[Ret 5.0 Pre-MoP] - Panda-ring to all players - Page 30

Podima · 09/14/12, 9:26 AM

Originally Posted by Krekko

Originally Posted by Blizzard Entertainment
It was just a few bugfixes to Divine Purpose, not a functionality change. Here's a more detailed description of what was fixed with Divine Purpose, for those curious:

-- Word of Glory was mistakenly giving two chances to proc Divine Purpose (one for the cast, one for the heal / damage). It now only gives one chance, as you'd expect.
-- As previously mentioned, Divine Purpose's proc chance is supposed to be reduced to 8.33% or 16.66% if you cast a finisher with 1 or 2 Holy Power. That was also incorrectly applying to casts of finishers with Divine Purpose active. For example, suppose you had 4 Holy Power, and cast Light of Dawn (dropping you to 1 Holy Power). That procs Divine Purpose. You then Light of Dawn again. Previously, it would see that you're at 1 Holy Power, and give that second cast only an 8.33% chance to proc Divine Purpose again. This fixes that, so that casts made with Divine Purpose have the full 25% chance to proc Divine Purpose again.
-- Some finishers would proc Divine Purpose instantly when you cast them, and some would have a moment of delay between the cast and you gaining Divine Purpose. Those have been fixed to always proc instantly; it should feel much more responsive now.
-Source

To follow up on the change earlier.

Going back to this briefly: I assume that this bug wasn't present in the Simcraft implementation? It's good that Blizzard is keeping a careful eye on DP, even if it's still currently the weakest of the 3 level 75 options (though not by much, per discussion in this thread).

Zephur · 09/14/12, 9:41 AM

Originally Posted by Belatar

Zephur, I get your point and that might enhance the result, but I am not yet convinced this is the optimum.

It depends on your old stat weights being "right" and optimal which I am generally challenging with my question. Or to put it differently: What do you do the first time when you do not have any prior weights/reforges/enchants/gems?

Take 5.0.4 for example. If we cant to come up with a baseline gearset, our choices are extremely limited due to the lack of options. Since higher ilvl is almost always an upgrade (and its usually possible to napkin math situations where this might not be the case, prompting further testing), the only changes that might make a difference right now are shoulders, weapon, and trinkets. This does not take long to test, and resulted in the current Haste > Mastery > Crit on fights that last more than a few minutes.

Originally Posted by Belatar

In my opinion simc would need to do its 50.000 iterations for each possible combination of gem/enchant/reforge and then compare the results. This would take ages of course to compute, but in my opinion is at present the only way of getting the optimum result with maximum confidence. And I believe this is not the way the sims work presently.

While I agree that something like a genetic algorithm to test priority lists and enchants/gems/reforges would be interesting and potentially extremely useful, that just isn't the way SimC works at present. It can simulate some reforging, but that's about it.

But what I said above applies here. With baseline gear, the best gems/enchants/reforges become apparent relatively quickly, usually due to a limited number of options. And in those cases where there are two or three interesting options, it doesn't take an inordinate amount of time to sim them.

Originally Posted by Belatar

You would of course try to reduce the number of combinations possible by ruling out obvious nonsense like i.e. "spirit", but if you do not start off clean every time you will get a biased result. You are after all using results from a different gear setup which are not accurate anymore to generate the new one. At least that´s still my opinion at this point.

Yes, there is bias in using the old gear weights, but it's justifiable bias. Changing a single piece of gear out of fifteen possible items just doesn't affect your weights enough to make the old weights worthless; your gear is still ~93% the same, so your old weights are still ~93% accurate.

Hinkyz · 09/14/12, 9:57 AM

Originally Posted by Elidra

I don't see any usage of Harsh Words, which could push ranged Ret even higher.

The damage from harsh words is much less than that of a wrath hit so we decided the only use for holy power was refreshing inq for the 10% crit.

Originally Posted by Balhale

Also, I see things like you having only 83% uptime on Inq, only have two AW in a 7.5 minute fight, only have 3 HA in a 7.5 minute fight, only five execution sentences in a 7.5 minute fight.

The inq uptime and cd usage seems weird to me, the only reason for that would be if we were testing something, which could entirely be possible (and another reason why the logs aren't really for comparing, just showing off the damage of wrath)

The execution sentence use I can definitely explain though. With this boss fight being a council fight, that gives 3 20% health phases, and its almost impossible to get an execution sentence off during them or when using AW as theres so many holy power generators which are higher priority.

Balhale · 09/14/12, 7:06 PM

ES is higher priority than HP generators, simply because it hits roughly four to six times as hard. That's actually largely why it's better than Holy Prism - it's larger damage advantage makes it strong enough to prioritize over the generators, while Holy Prism, which is just barely stronger than Hammer and Exo isn't strong enough to prioritize and so gets delayed heavily.

Charybdis · 09/15/12, 12:01 AM

Originally Posted by Belatar

Zephur, I get your point and that might enhance the result, but I am not yet convinced this is the optimum.

It depends on your old stat weights being "right" and optimal which I am generally challenging with my question. Or to put it differently: What do you do the first time when you do not have any prior weights/reforges/enchants/gems?

In my opinion simc would need to do its 50.000 iterations for each possible combination of gem/enchant/reforge and then compare the results. This would take ages of course to compute, but in my opinion is at present the only way of getting the optimum result with maximum confidence. And I believe this is not the way the sims work presently.

You would of course try to reduce the number of combinations possible by ruling out obvious nonsense like i.e. "spirit", but if you do not start off clean every time you will get a biased result. You are after all using results from a different gear setup which are not accurate anymore to generate the new one. At least that´s still my opinion at this point.

The "catch 22" of any simulator/spreadsheet's results are fairly well known. We can't get the best gear rankings without the weights, but we can't get the weights without the best gear.

Even with that slight problem, we do still get the best gear and stat weights eventually because people continuously iterate on the setups by starting out with any gear set and using the output weights for tweaking the next gear set. SimC gets updated quite often so chances are the generated weights will be good enough (~90% accurate+) for everyone but the most anal retentive barring major programming bugs.

If you're interesting in working on SimC, you're always welcome to contribute at Issues - simulationcraft - World of Warcraft DPS Simulator - Google Project Hosting. It would be more productive to post SimC ideas there rather than here on EJ unless someone makes an actual SimC thread here.

stRafaello · 09/15/12, 12:54 AM

Originally Posted by Belatar

@stRafaello:

Statistical significance - Wikipedia, the free encyclopedia
Statistical significance - Wikipedia, the free encyclopedia

By design, damage in WoW is largely RNG based, so there will never be empiric "proof" or "final" results. You can only try to maximize the confidence in your results by maximizing your sample size. There is no real "noise" in WoW given that you repeat tries (samples) with identical prerequisites under controlled conditions (which SimC or Excel sheets are able of doing).

SimC makes sure of all that while your individual runs are prone to error (buffs not present / forgotten, etc., errors in rotation) and actually have noise.
Regarding sample size, SimC does 50000 noise-free runs in about 1 minute... you´ll never even get close by doing this real-time.

Only advantage of doing it real-time in WoW would be, that you are using the actual engine vs. a "reverse-engineered" one. But this does not make up for the huge gap in significance.

What real-time testing IS good for is for double-checking mechanics, behaviour, etc. to validate your engine and base numbers or to double-check sim results whether they are within the predicted range of error.

If you are trying to compare your results to the simulated ones from SimC or Excels as equal, you are wasting your time.

This is basically how I've done the work posted:

Define a question
Gather information and resources (observe)
Form an explanatory hypothesis
Test the hypothesis by performing an experiment and collecting data in a reproducible manner
Analyze the data
Interpret the data and draw conclusions that serve as a starting point for new hypothesis
Publish results
Retest (frequently done by other scientists)

I haven't jumped any step in the way, and yet you're telling me that it's a waste of time to do observations because they're not 100000 iterations of hard data.

Scientific method - Wikipedia, the free encyclopedia

Belatar · 09/15/12, 3:50 AM

@stRafaello: That's what I am telling you still.

You are not deductively looking for a formula like e=mc2, you already have your formula(e) and are trying to weigh external RNG-influenced factors (stats). Thus you are dealing with statistics here and the more samples you get, the better the results.

stRafaello · 09/15/12, 7:21 AM

Originally Posted by Belatar

@stRafaello: That's what I am telling you still.

You are not deductively looking for a formula like e=mc2, you already have your formula(e) and are trying to weigh external RNG-influenced factors (stats).

Yeah, because the only reason someone would make use of the scientific method is if they're looking for a formula like e=mc²...

Originally Posted by Belatar

Thus you are dealing with statistics here and the more samples you get, the better the results.

Oh, you mean those 25.000 iterations samples I've already gotten?

I'll simplify:

Define a question: Are Mastery>haste builds stronger than haste on Tendons situations?

Gather information and resources (observe): Field-tested both builds a few times and made my calculations to gather good numbers for observation.

Form an explanatory hypothesis: [Ret 5.0 Pre-MoP] - Panda-ring to all players

Test the hypothesis by performing an experiment and collecting data in a reproducible manner: Did 25k iterations simulations of both builds ([Ret 5.0 Pre-MoP] - Panda-ring to all players)

Analyze the data: Haste (http://i.imgur.com/icRpV.png) and Mastery (http://i.imgur.com/ZMRu1.png)

Interpret the data and draw conclusions that serve as a starting point for new hypothesis: " Basically what I said before. Mastery builds are viable in these situations, but yeah, not really worth the effort of changing all the reforges." (See both previous links for the full conclusion).

Publish results: Loads of money, here I go!

Retest (frequently done by other scientists): Have got a few proof-reading of the results and feedback, but the conclusion still stands the same (for now).

• malthrin · 09/15/12, 9:45 AM

That's enough of the methodology argument. stRafaello, your SimC results were a good contribution. Thanks for posting them in full. Any further discussion on this topic should be about verifying that the configuration used is a good reflection of the game scenario under consideration.

The 'raid tests' were uselessly anecdotal. Your argument would have been more credible if you had not led with them.

A reminder to everyone: avoid excessive quote-splitting. If you find yourself wanting to respond to someone point-by-point, you're probably engaged in a stupid argument that doesn't contribute anything to the other readers.

Balhale · 09/15/12, 3:57 PM

Just found a major problem with your SimC results Rafaello. You are using a rather old build of SimC that still has the 2.1% conversion for mastery.

My 20 second sims with potions and guardian have haste well ahead of mastery. Partly this is the fix of mastery, partly the fact that guardian scales with haste and not mastery.

Links:
1) Using Creche, haste (h) vs mastery (m) vs haste str gems (hs) : Free Cloud Storage - MediaFire

Ranking: hs > h > m

2) Using Rotting, haste vs mastery vs haste str gems : Free Cloud Storage - MediaFire

Ranking: hs > h > m

3) Using Rotting, bloodlust and stormlash enabled, haste vs. mastery. vs. haste str gems : Free Cloud Storage - MediaFire

Ranking: h > hs> m

TheCrusader · 09/16/12, 3:48 AM

I have a question regarding Glyph of Mass Exorcism:
It says that Mass Exorcism will cause 25% dmg to all enemies within 8 yards of the main target. But will it deal 100% dmg to the main target ?

Hinkyz · 09/16/12, 5:49 AM

With the mass exorcism glyph, the tooltip changes to the following:

After a bit of testing with the dummies, I found that each individual hit has its own chance (based off of your crit chance) to crit, as shown here with one of the 25% hits critting:

Critting on the main target with the 100% hit will not affect the damage of the proceeding 25% hits, they will be based off of the non-crit amount:

TheCrusader · 09/17/12, 4:00 AM

Great!
Thanks

In that case I think, Glyph of Mass Exorcism will be one of my favorites.

aylen86 · 09/17/12, 5:52 AM

Yes, but of situational benefit. Losing the range of Exorcism is a straight DPS loss in situations without additional mobs on a regular basis.

Keep that in mind, changing talents is quite cheap in MoP.

Exemplar · 09/17/12, 10:01 AM

Hinkyz - I come up with Ret-Wrath of 13793 average damage and 280% Spellpower coefficient. I decided to do a quick napkin math - looks like the J are worth the usage for Inq (once Censure at 5, may get more from Exo on CD and J just to hit 3 HP/refresh Censure). More to bolster Exo/HoW/ES when used than for the crit, but the crit helps offset lost Wrath.

Rough examination shows that even accounting for crit rate/AW/uptime/etc, that Wrathspam ~= normal rotation. Not worth stealing the Symbiosis in general, but for some gimmick fight may be useful for a raid.

stRafaello - in the interest of improving any future testing: you cannot alter data. If a test has too much/little of X, you cannot tweak the data to remove/add any value of X. It is not sanitizing data, but polluting it. If you have too little/much of X, the only solution is to run more tests until X averages back out.

Belatar and Zephur - say you run stat weights. If stat A is 0.2 ahead of stat B with one setup, and you tweak things, rerun, and stat A is now 0.1 ahead of stat B - you wouldn't change your reforging. You're still going to try to reforge for as much A as possible (then as much B, etc). The only time it would matter is if you reforge for A, then suddenly B is better. You can catch those edge cases with Zephur's recommended reforge, run sim, update reforge (optionally run sim again). You don't have to explore the entire terrain of possible data.

Bottom line - precise values down to multiple decimal points generally don't mean much. Frankly all that's important for reforging is an order of stats - A > B > C.

Balhale - are you sure ES should be higher priority than generators? Delaying a short CD by a single GCD is a lot greater loss than delaying a minute CD by a single GCD. Example: losing 1/3 of a CS across an entire fight (and thus 1/9th of a TV), compared to losing 1/40th of ES. ES would need to hit 14 times harder than CS to be worth the loss (not counting the TV ripple) - I'm not seeing that's the case.