Enhance Shaman: The Collected Works of Theorycraft, Vol I - Page 333

frozndevl · 04/02/08, 5:34 PM

I don't even know where to begin with regards to statistically relevant populations in regards to 1 raid being enough to base a conclusion. In short 1 boss fight can fluctuate extremely based on our melee mechanics, therefore relying on 1 boss to fight to create a conclusion is just bad math/modeling/stats/etc.

And as is quoted on a blog I frequent, results based analysis does not adequately predict future performance.

◊ Shalas · 04/02/08, 5:41 PM

Originally Posted by Monkeysnarf

...

PRNGs do not work how you think they do. Statistics does not work how think it does.

Originally Posted by Toots Hepcat

Incidentally, the Scryer bonus on [Shattered Sun Pendant of Resolve] is sort of ridiculous. +100 Expertise Rating means that, when it procs, you'll be way OVER the expertise cap (it's a 6.25% reduction to dodge/parry, not counting the passive bonus). Either somebody at Blizzard is bad at math, or there's something about Expertise that we aren't getting.

The obvious explaination to me is that they consider the parry reduction much more useful than we do, and even with the proc active it's hard to cap that. The dodge cap is pretty well confirmed, so I'm not sure what else it could be other than just a badly chosen stat for the effect.

Rhaegal · 04/02/08, 5:43 PM

Originally Posted by Monkeysnarf

As far as feel goes, "feel" is just another word for experience. There is no random. Therefore a sample that's only 1 raid large is large enough to know what you need to know. It might not tell you everything you'd like to know about bosses that you have no idea of what their abilities and defenses are, but it will tell you everything you need to know about that specific raid.

No, it doesn't. Reading through this thread, and through personal experience, you would notice that enhancement DPS can vary significantly on the same encounter on different nights, by the same shaman in the same gear pressing all the right buttons at the right times. Our DPS is much more prone to random factors than most other classes simply due to Windfury mechanics.

Yes, I say random factors. Do some research on how "random" numbers are generated in computing, and you'll know that your point is not valid. Random elements generated by programs are just as prone to streaks as real-life "random" situations like flipping a coin (assuming, of course, it's using a halfway decent algorithm, but that's a different story entirely--I like to hope that Blizzard's RNG is halfway decent).

The sim tells us what we can expect to see on average under ideal conditions (i.e., not running around chasing a boss, warlocks/druids not allowing CoR/FF to fall off on accident, not threat-limited, etc.). In other words, it tells us what our best potential DPS could be, averaging out good/bad luck streaks. That sort of information simply can't be produced in-game.

Monkeysnarf · 04/02/08, 5:51 PM

Originally Posted by frozndevl

I don't even know where to begin with regards to statistically relevant populations in regards to 1 raid being enough to base a conclusion. In short 1 boss fight can fluctuate extremely based on our melee mechanics, therefore relying on 1 boss to fight to create a conclusion is just bad math/modeling/stats/etc.

And as is quoted on a blog I frequent, results based analysis does not adequately predict future performance.

I know where you can begin; you can start with the fact that the random number generator is not random. Then you can progress with the fact that you and everyone who has EVER played this game has never observed 2 boss fights that were substantially different when ALL ELSE WAS EQUAL. You can finish with the fact that real world math/modeling/stats/etc loose their value when you remove the "real world" from the equation.

• Toots Hepcat · 04/02/08, 6:01 PM

I've been to Magister's Terrace three times in three days. Same gear and group each time.

One night I did 921 dps. Once, 814. And once, 791.

That's a deviation of 17%! That's why a single parse is irrelevant to theorycrafting. If I had to rely solely on parses to evaluate gear, specs and strategy, I could never do so reliably. As much as you seem to think there's some magic behind computer-generated randomness that makes it different from a roll of the dice (Stephen Wolfram would agree with you), the fact is that everybody has good nights and bad nights.

Furthermore, the equations we use to calculate things like "DPS" can be gamed through playstyle. For example, if I have to run out of melee range and stay there for longer than 3s, I'm going to shoot off a lightning bolt. Guess what? This means my DPS stats will actually be lower than if I hadn't done anything.

This is why we have formulas. This is why we have EP. This is why we have the simulator. And they all work very well.

The second you start trusting your "feelings" more than reason, you're resigning yourself to making poor choices. Such as making foolish statements like "everyone who has EVER played this game has never observed 2 boss fights that were substantially different when ALL ELSE WAS EQUAL." On the contrary; i don't think I've ever had two such fights where the results were the same.

Monkeysnarf · 04/02/08, 6:18 PM

Originally Posted by Rhaegal

No, it doesn't. Reading through this thread, and through personal experience, you would notice that enhancement DPS can vary significantly on the same encounter on different nights, by the same shaman in the same gear pressing all the right buttons at the right times. Our DPS is much more prone to random factors than most other classes simply due to Windfury mechanics.

Yes, I say random factors. Do some research on how "random" numbers are generated in computing, and you'll know that your point is not valid. Random elements generated by programs are just as prone to streaks as real-life "random" situations like flipping a coin (assuming, of course, it's using a halfway decent algorithm, but that's a different story entirely--I like to hope that Blizzard's RNG is halfway decent).

The sim tells us what we can expect to see on average under ideal conditions (i.e., not running around chasing a boss, warlocks/druids not allowing CoR/FF to fall off on accident, not threat-limited, etc.). In other words, it tells us what our best potential DPS could be, averaging out good/bad luck streaks. That sort of information simply can't be produced in-game.

This is actually the biggest problem with this thread. It gets ahead of itself. If you saved every WWS and gear selection from every raid you've ever been on, you would have a 100% sample size of your performance. To then say that those numbers don't matter because you only went on 1 raid a week for 4 years = 208 raids instead of 10,000 raids (1 raid a week for 192 years) is just silly. Worse than silly, it just doesn't matter.

Listen, to take information obtained from a sim over actually information obtained in-game is nonsense. The sim does not kill bosses, your actually performance does. The fact that you might have to run around and your only dps, while you're moving will be shocks is important. It might not be important to tank and spank fights, but it's important to movement fights.

If you're after your average expected output under ideal conditions, you wouldn't use a sim; you’d use a spread sheet. It's a lot easier to calculate a 30% proc rate in a spread sheet than it is in a sim. This may come as a surprise, but a spread sheet will return 30% if you ask it to calculate a 30% proc rate.

A sim is trying to tell us what we'd actually expect to see in a raid. So then you guys take what you actually saw in the raid and say it means nothing because your sim said something different, as if the sim is going to magically post-hoc ankh those bosses you killed and undo everything you did.

iconocclast · 04/02/08, 6:22 PM

Just looking for a few ideas as to after clicking "Theorycraft" is there a typical time that Yo!'s takes to complete?

Monkeysnarf · 04/02/08, 6:25 PM

Originally Posted by Toots Hepcat

I've been to Magister's Terrace three times in three days. Same gear and group each time.

One night I did 921 dps. Once, 814. And once, 791.

That's a deviation of 17%! That's why a single parse is irrelevant to theorycrafting. If I had to rely solely on parses to evaluate gear, specs and strategy, I could never do so reliably. As much as you seem to think there's some magic behind computer-generated randomness that makes it different from a roll of the dice (Stephen Wolfram would agree with you), the fact is that everybody has good nights and bad nights.

Furthermore, the equations we use to calculate things like "DPS" can be gamed through playstyle. For example, if I have to run out of melee range and stay there for longer than 3s, I'm going to shoot off a lightning bolt. Guess what? This means my DPS stats will actually be lower than if I hadn't done anything.

This is why we have formulas. This is why we have EP. This is why we have the simulator. And they all work very well.

The second you start trusting your "feelings" more than reason, you're resigning yourself to making poor choices. Such as making foolish statements like "everyone who has EVER played this game has never observed 2 boss fights that were substantially different when ALL ELSE WAS EQUAL." On the contrary; i don't think I've ever had two such fights where the results were the same.

I bet if you posted those 3 WWS we could see exactly where YOU (party make up, buffs, other players, etc.) made the difference through play style and button mashing than where the random generator screwed you on 1 day and bless you on the other.

◊ Shalas · 04/02/08, 6:26 PM

Originally Posted by iconocclast

Just looking for a few ideas as to after clicking "Theorycraft" is there a typical time that Yo!'s takes to complete?

10k hours with stat weights enabled takes about 5 minutes on a 2 Ghz Pentium M for me.

Rhaegal · 04/02/08, 6:36 PM

[e] Nevermind, I'm stepping out of this argument.

◊ Kasi · 04/02/08, 6:38 PM

Originally Posted by Monkeysnarf

This is actually the biggest problem with this thread. It gets ahead of itself. If you saved every WWS and gear selection from every raid you've ever been on, you would have a 100% sample size of your performance. To then say that those numbers don't matter because you only went on 1 raid a week for 4 years = 208 raids instead of 10,000 raids (1 raid a week for 192 years) is just silly. Worse than silly, it just doesn't matter.

Listen, to take information obtained from a sim over actually information obtained in-game is nonsense. The sim does not kill bosses, your actually performance does. The fact that you might have to run around and your only dps, while you're moving will be shocks is important. It might not be important to tank and spank fights, but it's important to movement fights.

If you're after your average expected output under ideal conditions, you wouldn't use a sim; you’d use a spread sheet. It's a lot easier to calculate a 30% proc rate in a spread sheet than it is in a sim. This may come as a surprise, but a spread sheet will return 30% if you ask it to calculate a 30% proc rate.

A sim is trying to tell us what we'd actually expect to see in a raid. So then you guys take what you actually saw in the raid and say it means nothing because your sim said something different, as if the sim is going to magically post-hoc ankh those bosses you killed and undo everything you did.

You seem to be missing something here. Of course there are things like AoEs, stuff like being slept on Anatheron, where the Gathios kiter moves him, other boss movements, being the victim of the RNG, having horrible luck on crits. None of these you have much control over. And the ones you do can never be explained through a simulator or TC anyway. I play mostly as elemental, but what explains that I do more dps than the mages in my raid? Because I do things to maximize my dps both pre battle and during battle. During battle is the things I listed above. Keeping uptime on bosses, not wasting time, minimizing movement, knowing when to redrop totems, knowing when to pop trinkets/cooldowns, knowing how to stay alive. None of this can ever be explained in a theorycrafting thread other than basics. (well the trinkets can, little else though no) For damn sure none of it can ever be measured and quantified.

What this thread is for though is the pre fight preparation. What buffs to use, what gear to use, group comp, enchants, gem choices. This is to ensure that before you go through all the random crap that deviates you from doing what you want (staying on a bosses ass never moving and doing your cycle perfectly as you can) your dps is as high as it can be. This thread deals with the pre fight work. To say that it doesn't matter is stupid.

Joy · 04/02/08, 6:48 PM

Having just done Brutallus for the first time yesterday it must be said that this fight is amazingly fun for melee.

I respeced an elemental subspec for the first time and had moderate results Wow Web Stats .

I was consistently able to sustain just under 2100 dps while twisting, on the kill I forgot to repot after my elixir of demonslaying wore off so lost a bit of DPS there. I had never been ele before so wasn't sure about whether to keep a 1;1 rotation so ended up doing 1;2 letting Flameshock get it's last tick. I also ran oom for about 10s because JoW was knocked off and not reapplied.

There were a few times my rotation fell apart because I was leading the raid and had to move groups or type something out. 3 drummers in the group and about every buff imaginable bar world buffs.

With a Mounting Vengeance and a SoC I think I could comfortably get around 2120 dps.

Anyway as I said I loved this fight and will look forward to it every week.

Edit -sorry linked a wws that didn't include pets. Wow Web Stats should be correct.

philomon · 04/02/08, 7:59 PM

Originally Posted by Malan

If you want to argue that the coefficients for Flame Shock are wrong on wowwiki etc, than we can explore that. If you are arguing that my math is wrong with the assumption that the coefficients are right, you're incorrect. I did it both ways (split the DD and the DoT and added the dmg together vs doing them in one calc) just to be sure and they both came out the same.

I'll make a case for differing coeffecients in a subsequent post with a little more rigor (thankfully there's very little variance to account for =D) and various levels of +dmg to verify everything. I linked to the article to display older findings, but then posted personal findings that differered. The point of either number set was to point out that a correction was needed in flame shock's 6 sec dps.

But using the numbers from wowwiki (15 and 52), the 6 sec dps of flame shock is once again:

(((377 + (750 * .15) + (.5(420 + (750 * .52))) / 6 ) * 1.36 =
((489.5 + 405) / 6 ) * 1.36 = 202.75 dps

(not 246.9, and now less than earth shock in terms of 6 second output)

With a split coeffecient, the instant portion of the damage is devalued and the dot is meatier, rather than your current inception that results in the instant portion doing a lot more than it should, thus skewing the 6 second output.

Yakout · 04/02/08, 11:04 PM

[e] Deleted. No good reason to waste thread space trying to clarify basic principles of statistics or clear up misconceptions regarding the fundamental nature of PRNGs.

◊ Shalas · 04/02/08, 11:20 PM

Streaks are not "non-random". Not seeing streaks in small sample sizes incredibly frequently would be non-random.

Barring data corruption PRNGs are technically deterministic, but I don't think it's any coincidence that every person I've ever seen point out that PRNGs are not random has demonstrated a complete lack of understanding of how PRNGs work. In WoW's case, (++i % 100) would probably be indistinguishable from a real RNG in most cases. The timing data from user input is considered random for the same reason that rolling a die is considered random, and WoW's combat system is full of that. With a real PRNG like IBAA combined with that random data the result is nearly identical to a quantum RNG.

• Disquette · 04/02/08, 11:39 PM

I can't believe I'm taking a bite on a total troll cookie, but here goes...

Originally Posted by Monkeysnarf

This is actually the biggest problem with this thread. It gets ahead of itself. If you saved every WWS and gear selection from every raid you've ever been on, you would have a 100% sample size of your performance. To then say that those numbers don't matter because you only went on 1 raid a week for 4 years = 208 raids instead of 10,000 raids (1 raid a week for 192 years) is just silly. Worse than silly, it just doesn't matter.

Wrong. It's not a 100% sample size of your performance. It's a 100% sample size of your past performance. Hey, if you flip a coin 3 times and it's heads every time, are you going to tell me that real life experiences mean that you should expect heads 100% of the time for the next 3000 flips? No, of course not, because you're smarter than that.

Also, it's just plain ignorant to say "well it's all in the wws, so you can account for all the variables". We would get no where if the only way to predict dps was by running regression analyses with the variables of all the movements, party makeup, consumable uses, etc etc. A sim is a way to look at your own dps, ceteris paribus.

Listen, to take information obtained from a sim over actually information obtained in-game is nonsense. The sim does not kill bosses, your actually performance does. The fact that you might have to run around and your only dps, while you're moving will be shocks is important. It might not be important to tank and spank fights, but it's important to movement fights.

True, the sim can't kill things for you, but it can tell you what the optimal gearing is for a controlled situation, and we have taken it as a given that the best gear for a tank 'n spank is often going to be the best gear for most fights. If you're basing objections on the fact that sometimes we're not able to do our job, then that's your prerogative, but I personally would find it silly to discount all the rogue theory crafting based on the fact that I would shoot my bow for a lot of the time on Baron Gheddon.

If you're after your average expected output under ideal conditions, you wouldn't use a sim; you’d use a spread sheet. It's a lot easier to calculate a 30% proc rate in a spread sheet than it is in a sim. This may come as a surprise, but a spread sheet will return 30% if you ask it to calculate a 30% proc rate.

If you don't realize why the majority of people here are laughing at the naivety contained in the paragraph quoted above, then you obviously don't understand shaman issues. With a nod to Tornhoof, who has done an excellent job, the sims don't have to worry about being good approximations, they can simply be the best that we have available.

A sim is trying to tell us what we'd actually expect to see in a raid. So then you guys take what you actually saw in the raid and say it means nothing because your sim said something different, as if the sim is going to magically post-hoc ankh those bosses you killed and undo everything you did.

See my first example with the coin flips. If I get really high dps 3 raids in a row, should I expect it in the fourth? Maybe I was lucky and didn't get ported? Maybe it was good group composition? Maybe I had extra consumables? Maybe it was my stormstrike timing? Maybe it was a nice string of crits? Maybe it was a combination of all the above?

The sim takes all of the variance that is possible out of the equation so that you don't have to spend quite literally hours of time scrutinizing every aspect of a WWS parse, combined with frame-by-frame Fraps, in order to achieve the same result.

Sorry to the rest of you for prolonging this garbage of a sub-thread.

Vellius · 04/03/08, 1:59 AM

Originally Posted by Shalas

PRNGs do not work how you think they do. Statistics does not work how think it does.

The obvious explaination to me is that they consider the parry reduction much more useful than we do, and even with the proc active it's hard to cap that. The dodge cap is pretty well confirmed, so I'm not sure what else it could be other than just a badly chosen stat for the effect.

I... would imagine that the far more obvious explanation is that it's intended for tanks, given the 48 stamina and complete lack of pure DPS stats; hit and expertise have a definite place for threat sets, after all.

Also, I guess, there's the fact that the Scryer proc is +100 dodge, which couldn't even begin to be construed as a DPS proc.

Lethnon · 04/03/08, 2:58 AM

good day everyone!

I have a simple question today. Does anyone know of an addon that will let you know of the cataclysm 2 piece bonus proc? I realize it isn't all that, but hey, an instant heal is an instant heal, might as well take advantage of it. I haven't had any luck on WoWAce or Curse.

Thanks in advance!

iconocclast · 04/03/08, 3:04 AM

Originally Posted by Macar

Try using Yo!'s simulator.

Edit: automatica beat me to it.

Originally Posted by Lethnon

good day everyone!

I have a simple question today. Does anyone know of an addon that will let you know of the cataclysm 2 piece bonus proc? I realize it isn't all that, but hey, an instant heal is an instant heal, might as well take advantage of it. I haven't had any luck on WoWAce or Curse.

Thanks in advance!

Enhancer

Mengus · 04/03/08, 3:09 AM

Originally Posted by Vellius

I... would imagine that the far more obvious explanation is that it's intended for tanks, given the 48 stamina and complete lack of pure DPS stats; hit and expertise have a definite place for threat sets, after all.

Also, I guess, there's the fact that the Scryer proc is +100 dodge, which couldn't even begin to be construed as a DPS proc.

I'm scryer, and it proced a 100 Exp Rating proc for me... I use it in PVP, where it can come in handy... I'm not thinking of using it for DPS, but since pvp gear doesn't have expertise, it's STA and static Expertise along with the proc makes it worth consideration.

Atren · 04/03/08, 5:32 AM

Originally Posted by Lethnon

good day everyone!

I have a simple question today. Does anyone know of an addon that will let you know of the cataclysm 2 piece bonus proc? I realize it isn't all that, but hey, an instant heal is an instant heal, might as well take advantage of it. I haven't had any luck on WoWAce or Curse.

Thanks in advance!

Just a little remainder that the instant heal will reset swing timer.

Expertise is very powerful tanking stat for threat and avoiding parry burst. Since parry is a lot higher than dodge then it also has much more room before cases to be useful for tanks.