Cataclysm Mage Simulators and Formulators - Page 18

Light4 · 03/23/11, 2:44 PM

Have you checked that SimC is using Molten armor all the time? At least for my toon, it switches to Mage when low on mana and the resulting loss of 3% crit can explain the sub-100% crit of DF and FFB.

DrRumpus · 03/23/11, 2:49 PM

Er, ok now I get it... from playing frost on live I very rarely needed to switch armors, usually 1-tick evocation was enough. Thanks for the answer.

revulva · 03/28/11, 11:19 AM

I read some discussions here and on mmo-champions about frost mages, and I started using SimC to try to achieve some good stat weights to find a BiS set. Frost has an interesting stat priority since the crit cap for a few spells is actually achievable through shatter - and the 2 piece t11 bonus creates an extra gearing break point. After hit rating, we are trying to follow a priority like this: "crit to 2p lance cap > haste > crit to DF/FFB cap >= mastery > more crit".

After a couple of iterations of testing out different BiS sets, I ended up with stat weights like in this link:
Frost Mage - Mr. Robot - World of Warcraft
Here is the set again if you want to import into SimC:
chardev 8 - WoW Cataclysm

This set of gear actually performs significantly better than the current default SimC BiS profile, as it should since it follows the latest theory on gearing more closely. This holds true for the PTR and Live. You can see results here (Kuni's Frost Mage PVE Guide) that Kuni Zyrekai on the mmo-champions forum made. Read up a couple of posts on that thread if you want to see the full discussion.

I am consistently able to use SimC along with an iterative process to find stat weights which Mr. Robot's algorithm is then able to use to find BiS or very, very near BiS sets of gear - often over the full range of currently available 346+ gear. I have shown this process in detail for shadow priests, elemental shamans, and affliction warlocks on our site - and I have a similar write-up forthcoming for mages and the other two warlock specs. I know Mr. Robot is not a "simulator" or a "formulator", but, it is turning out to be an extremely viable companion tool for finding BiS gear. We've been in some dialogue with the SimC developers to see if we can implement cross-integration of the two tools... hopefully we can figure it out sometime soon.

The results I posted up for other classes can be found on our home page if you click on the link in my sig below.

Please note that our site's default stat weights for mages are not yet updated with my latest findings. We hope to have those up relatively soon - you'll have to use the link I provided above and enter in those weights manually if you want to optimize your gear with these better frost stat weights for the time being.

dedmonwakeen · 03/28/11, 1:16 PM

Originally Posted by revulva

Here is the set again if you want to import into SimC:
chardev 8 - WoW Cataclysm

Maintaining BiS sets is the bane of our (SimC developers) existence.

Frost and Frostfire profiles updated using this chardev profile as a base. (r8245)

Nathyiel · 03/29/11, 8:40 AM

With the hit cap, we can gain another 103 dps. We have to change boot's enchant from +35 mastery/run speed to +50 hit rating.

the new chardev: chardev 8 - WoW Cataclysm.

And for information, if you switch from Theralion's mirror to Bell, it's a loss of 186 dps without reforge and 130 dps with haste or mastery reforged.

(SimC 406.16 / 25k interations)

revulva · 03/29/11, 11:55 AM

Originally Posted by Nathyiel

With the hit cap, we can gain another 103 dps. We have to change boot's enchant from +35 mastery/run speed to +50 hit rating.

the new chardev: chardev 8 - WoW Cataclysm.

And for information, if you switch from Theralion's mirror to Bell, it's a loss of 186 dps without reforge and 130 dps with haste or mastery reforged.

(SimC 406.16 / 25k interations)

I was under the impression that having run speed is always a DPS gain in most of the fights in the game right now, although it isn't actually calculated in SimC. That is why I always go with lavawalker for a realistic recommended BiS setup for any mage.

dirby · 03/29/11, 12:48 PM

Taken from the Fire Compendium:

Time spent moving is effectively lost or lower DPS. Although you can still cast Scorch while moving, it's still lower DPS then a Burn rotation. Runspeed enchants therefore can be considered a DPS enchant for many encounters, and a strong one at that, with the actual value varying depending on the unique movement requirements of each encounter. In a Patchwerk style fight, however, where there's no movement at all - there are 50 Hit / Mastery enchants available which would strictly be superior for DPS. Generally speaking however, the vast majority of encounters require sufficient movement to make Runspeed the best enchant for boots by far.

dedmonwakeen · 03/29/11, 1:20 PM

In SimC one can create movement raid events based upon distance as opposed to time. In addition the core architecture supports a composite_movement_speed() API at the player_t level. However, the individual class abilities and enchants have not been implemented yet. (Each of the classes needs to override composite_movement_speed() appropriately.) With the hard work done, the remaining coding is mostly mechanical. It is just a manner of too much work and not enough hands. If somebody is particularly interested, please drop me a line.

Nathyiel · 03/30/11, 3:54 AM

I have made a test to know the precision of the DPS with 10k iterations.

For 100 samples of the same BiS:

DpS minimal: 28714
DpS maximal: 28745
DpS: 2730.69
Spead (max-min): 31
Range (max-min)/2: 15.5
Range%: 0.11%

In conclusion, we can't compare results with a difference lower than 30 dps.

dedmonwakeen · 03/30/11, 8:15 AM

There is a new section in the report labeled Statistics and Data Analysis. It generates tons of statistical information, providing iteration targets to reach certain levels of precision, etc.

Nathyiel · 03/30/11, 10:45 AM

Originally Posted by dedmonwakeen

There is a new section in the report labeled Statistics and Data Analysis. It generates tons of statistical information, providing iteration targets to reach certain levels of precision, etc.

It's this new function that give this idea. This statistical informations is about the simulations on him-self. It's quite useful to see where is the limits, what data can use, etc. It tell the precision about the simulation.

But what is the precision between 2 simulations or more? Is a comparison between 26500 and 26550 show a dps upgrade or just RGN? So I have run a test that show that I was on a precision of +/-15.

Or maybe, this information is already generate but under a label of not comprehensible for those (like me) who aren't familiar with statistic analysis.

revulva · 03/30/11, 10:58 AM

We updated the code on our site to take into account the 2 piece t11 bonus for frost mages. What it will do is actually bump up the weight on crit before the ice lance crit cap, then drop it down once you hit it. So, to use the site, you want to enter in the weight for crit as if you are ignoring that set bonus - it will do it in the background. So, here are the new results:
Frost Mage - Mr. Robot - World of Warcraft
chardev 8 - WoW Cataclysm

This set is actually even better than the last one we found, at least with SimC's default frost rotation.

One could make an argument that locking in Theralion's Mirror instead of Bell of Enraging Resonance slightly increases DPS... but when I tested with the default rotation the increase was too slight to overcome the statistical differences in 10k iterations. Because of the interesting interaction with the crit caps that frost mages have, I think we can call the Bell and Mirror statistically equivalent in the context of the entire set.

I am going to update our site with the stat weights that create this gear set next time I am able to push new weights to the server.

Gavinas · 03/30/11, 11:34 AM

Originally Posted by Nathyiel

It's this new function that give this idea. This statistical informations is about the simulations on him-self. It's quite useful to see where is the limits, what data can use, etc. It tell the precision about the simulation.

But what is the precision between 2 simulations or more? Is a comparison between 26500 and 26550 show a dps upgrade or just RGN? So I have run a test that show that I was on a precision of +/-15.

Or maybe, this information is already generate but under a label of not comprehensible for those (like me) who aren't familiar with statistic analysis.

That information is all already in the report, the lines you're interested in are the confidence intervals.

Ocrist · 04/03/11, 5:58 PM

I've noticed that the SimulationCraft action priority list for arcane mages has changed over the last month or so.

Out has gone the lines:

actions+=/arcane_blast,if=buff.clearcasting.react&buff.arcane_blast.stack>=2
ie react to clearcasting procs to get a free AB3 or AB4 cast

and

actions+=/arcane_blast,if=mana_pct>94
ie even if you have reached the AB stack where you would normally clear and if you've been lucky and are still at almost full mana, squeeze out another AB (you could also find the approriate mana percentages when various int procs are up).

In has come the line:

actions+=/sequence,name=conserve:arcane_blast:arcane_blast:arcane_blast:arcane_blast,if=!buff.b loodlust.up

which - correct me if I am wrong - is telling the sim to perform a simple rotation (AB4 or AB3 in BL).

I find these changes difficult to fathom and rather depressing. I was under the impression that good players would make such fluid decisions regarding clearcasting and mana levels - as Lhivera recently said in the forums:

"One of the reasons that the simulator does better than humans is that it doesn't use a conservation cycle - it makes a cast-by-cast decision about whether or not it can afford to cast another Blast."

Obviously this is no longer the case. I know that the idea behind the simulator is simply to test such scenarios and come up with the pragmatic answer as to what optimal play genuinely is. I tried putting these lines back into the action priority list and - low and behold - dps went down. I just don't know how clearing your AB4 stack even though clearcasting is up is ever going to be better than getting out a free AB4. Similarly, as is quite often the case, it is quite possible to get to AB4 and through luck be at a mana level where even if you are in a conserve phase, you can get out a fifth or sixth AB and still remain at 90% mana (or the equivalent percentage after calculating int procs).

I really do hope that this is the case. I would be dismayed if such "good player" decisions were not rewarded at all, and that a dull AB4 rotation was definitely optimal.

dedmonwakeen · 04/04/11, 12:55 PM

Looking over the svn history, I see Silverwind has been tweaking the Arcane action lists so hopefully he can comment. It may sound counter-intuitive, but the implementation of sequences in SimC is very painful, often with unexpected runtime/functional behavior. Usually, sequences are simply place-holders (or experiments) on the step to simple priority lists, however sometimes the translation is non-trivial.