• Aldriana

Oh, I'm sure we could get some upper-division Roguecraft threads going; the issue with such is that this sort of analysis isn't being done continuously. We go a few weeks without doing much at all, and then an interesting/hard problems come along so we have to break out the big guns. Hence, this sort of analysis tends to crop up every so often in all the various rogue threads, depending on where the question arose. But, if people feel a need for a thread devoted specifically to heavy-duty modeling of rogue mechanics, I'm up to the challenge .

Well, right; I mean, the attacks will on average be more or less randomly distributed across time - but 'randomly distributed" is sort of a different ballgame then "periodically every t seconds". Now, I fully expect the answers to be highly similar, and, frankly, I've been using this exact model in the gear sheet for a year (as in, dating back to before I made the sheet publicly available, back when I was just using it for my own purposes) without a second thought for it's accuracy. If the MH case is different by more than a percent or two, I'll be surprised.

On the other hand, I've been surprised before, so it probably *is* worth doing a quick simulation as a sanity check; alas, that simulation is significantly harder to write.

Regarding the cycles effecting uptime effecting cycles issue: yes, this circular reference is the essential problem of rogue modeling. It's, in fact, the reason I gave up on the Rogue DPS sheet and released the Rogue Gear sheet, since, at the time, it was not being handled well at all in the DPS sheet. The Gear sheet resolves this by iterating cycle computation a couple of times until the answer converges on a particular cycle, and uses that. This process, in fact, comprises a good solid third of the computations in the Gear sheet. A direct analytic answer would of course be better, but iterating to convergeance works pretty well and is a heck of a lot easier. I'm not sure how the DPS sheet handles it these days.

Note, however, that haste in particular tends not to be a major issue for cycle determination, given that it proportionally reduces the proc rate on PPM effects so procs per unit time are largely unchanged; the only feedback mechanism that allows haste effects to change the cycle is Combat Potency. So it's perhaps not as much of an issue as one might expect, but it does still require some careful handling.

 

Dontmindme

I agree that the MH hits should do little to skew the data. One can say that MH specials are "randomly" distributed but they are far less random than purely random given one usually has to wait for the energy to use them. The gaps between them may change in size, but they are still likely at least fairly evenly distributed throughout the timeline.

I'm a little more worried about the effect of Sword Spec procs / proccing Windfury / proccing Sword Spec again, but I'm not sure there is a really good way to avoid this or easily calculate this. I still don't believe this will throw the computations off all that much although maybe worth testing these spikes. OHs should work very well given a 100% SnD time only differing for iteration when Haste procs or on use Haste items kick in.

In terms of procs affecting cycles, my planned way to implement this on the DPS spreadsheet is to estimate the uptimes so the cycles have something to go by. Then come to an exact computation which will add or subtract dps based on the difference from the estimates. Following that, I plan to add a Macro that essentially will switch the computed uptimes in as the estimates (essentially iterating by Macro). There are too many items that are just calculated after the fact or which have estimated uptimes that could be made more accurate. The short answer ito how its currently modeled in the DPS sheet is that the estimated uptimes are probably working pretty decently in terms of cycle effects, but are not as accurate in terms of item value. Whereas the items that are formulated after the fact are probably pretty reasonably estimated in terms of item value but are not feedbacking into the cycles. As you might notice, I'm putting a sharper eye on accuracy of the proc mechanics. Once I implement more accurate formulas, item-by-item I can follow through on the above-mention plan of action.

Now, talking Deadly Poison. One can essentially determine the uptime by the previously mentioned formulas which also equates downtime. Which tells you that x % of the time you have 0 stacks up and 1-x % of the time you have 1-5 stacks up. If we subtract off the 0 stack subset and only look at the 1-5 stacks, wouldn't the % downtime equate to the same % downtime at 0 stacks?

I think it does, as the % time stuck at 0 stacks vs the rest of stacks seems like it should equate to the same percent of time stuck at 1 stack vs 2 or more as the chance to elevate to a higher stack should be the same as elevating from 0 stacks. If one continues that up the line you get...

0 stacks = %downtime determination -> D0 which is also the probability at 0 stacks
So wouldn't it follow that probability at the other stacks would be...
1 stacks = (1-D0)*(D0) -> D1
2 stacks = (1-D0-D1)*(D0) -> D2
3 stacks = (1-D0-D1-D2)*(D0) -> D3
4 stacks = (1-D0-D1-D2-D3)*(D0) -> D4
5 or more stacks = (1-D0-D1-D2-D3-D4)

Is there some flaw with this logic? If not, this seems maybe a more accurate way of determining uptime at different levels. Granted, it doesn't account for the damage being determined every 3 seconds from the start, but it may make sense that the percent time spent at a given stack size might be equivalent to the number of damage procs at a given stack level. Thoughts?

 

• Aldriana

So, I agree that the distribution of attacks isn't totally random - however, it's certainly very complicated, and "random" is probably a better approximation than "periodic" given that you have:

1) Autoattacks, which are periodic;
2) Yellow attacks - both your CPGs (Combo Point Generators) and finishers, which are... roughly periodic, but not precisely due to energy queueing, frequency of Combat Potency procs, etc.
3) Two forms of procs, which occur on a random selection of those attacks, and occur up to 1 second after the initial attack, some of which are generated by OH attacks which are on yet a third periodicity.

Now, even neglecting the fact that 2 isn't totally periodic, we have two periodic effects (with different periods), with two interwoven more-or-less random ones (they're not randomly distributed across all time, but I suspect if I gave you a list of times at which SS procced for me in a given fight and a list of random numbers, you'd be hard pressed to tell which is which.

So, it's not totally random - there *is* quite a bit of structure - but it's complicated enough structure that it's probably just better to think of them as random. However, I still don't think the answer will be very different from the periodic case - within a couple percent, to be sure.

Regarding the Deadly Poison estimate: I think the difference between the frequency of having no poison up and the frequency of having some other number of stacks up is not the same, as the transition table is different, as it's not possible for the number of stacks to drop from, say, 2 to 1 or 5 to 3. If poison stacks dropped off one at a time, something like that would be possible; since they all drop at once, I don't think it quite works (although I'm having a hard time explaining why).

That said: I did just have an idea how to improve this model to more accurately reflect the real-world case, but it's getting rather late so I don't have the brainpower to push it through to a solution. I'll try to do that tomorrow when I'm more awake.

 

Pale

20/0/41 beats 0/20/41 by about 2%. 11/9/41 does about 1% more dmg than 20/0/41.

Sorry if it looked as I tried to do that. I merely wanted to know why the spreadsheet gave almost equal damage for combat/swords and assassination/shadowstep. I know combat/swords is still the cookie cutter build for raiding.

Ghostly Strike, Evasion, and I can live for 15 more seconds, giving the raid time to recover from whatever mistake it was we made. I believe in survivability, dead rogues deal no damage. You believe in doing 0,4% more dmg, fair enough

The difference is minimal in damage output and as you said, depends on build. Putting my gear (with a S2 OH) through the spreadsheet gives the following results:

20/0/41 agi gems
730,98 (baseline)
1297,51 (baseline)

20/41/0 agi gems
736,77 (0,79%)
1311,98 (1,10%)

20/0/41 hit gems
729,06 (-0,26%)
1295,36 (-0,17%)

20/41/0 hit gems
736,33 (0,73%)
1313,18 (1,19%)

Replacing a 8agi with a 8hit results in -0,4% dodge, and buffed 0,04% dps increase, but only when fully raidbuffed.

As for my original question "why does the spreadsheet show such a minimal difference between combat/swords and some shadowstep build?", Dontmindme gave the answer:
So what the sheet does is allow you to eat all your own hemo charges, which in most raids won't happen.

So as I see it, there's nothing wrong with one shadowstep rogue in the raid. Your overall damage contribution to the raid won't differ much (-1,19% with my gear and some gem corrections) and it's a nice change of play. However, if you care a lot about being high on Recount/DM/etc.. go combat/swords.

 

koaschten

Corrected that sentence there for you. You only end up tanking when pulling aggro, 15 seconds won't change a thing. The only time it's not your fault is when the tank dies... and then it's over anyway.

 

songster

True, at low gear levels, ShStep can end up only a little behind combat swords. However, at that gear level, tri-spec Hemo is well ahead of both, so that would be your preferred option. Since the damage added by the Hemo debuff is static and does not scale with gear, Shadowstep will only drop further and further behind. At around T5/early T6 level, even tri-spec Hemo stops being worth the tradeoff.

 

Pale

Not appreciated, and not correct either. When on new content, mistakes are made, extra time means extra chances to learn/practice. I've tanked Mag for a full minute when the MT went down, OT'd Gruul when a tank went down. Immediately hitting vanish would only have meant that someone that doesn't have the escape routes ends up in trouble.

 

Ozzmar

And dead tanks don't tank anything... I don't mean to be disrespectful by any means, but offtanking (or compensating for mistakes) is not your job - damage is. If a tank dies, a Feral druid could much more easily soak up the damage adequately. Not to mention, he's going to actually be able to finish the fight - not just provide a band-aid that's just going to get bled through in 15 seconds (if not sooner).

You've made it abundantly clear that you think the miniscule amount of dodge gained at the cost of damage output is your decision, and that's fine. But it's an intangible benefit based totally on luck every time the boss takes a swing, and trying to argue that it's a justifiable approach to speccing and gearing your raid rogue isn't going to hold up in here, I'm sorry.

 

Dontmindme

The thing is that when a stack drops from 2 to 0 or 5 to 0 it does come back up through 1 stack or 3 stacks (at least most of the time). So, to some extent, it is dropping to 1 or 3 just through the predecessors (with a certain dropoff as sometimes it drops before it gets there). That dropoff can be reflected in the difference between the probability uptimes between a certain stack and its predecessor. The problem I've seen with other models is that the uptime at various intermediate stacks is generally considered constant which we know can't be the case, as we know it does sometimes drop from say 2 stacks before it reaches 3 stacks. So, certainly there should be a level of degradation as the stacks increase which this model reflects (the question being whether it reflects it correctly). I've got some ideas on how to test this model, but not sure when I'll have the time to put it together.

 

◊ Hanos

What you are still missing is that there is a "small" DPS loss with the best possible scenario, which would be even smaller with a Tri-Spec (Hemo Buffbot) Build, and you would still get Ghostly Strike, however, in the real world that "small" DPS loss from my experience is really more like 10-15% instead of 1-2%.

The ability to stack and chain cooldowns is huge. If you are looking at a fight like Patchwerk, yes Hemo comes close, however if you look at a fight like Gurtogg or Archimonde or Illidan, where you are running in and out, or you have a phase where you want to control aggro followed by a zerg, then Combat really pulls ahead. The power of Combat is the highly controlled burst, on Archimonde I can wait for the fear and then blow AR, BF, Drums, and call for Heroism and blow up, where you could... Hemo and S&D and Rupture.

As far as having a contingency for screwing up on T4 content... that isn't what most of us are concerned about. If your tank dies, it isn't the rogue's job to tank, it is the rogue's job to vanish and let the Feral or Warrior get aggro and restabilize. The truth of the matter is the most rogues who want to raid Shadowstep just don't want to eat the cost of respec'ing for PvP, and they want to PvP as Shadowstep. If that is the case and the extra DPS doesn't matter (and it doesn't in T4) then more power to you. But don't try to advocate it for high end raiding, I showed a number of parses of 20/0/41 vs Combat a week or so ago and the results were 10-15% not 1-2%.

 

◊ Shaker

So, just for further reference, because I suppose the NAME OF THE FORUMS didn't make it abundantly clear:

Yes. We, the rogue community at the Elitist Jerks forums, are very interested in DPS improvements for our characters at granularities of less than 1%.

Consistency. It's only a virtue if you're not a screwup.

 

Pale

@hanos and shaker: once again, i'm not trying to get shadowstep established as a raid build.

My rogue is my alt. I was just trying to figure out if the original post of this thread is correct with it's -12% estimate, and it's not. The personal dps might drop by 12% but the hemo debuff gives something substantial in return. The reason I asked is that we're currently at Vashj, and shadowstep has a distinct advantage for a rogue if he needs to be on elementals.

@Ozzmar: please try to know what you talk about before replying, saves everyone work

@Dontmindme: /cheer for actually knowing what you talk about and phrasing it in a helpful way

 

Stabmaster

It's irrelevant that you offtanked those bosses for a few seconds. If your tank dies, it's a wipe, and your offtanking isn't going to have any value, except as comedic relief as people shout "ROUGE TANK LAWL"

You don't learn or practice anything, either, as Rogue tanking has no place in raids. What can you possibly learn in a handful of seconds Evasion tanking? Nothing, aside from the fact that you get one-shotted, and that the healers need to keep the MT alive in the first place. If your 15 seconds of Evasion tanking mean the difference between a 1% wipe and a 1% win, then the extra (considerably higher) dps of Sword spec will be more effective in dropping him before a complete wipe.

You can try and justify Lolstep in raids as much as you like, but it's not good or DPS-competitive, except for, perhaps, very low-level raids that are easily puggable -- and in that situation, who cares? Maximizing DPS is important for the many fights where you're under an enrage time limit or other timing factors.

Dodge is useless for a raiding Rogue now that cleaves have basically been eliminated. There is really no effective argument to the contrary, despite your assertions.

 

• Vulajin

Relevant part bolded. And trust me, it's correct.

 

Ozzmar

Well if you'd kindly point out where I've contradicted anyone else in this discussion, I'd welcome the correction. I've only tried to be completely honest and helpful thusfar, and your dismissive tone is slightly disrespectful.

 

Ruqas

If you're not trying to get it established as a raid build, what are you doing posting? There is no clarification needed here. We all know the difference between Hemo builds' personal DPS and the DPS contributed to the raid. And a "nice change of play" is "ok" on farm content, but since you seem to be advertising the use of agility for mistakes made on progress content, it isn't. While it is true Hemo does give something "substantial" in return (substantial being used in this case as an excuse for doing less DPS overall), tri-spec still beats out Shadowstep.

What you are saying to Ozzmar is what you should be saying to yourself. You wasted everyone's time on making posts to show you how you either didn't think it through, or how survivability can easily replace stats with common sense and paying attention.

Even if you are on Lady Vashj running Shadowstep because it may prove more useful, you may not have the rest of the bosses on farm, which means that you are ruining your performance on the rest so you can perform well on another. Otherwise known as an exception to a rule. Of course, you may also consider the rest of bosses on farm. In that case, consider this last paragraph moot.

Now how she taketh mine eye.

 

◊ Shaker

No, you're just shitting up a thread which (admittedly) is prone to being shit up all too regularly, all because a) you couldn't read the first post, and b) like to get defensive about your build on a board where you disagree with the basic principles of a larger portion of the community.


(Edit to add) Just for clarification, we understand that you value the survivability and the mobility of that build more than the DPS increase you'd see from a different one. We just don't care, because in our opinions, it is not useful.

Consistency. It's only a virtue if you're not a screwup.

 

◊ Hanos

At extremely low gear levels the loss may only be 1-2% best case, however, the hemo debuff does not scale with gear, while everything about combat does (for example 2 piece T6 and Combat Potency). I would still argue that while the "theoretical DPS loss is only 1-2%" with your gear, the real world loss is far higher, the DPS of Hemo is very consistent, while Combat has far more burst. The more movement involved, the more value that burst has, for example Blade Flurry can be up 12.5% of the time, now if you are moving 50% of the time (not unreasonable on Vashj P2), now BF is up 25% of the time you are DPSing, same with AR.

For a fight like Vashj you are running all over the place, the ability to blow cooldowns to drop an Elite, or to burn her down during the enrage on phase 3 far outweighs the utility of being able to do something stupid like putting rogues on elementals in phase 2, you are better off sitting the rogue and bringing in an alt warlock if that is what you are trying to do.

 

Dontmindme

While doing a little math at work during lunch I discovered a flaw with the above mentioned Deadly Poison model. It looks to be yet another simple formula that would be correct of the frequency of one's attacks were 0. Without boring anyone with excess mathematical detail, that formula does not work and Aldriana's initial instincts were correct. Still some hope of making a similar approach work, but if I derive the details, I'll open up a Roguecraft 201 thread...

Nothing to see here...move along...

 

Ozzmar

Phht, post it I say! I love reading these discussions between you guys. They're interesting!

 

• Aldriana

On the Deadly Poison debate: I have good news, bad news, good news, bad news, good news, and bad news.

Good news: I'm pretty sure I've figured out a way to solve the problem in the general case.

Bad news: Doing so involves computing the eigenvalues of a 20x20 (or so) matrix - which can't be done symbolically in general, meaning that the best we're going to do is a numerical approximation... which is utterly useless, since we can already get that by simply running a simulation.

Good news: The approach simplifies nicely in the case where poison uptime periods are "long" - that is, where poison usually has time to stack up to 5 before it drops again.

Bad news: Even solving that requires computing the eigenvalues of a 4x4 matrix, which, while possible in the general case, is so unimaginably messy that it's really not any better as a solution.

Good news: I think there's a way to bypass the matrix entirely in this simplified case and directly compute what you need.

Bad news: The details of doing so are sufficiently messy that it's gonna take me a couple days to work it out.

So, long story short: I have an idea for an approach that I'm pretty sure gets a better answer than we currently have; if you're interested in knowing the details and/or would like to assist with the math, PM me; if there's enough interest, I'll create a upper-division rogue math thread (which I think should probably be called something other than "Roguecraft 201" to avoid confusing the casual forumer) and explain the approach there.

 

Pale

Thanks, point taken.

 

NvidiaN

A question for you intelligent folk whom I rather look up to:

On my forum we have a "DPS Wall of Shame" thread, in which we post WWS reports for various fights and shine a light on melee players who are slacking. After comparing my "buffed DPS" which from my understanding is the max possible DPS I can do, to my obtained numbers on fights such as Leotheras the Blind in SSC, I see that my obtained DPS is lower which only makes sense.

Is there some sort of generally accepted "formula" for determining the amount of DPS lost due to downtime? I understand that some fights just aren't rogue friendly, but I'm trying to find a way to put exact numbers to this idea.

 

• Aldriana

Short answer: no.

Long answer: For any specific fight, with not unreasonable effort, one could probably come up with a reasonable downtime estimate. However, in practice, since every fight is different, no concerted effort to do this has occurred. It's sort of on my to-do list, but it's pretty darn far down the list.

 

Busko

Hi guys i been trying to figure out how good haste is for us and how it compares to other stats like hit and ap etc. At the moment with my current gear it says that for me the Shadow-Walkers Cord is the best.