stats, so we could usually just pick our best and stack it and spell power more or less to the exclusion of everything else. When the stats are close, you don't just go and reforge everything to the one that is .03 higher than the others. Stat weights don't happen in a vacuum and when you reforge 90 haste into mastery it affects the stat weights that gave you the formula you are using. This is a problem that melee DPS (namely our Enh brothers) have encountered for years. Players would see that StatA would overtake StatB, then go and re-enchant and re-gem for that stat and be astounded when their performance and the sims would now tell them it was a DPS loss.

This is the key. Elemental players need to change how they view stats - in the past we've been content to take snapshots of gear profiles, sim them, and take stat weights from that, when in reality that has almost no use for a player seeking advice 99% of the time. It worked before because the results were always the same for all gear levels: spellpower and haste were king, no exceptions. Now relative values between certain stats can fluctuate, as Moshne's graph shows very well.

This means that people asking "is haste better than mastery?" need to understand, as most enhancement shamans do, that the answer depends entirely on their specific gear, raid composition, and fight conditions. In that sense I suppose the stat priority list that we've kept around for the last two years is a little outmoded.

Hoverpuma · 02/22/11, 3:46 PM

While the "near-comparable stats" issue may be new to Elemental shamans, the inability of fixed stat weights to provide accurate predictive value has been well-known for years, in the context of hard caps (eg. hit rating) and soft caps (eg. haste and a 1.0 sec GCD). The suggested "TLDR" stat weights at the top of the linked article don't even accurately account for the relatively simple behavior of hit rating (highly valued below cap, useless above it), instead assigning hit/spirit a uniform weight slightly higher than haste.

Another issue I have with the linked article is that it attempts to draw general conclusions about the behavior of a function of 6 variables (int, sp, hit, crit, haste, mastery) by examining local gradients (the stat weights generated by SimC) at only 9 data points in a 6-dimensional domain. It identifies a single inflection point in that function between mastery and haste, and then jumps to the sweeping statement that "we always need subantially more haste rating on the gear compared to mastery to produce an optimal DPS result" based on the values of haste and mastery at the 9 test data points. The statement may or may not be true (or even roughly true), but justifying it based on such a sparse dataset is somewhat tenuous.

revulva · 02/22/11, 4:10 PM

I understand the theory behind what you guys are getting at. I understand that stat weights, in the past, have been shown to be a "poor" way to optimize gear. I've been doing theorycraft for a long time too - and I've made simulators, spreadsheets, etc. This new tool that we've created is taking a different approach to an old way to solve the gear optimization problem - and we're finding that we can achieve much higher accuracy than is widely believed possible using stat weights. I think this is helped in large part by the way that Cataclysm has itemized gear.

I think that you are misunderstanding what I am doing: I am doing empirical analysis of how stats affect elemental shaman in order to find some general stat weights which we can use to optimize gear across the board. Will it be perfect? No. But, if you want it to be perfect then you will need to manually optimize your gear. If I can create a tool/method that gets you within, say 0.25% of the theoretically maximum possible DPS for your set of gear... would that not be valuable? At some point we're so close to "THE" optimal solution that we can use it with no observable effect in the game. I know that many theorycrafters are unhappy with any solution except THE solution... but, you all know your math and hopefully can concede the point that in a practical (in-game) setting a small degree of approximation is truly negligible.

Lets look at some examples (not all of you are logged out in your elemental gear or have your elemental shaman listed, so I can't check everyone...) where I test Armory gear vs. Mr. Robot's suggested setup using my proposed default stat weights (I'm ignoring enchants with maelstrom crystals that are not already on gear in the Armory):

Binkenstein (aka Askledarea)
Armory: 23030 DPS
Mr. Robot: 23013 DPS (Mr. Robot - Character Profile - World of Warcraft)
0.07% difference

Hoverpuma (aka Kitoshi)
Armory: 22070 DPS
Mr. Robot: 22207 DPS (Mr. Robot - Character Profile - World of Warcraft)

Gravenimage
Armory: 22069 DPS
Mr. Robot: 22041 DPS (Mr. Robot - Character Profile - World of Warcraft)
0.13% difference

Achertite
Armory: 24581 DPS
Mr. Robot: 24606 DPS (Mr. Robot - Character Profile - World of Warcraft)

Using a generic set of stat weights I can improve or stay within an extremely small margin of error - even on the gear of people who really know what they are doing.

I would like to point out that empirical analysis like this is not unprecedented! I know that the number of trials I am running is low... but we understand a lot of the math behind what is going on, so it isn't necessary to do hundreds of trials to propose a hypothesis. I'm trained as a structural engineer - and I can tell you from experience that many of the methods and formulas used to hold up the buildings you are sitting in are based on empirical formulas and rules of thumb. Is it really so much of a stretch to apply the same type of analysis to a video game?

Right now all the "advice" out there for elemental shaman just tells them that haste = mastery and we don't really know because it is hyper-specific to your character. That is misleading. We know we need more haste on our gear, and we know at certain points they become competitive with each other... but I've also shown that you can simply favor haste and be just fine. Why not offer up favoring haste as the advice with a caveat: check your gear with sims if you want to squeak out that last 30 DPS?

Also - Mr. Robot doesn't simply apply stat weights for stats such as spirit and hit blindly - we have an algorithm that finds a solution which will get you very close to the caps - it then ignores those stats once you are capped. This is why we can provide a weight for spirit/hit and still find good results.

Hoverpuma · 02/22/11, 5:38 PM

Originally Posted by revulva

I would like to point out that empirical analysis like this is not unprecedented! I know that the number of trials I am running is low... but we understand a lot of the math behind what is going on, so it isn't necessary to do hundreds of trials to propose a hypothesis.

Yes, but there's a difference between proposing a hypothesis and leaping to a conclusion. The concern I cited was for a conclusion based on a sparse data set and no supporting math, not a hypothesis.

It's not wrong to make use of local stat weights in the context of global optimization. I can't recall whether Rawr does gradient-descent or something else, but it's an accepted optimization technique with known strengths and weaknesses. If you want to substitute a single global set of weights as an approximation for local gradients, as I believe you are proposing, it's just important to understand how your global weight vector compares with the variations in the local trend lines. If you can demonstrate that a single global weight vector is "close enough" to within an acceptable margin of error across a reasonable domain (say, ilvl 346-372 gear), that could be useful. The important thing, though, is that you would also need to make sure you're not extrapolating from a small local variation. Have a look at Moshne's graphs - while there's a pretty clear average trend, the actual lines spike up and down a decent amount, corresponding to fairly sharp shifts in the local gradient. In addition, this approximation would only be appropriate if the relationship between DPS and haste/mastery is roughly linear across the domain of interest, which you would need to demonstrate.

Now, if the tool supports stat caps in its global optimization, why do you need to artificially reduce the value of hit in the weights you list? If you're under the cap, hit is more valuable than int. The only subtlety is explaining that, since you can gem for int but not reforge for it, it's a better use of resources to reforge for the hit cap than gem for it. But that should be covered if the optimization algorithm is working properly. It was that kludge that led me to believe this was a gear-ranking tool (a la WoWhead's gear weights tool) rather than a global optimizer.

(Also, please don't cite my gear as an example of someone who really knows what they're doing for pure DPS minmaxing - I still have gear shared between Elemental and Resto and have had to make some compromises because of that.)

revulva · 02/22/11, 6:30 PM

And what do you think would be an acceptable way to show that these stat weights actually do work like I propose they would?

I've run tests on dozens of different sets of gear all across the range of stats from 346 to the 372 item level. I have also done similar tests to even greater extents for other classes and specs. I keep finding the same results... stats just don't change that much from 346 to 372 gear. Only tanks show a significant crossover point so far which we have needed to account for, due to diminishing returns on dodge/parry and mastery mechanics.

I changed the weight for hit for the exact reason you said: it is sub-optimal to gem for hit. Saying that hit/spirit is the best single stat would lead any general optimization algorithm to assume that gemming for hit/spirit would be optimal. This algorithm we have created works for every single class and spec in the game and any combination of stat weights a person comes up with. It simply takes a set of stat weights and creates the best set possible, taking into account applicable hard and soft caps. The stat weights you plug into it need to make sense, though. If what you want to do is: gear/gem/enchant for int, reforge for hit - the stat weights have to say that. Saying hit/spirit > int > other stats does not say that. Int > hit/spirit > other stats does say that. There is no "kludge" there - it is a better relative value of the stats since intellect IS more important than hit/spirit.

You could create stat weights customized to your character and plug them in to the site if you want, they are fully editable. We just provide defaults based on the latest theory and simulation tools so that people can optimize quickly without the need for personally dealing with the simulator, spreadsheet, etc. It just so happens the defaults are very accurate over a wide range of gear - and I thought that would be valuable information for the community.

In Moshne's graph, the only way that mastery is slightly better than haste is if he were to keep all other stats stable and add between 130 and 160 mastery. You can't even do that in reality. I think you are getting TOO bogged down in complicated analysis when a much simpler and more empirical analysis can lead to extremely accurate results. Just my two cents, and I know I am most likely in a minority on this forum... but so far results back up the hypothesis.

If someone finds an example where using these stat weights have a significant negative impact on their gear - I would like to know so we can tweak them. So far I just haven't found a case where that happens.

Binkenstein · 02/23/11, 2:48 PM

You're approaching the Stat Weight issue from too much of a technical viewpoint. I have always hated giving stat weights because they always fluctuate, and are always different for different players.

The general weights that I have been posting of late are general approximations of the values I'm getting from my spreadsheet (although when I get the time I need to do some modifications to how it's handling FS refreshes, which may be unlikely given the earthquake on Tuesday).

A weighting that suggests that haste is 5% better than mastery will see people stack haste over mastery, even to the point where the inverse becomes true. People use a single stat weight across the board, rather than tweaking it on a frequent basis. It's this reason why I have gone with a "Haste = Mastery" weighting, because in the long run the variations will be minor between the two stats.

madmaveric · 02/26/11, 5:46 AM

Originally Posted by Argannor

When regarding Cho'gall, there are 16 adds (4. cultist) and the space in between sometimes is really big... so that EQ won't hit more then 6 adds an CL is won't hit more then 4 adds. The AoE-Totem got the same problem -.- You got an idea how to fix? (tactics: a)

1) Position boss and raid on the side of the room rather than the center
2) put adds (and therefore spawn points) in the doorway
3) line of site adds round the edge of the doorway

Afakaz · 03/01/11, 6:02 AM

Apologies if this is the wrong thread, but I didn't see one for 4.1 mechanics.

Has anyone done the math on the Earthquake changes to see what kind of DPS change this will represent in its current form? I did a little bit of napkin math that follows; my schedule hasn't allowed for any progression past heroics so I don't know what absolute gear levels will look like, math is based strictly off of the tooltip.

Old earthquake: channelled for 8 sec, 542 damage/tick. 542 x 8 = 4336 total damage, 542 DPS, 542 damage per second casting. Channelled, unable to do anything while it's going (except that you can move if you cast Spiritwalker's Grace), and a HUGE DPM loss if you take any damage while casting. The only advantage I see to this is that it gains more strongly from haste, due to a longer cast time.

New earthquake: cast for 2.5 sec, 325 damage/tick. 325 x 10 = 3250 total damage, 260 DPS given that with a 10 sec cooldown you can't start casting another one until the previous one has ended, leading to a 12.5 sec interval between beginning one earthquake and beginning the next. But here's the important figure; even at the most unflattering DPS figures, that's still 1300 damage per second casting, and it allows you to continue other damaging abilities/movement while earthquake is going. If you can make up even a third of Earthquake's total damage with your normal damage rotation, this seems to me to come out to a fairly significant damage gain, albeit not from earthquake itself strictly in a vacuum.

Sorry if my math fails anywhere, I don't do my own calculations too often since I'm usually late enough on the scene that one of the fantastic posters around here has already done the math by the time I think of a calculation that should be done.

Kaideq · 03/02/11, 1:29 AM

Currently glyphed chain lightning is a better option for everything other than Magmaw hc/Cho'gall aoe wise. The new version of Earth Quake works really well on those 2 fights as you know exactly were you want to AOE in advance, as such you can get the maximum benefit out of the new design easily on those 2 fights.
With the constant moving on Maloriak due to the slime pools I'd say the new design is pretty nifty too , but all in all the ability is still a piss poor master talent tough.
The new fire nova strengthens the current; flame shocking multi targets and spamming chain lightning and lava burst even more, making earth quake even more of a gimmick than it already is.

Krusader37 · 03/02/11, 11:47 AM

The new fire nova strengthens the current; flame shocking multi targets and spamming chain lightning and lava burst even more, making earth quake even more of a gimmick than it already is.

Actually, from what I understand earth quake is not going to be a channeled spell anymore. It's just going to have a 2 second cast time with a 10 second cool down. I see it very viable in an AoE rotation with CL and flame shock/fire nova.

geeeeiono · 03/03/11, 6:59 AM

I've got a question regarding how mastery works. Do I need to reach the next mastery "point" to get the maximum benefit of the mastery? I was told this was the case, and I'm looking for a confirmation/deny with some evidence to back it up. For instance I was told that 13 mastery and 13.55 mastery would give me the same chance for the mastery proc.

• Jessamy · 03/03/11, 8:40 AM

You were told wrong. This is easier to see with enhancement spec than resto or elemental. It's true your character sheet will only update for whole number amounts of mastery. However the enhancement spec mastery increases elemental damage, and any amount of mastery will show up as tooltip changes for things like lightning bolt.

insylogo · 03/05/11, 5:04 PM

I'm confused about our 4pc set bonus. 10% off a 2 second cast should be .2 seconds off the cast time. If it's based on current cast time of 1.6 seconds, it should be .16 seconds off cast time. However, I'm only getting about .08 seconds off my cast time, or 5% less. What gives? Am I miscalculating how this haste works, or is the tooltip wrong or what?

Edit: Nevermind, the tooltip was wrong. My cast rate is really 1.47 down from 1.64, which looks a lot like .164seconds off.