Cataclysm Mage Simulators and Formulators - Page 13

dedmonwakeen · 02/01/11, 5:16 PM

Originally Posted by Kavan

An additional complexity for fire when trying to use EP values even if you try to do some dynamic adjustments on them is that there are several discontinuities that might not even show in EP values depending on the granularity of differentials. So it's not really that the EP values are wrong, you just can't use them directly to get total dps (at least not in any easy way). If you do try to use EP values a slightly more correct way would be to know the stats that were used to obtain the EP values and then when you evaluate your gear compute the differentials from gear to the stats used for EP values and then compute linear combination on differentials only.

I toyed with making a Rawr.BlackBox module that took Stat+StatWeight pairs as input just so I could experiment with Rawr's optimizer. However, discretion seemed the better part of valor: It is never a good idea to poke a Bear with a stick.

SpartAdept · 02/01/11, 6:31 PM

Originally Posted by revulva

I did my analysis on the Fire Mage 372 BiS gear setup as an example. I used SimulationCraft build 403-29 on the Live version of the game for my DPS tests. [b]I know the patch is going to change some things as far as DPS numbers and BiS setups, but, for now I figured testing on Live is still the best for discussion purposes.

Everything I've done has been in consideration with the upcoming changes, I personally put all my thought and limited time into BiS Mage taking into account maybe 0.1% of the WoW Mage population will have / or has access to all of the current potential BiS gear prior to the patch. I don't see much purpose in figuring out information that will never be used, other than just satisfying thirst for knowledge.

That being said, just referring to the post Erdluf made, it does appear that this program is using fixed stat values, instead of dependency values (correct me if i'm wrong if this does), while one stat changes to the extreme (Example; 217 Mastery -> 217 Crit) this doesn't seem to effect the stat weights. Mathematically and logically it should, yes?

revulva · 02/01/11, 7:21 PM

We do something in between fixed stat values and dependency values. We take into account the large discontinuities found around soft and hard caps, diminishing returns on stats like parry/dodge, etc.

That being said, the website lets you put in whatever stat weights you want. The "default" weights are optimized around t11 heroic BiS gear. Since stats don't change all that much between 346 and 372 gear, we find that you still get good results for most gear with the "default" weights. If people want to, they can run SimC on their current character gear setup, get stat weights, then plug them in to the site to optimize gems, enchants, and reforges.

I must have misread your post on the BiS mage gear setup - It looked to me that you were proposing a "Best in Slot" setup. Most discussions of that sort are theoretical discussions on what the best gear would be. Regardless of the stat weights in use, or how I found the resulting set - I found a set that does more damage. I would consider that a "better" Best in Slot set.

SpartAdept · 02/01/11, 9:38 PM

Originally Posted by revulva

We do something in between fixed stat values and dependency values. We take into account the large discontinuities found around soft and hard caps, diminishing returns on stats like parry/dodge, etc.

That being said, the website lets you put in whatever stat weights you want. The "default" weights are optimized around t11 heroic BiS gear. Since stats don't change all that much between 346 and 372 gear, we find that you still get good results for most gear with the "default" weights. If people want to, they can run SimC on their current character gear setup, get stat weights, then plug them in to the site to optimize gems, enchants, and reforges.

I must have misread your post on the BiS mage gear setup - It looked to me that you were proposing a "Best in Slot" setup. Most discussions of that sort are theoretical discussions on what the best gear would be. Regardless of the stat weights in use, or how I found the resulting set - I found a set that does more damage. I would consider that a "better" Best in Slot set.

So, you're saying that the best in slot setup, isn't actually best in slot, because "stats don't change that much from 346 to 372" and you use an educated guess for a number in between 346 and 372 gear for stat weights, so no matter what.. its wrong? I must of misread your post. Most discussions on what the best gear would be, regardless of the stat weights in use, or how one comes to the resulting set is explained by math, or answers to equations, not assumptions that are completely silly.

Cause it looks to me your program is rating Stump of time over Heroic Bell. Even after hit cap and 300 hit being useless, plus that whole 252 extra spellpower proc thing as well, would love to know the math behind that.

Maybe your trinket proc's are off / not functioning properly?

Just a quick reminder, Bell will no longer require a crit to potentially proc, I assume since you're after the answer with current standards on Live, this could also pose a discrepancy between our two assumptions.

Muphrid · 02/02/11, 12:03 AM

If revulva's found a set that produces higher DPS according to simcraft (and is conclusively so--i.e. a statistically significant difference at some high degree of confidence), then it doesn't matter what his methodology was to find that set. The proposition can be tested directly to verify whether it is or isn't a better set.

What does matter is whether that technique is mathematically valid and general--if it will reliably optimize items and sets. That is a tougher task. I understand the impetus, but I'm concerned that the approach is not well-founded. What does this EP number derive from? A simple multiplication of stats by their associated (DPS-normalized) weights? Something more complex than that to account for the inherent nonlinearity in the problem?

revulva · 02/02/11, 1:40 PM

Originally Posted by SpartAdept

So, you're saying that the best in slot setup, isn't actually best in slot, because "stats don't change that much from 346 to 372" and you use an educated guess for a number in between 346 and 372 gear for stat weights, so no matter what.. its wrong? I must of misread your post. Most discussions on what the best gear would be, regardless of the stat weights in use, or how one comes to the resulting set is explained by math, or answers to equations, not assumptions that are completely silly.

Cause it looks to me your program is rating Stump of time over Heroic Bell. Even after hit cap and 300 hit being useless, plus that whole 252 extra spellpower proc thing as well, would love to know the math behind that.

Maybe your trinket proc's are off / not functioning properly?

Just a quick reminder, Bell will no longer require a crit to potentially proc, I assume since you're after the answer with current standards on Live, this could also pose a discrepancy between our two assumptions.

I found a set that simulates out to higher damage with SimC. It is a slightly more optimal set, based on all of the assumptions that have been made so far in this thread and in other threads on this forum with regards to fire mage theorycraft. You used stat breakdowns to find the original set - why is it invalid for me to use stat breakdowns to help me find a more optimal set?

The item lists on the Mr. Robot web site are ranked independent of any other gear you have on. There is a note that tells you that right on the list. I have a whole post that goes over the math of how we estimate the value of trinkets right here: How does Mr. Robot estimate trinkets?. Use your own stat breakdowns and you will see the sound math that let me to the idea to try different trinkets than you had suggested. If you don't agree with the math, that's ok - but it is math that has been suggested by many other people for estimating cataclysm trinkets.

I then ran a simulation. The simulation shows a DPS gain. You can't argue with that. You could argue with the accuracy of the simulator... but, you have to use SOMEthing to decide which sets are optimal.

Silverwind · 02/03/11, 12:06 PM

I did a short comparison of racial bonuses for the Arcane module, that might be interesting for EJ readers:

Gnome = 5% max mana = ~400 dps (simcraft run)
Draenei = 1% hit = 102.45 hit rating = 287 dps
Worgen = 1% crit = 179.34 crit rating = 203 dps
Orc = 584 sp / 15s / 2 min cd = 73 spell power = 176 dps
Troll = 2562.5 haste / 10s / 3 min cd = 142.36 haste rating = 172 dps
Goblin = 1% haste = 128.125 haste rating = 154 dps
Blood Elf = 6% mana restore / 2 min = ??? dps

No surprise here, Gnomes come out on top for Arcane and I suspect that Draeneis are the best race for Fire/Frost. The Blood Elf racial is missing because it's hard to estimate and simc currently doesnt use Arcane Torrent per default. Maybe someone can provide a rough estimate.

Zimeron · 02/03/11, 12:50 PM

Originally Posted by Silverwind

The Blood Elf racial is missing because it's hard to estimate and simc currently doesnt use Arcane Torrent per default. Maybe someone can provide a rough estimate.

If you're a Blood Elf it will automatically add it upon import. If you change the race without regenerating the profile, it won't use it without adding it to the action list. Simply add /arcane_torrent to the action list or run with default_actions=1 and it'll use it as expected .

Nathyiel · 02/04/11, 2:53 AM

For Lhivera, do you store your result each time you make a new set of test? Or do you only keep your last best result? And why did you made test for Troll only for Frost?
And another question, what is the use of the "mana_pct" action ?

Ektoplasme · 02/04/11, 4:14 AM

Originally Posted by Silverwind

Draenei = 1% hit = 102.45 hit rating = 287 dps

You're overevaluating the Draenei racial. All characters in BiS 372 will be hitcapped so the draenei racial doesn't add 102 hit rating but 102 points of whatever rating you reforged your excess hit into. For arcane, best case scenario is when you managed to reforge this 102 rating into 102 mastery rating and worst case 102 crit rating.

This makes the value of the draenei racial somewhere between 112 DPS and 136 DPS. The draenei racial is thus one of the worst racials for arcane.

Caerdor · 02/04/11, 6:48 AM

The troll and orc racials are undervalued if calculated as an average uptime because they can generally be used alongside other strong cooldowns. For example the orc racial can be combined with every burn phase and the troll racial with every second arcane power and ideally heroism/pot/mana gem etc. It obviously depends on the gear used and the fight they are applied to but considering how strong the arcane cooldowns are it wouldn't surprise me if they proved to be well ahead of the worgen racial in most cases.

Maje · 02/04/11, 11:02 AM

I've been playing about with tracking Ignite damage on a target, and one thing is obvious blizzard has done some extreme tinkering. As is known the debuff is supposed to have a 4s duration however for whatever reason (some speculate to reduce the burst) blizzard decided to increase the duration to 6s in some cases (refreshes).

What is interesting however is how it's implemented on the UI side; calling UnitAura when Ignite is just applied returns the expected 4s duration however when Ignite is refreshed (for the first time) the duration isn't infact updated to 6s flat but rather to 4+2-(time that passed from the application).

As noted before some criticals are outright ignored (rarely) others change the damage in an awkward way even considering the new duration.

For instanse look at the following example of the output:

[17:44:50] igniter: ( 4270 , 5.907 , 3.9890000000014 )
[17:44:52] igniter: ( ~4270 , 5.907 , 1.9979999999996 )
[17:44:52] igniter: ( 4269 , 5.907 , 1.982 )
[17:44:54] igniter: ( ~4269 , 5.907 , 0 )
[17:44:54] igniter: ( 4269 , 5.907 , -0.014999999999418 )
[17:44:54] igniter: ( 0 , 0 , 0 )

What the addon does is record the last Ignite tick from the combat log and prints out the following (amount, total duration, time left) the (0, 0, 0) line means SPELL_AURA_REMOVED and a ~ prefixing the amount means it is from SPELL_AURA_REFRESH.

Now the above output doesn't have the criticals that actually procced it but they are of no consequence at the moment, note that the amount remains the same during all of the ticks despite the two _AURA_REFRESH events in between.

The full output is here:

[17:44:30] igniter: ( 0 , 0 , 0 )
[17:44:34] igniter: ( 0 , 4 , 4 )
[17:44:36] igniter: ( ~0 , 4 , 1.9890000000014 )
[17:44:36] igniter: ( 3229 , 4 , 1.614999999998 )
[17:44:38] igniter: ( ~3229 , 3.987 , 3.987000000001 )
[17:44:38] igniter: ( ~~3229 , 3.987 , 3.982 )
[17:44:38] igniter: ( 1568 , 3.987 , 3.982 )
[17:44:40] igniter: ( ~1568 , 3.987 , 1.9559999999983 )
[17:44:40] igniter: ( 1568 , 3.987 , 1.7189999999973 )
[17:44:41] igniter: ( ~1568 , 4.735 , 4.7350000000006 )
[17:44:42] igniter: ( ~~1568 , 4.735 , 3.9979999999996 )
[17:44:42] igniter: ( 6169 , 4.735 , 3.9799999999996 )
[17:44:44] igniter: ( ~6169 , 4.735 , 1.9979999999996 )
[17:44:44] igniter: ( 6169 , 4.735 , 1.9799999999996 )
[17:44:45] igniter: ( ~6169 , 4.708 , 4.7079999999987 )
[17:44:46] igniter: ( ~~6169 , 4.708 , 3.9969999999994 )
[17:44:46] igniter: ( 3796 , 4.708 , 3.9959999999992 )
[17:44:48] igniter: ( ~3796 , 4.708 , 1.9989999999998 )
[17:44:48] igniter: ( 3795 , 4.708 , 1.9979999999996 )
[17:44:48] igniter: ( ~3795 , 5.907 , 5.9069999999992 )
[17:44:50] igniter: ( ~~3795 , 5.907 , 3.9980000000032 )
[17:44:50] igniter: ( 4270 , 5.907 , 3.9890000000014 )
[17:44:52] igniter: ( ~4270 , 5.907 , 1.9979999999996 )
[17:44:52] igniter: ( 4269 , 5.907 , 1.982 )
[17:44:54] igniter: ( ~4269 , 5.907 , 0 )
[17:44:54] igniter: ( 4269 , 5.907 , -0.014999999999418 )
[17:44:54] igniter: ( 0 , 0 , 0 )

As an aside the addon was written to actually try and ease of when to use Combustion however the output I'm seeing doesn't make much sense to me in terms of predicting the amount of damage Ignite is going to do, it's porbably more useful to just do it blindly after good crits and hope.

Muphrid · 02/04/11, 12:46 PM

It seems like in the cases where the Ignite gains time (is refreshed) with no additional damage/tick like it could be a bug in Blizzard's new DoT clipping mechanics--rather than adding damage to the ignite, the damage is just being lost as if Ignite is a run-of-the-mill DoT and the leftover duration before the next tick is being added to the 4 second standard.

Silverwind · 02/04/11, 2:13 PM

I've commited some fixes for the racials to the Arcane module. Here's the updated racial ranking using Undead 372 PTR as a base:

Gnome = 5% max mana = ~400 dps (simcraft run)
Orc = 584 sp / 15s / 2 min cd = ~ 300 dps (simcraft run)
Draenei = 1% hit = 102.45 hit rating = 290 dps
Troll = 2562.5 haste / 10s / 3 min cd = ~220 dps (simcraft run)
Worgen = 1% crit = 179.34 crit rating = 193 dps
Goblin = 1% haste = 128.125 haste rating = 148 dps
Blood Elf = 6% mana restore / 2 min = ~140 dps (simcraft run)

Orc comes out pretty high when using Blood Fury during each burn phase. Troll uses Berserking every second burn. Blood Elf uses Arcane Torrent when it drops below 75% mana, but the dps gain seems a bit disappointing. The other 3 races are calculated purely through the scale factors.

Maje · 02/04/11, 2:36 PM

Your non-sim runned numbers are wrong, espesially Goblin and Draenei, Goblin because 1% haste is not 128.125 since haste scale multiplicatively if the mage had 10% haste then 1% would translate into 140.9375 haste rating, Draenei because if you aren't you aren't losing 1% hit but rather 102.45 rating points of your least valued stat since that's what you'll probably reforge to hit.