Guest Post: Yuki Learns You Stat Weighting (p3)June 25, 2009
As stated last week, here’s the post outlining how to determine weights to plug into lootrank to figure out your best in slot. It’s a general list written by (alas) a DPS-type, so it’s not aimed at healers specifically, but I hope you all can still find it useful. Yuki’s been my guildie, my buddy, and enabler of my anime addiction for several years now. He’s also the first person who told me that such a thing as mathcraft exists. Be warned, there’s numbers ahead.
Step 5: Final Complications
There are basically two things left that haven’t been modeled. The first is mana regen, which creates some difficulties, and the second is trinkets and the like with activated or proc effects, which presents others.
5a: Mana Regen
The main thing about mana regen is that it only matters if you run out of mana. This can make it hard to model, since it’s not something that easily converts directly into spellpower or attack power.
Probably the most common way of modeling mana regen is in terms of how many extra spells it lets you cast if you WERE to cast until you ran out. Basically, you figure out how long you can cast without running out of mana, and how much effective healing or damage you’d do on average. Then you see how much 1mp5 would let you cast. You compare how much a point of spell damage would have added vs. the amount the mp5 added to determine the effective spellpower value of a point of mp5. This is a popular method because it’s pretty straightforward, and you can modify it a little to include out of 5 second rule regen if you’d like.
Another method is to work backwards: essentially, you take a length of time longer than the amount of time it would take you to run out of mana chain casting and then calculate how much less time you have to spend on 5-second-rule regen to determine how many more spells you could cast, and use that to derive the relative value.
As you can probably guess, there really isn’t a great way to graft mp5 onto spellpower. Still, these two methods at least produce something, which can be modestly useful for modeling gear.
So, we’ve basically covered how to model static stats; what about ones that aren’t so static? Generally these are in the form of either active abilities (clicky trinkets, for instance) or passive procs (chance-on-hit/crit trinkets). How to model these?
Actives are actually pretty easy. You can basically model them as “permanent” power by dividing their effect by the amount of time it spends on cooldown. So, for instance, a trinket that provides 200 spellpower for 15 seconds on a 1 minute cooldown is effectively the same as one that grants 50 spellpower all the time. This isn’t wholly accurate (since you can stack some cooldowns to get more out of them, like an unholy DK blowing Gargoyle while her trinket buffs are up), but it’s good enough for government work.
Passives are trickier. You need to know the proc rate and internal cooldown, which are almost never listed; you’ll need to get them from datamining or testing. Use these to calculate the average uptime % (which is the proc duration divided by average time to proc + internal cooldown), then multiply the value of the proc by that. Since this requires you to use the average time to proc, which can vary wildly for low-chance trinkets, and it shares the flaw of not modeling stacked-burst very well with the active version, it’s not wholly accurate, but it gets the job done well enough to compare trinkets.
Step 6: Putting it all together
At this stage, you have enough information to generate final stat weights: for each overall stat, add up the weights you calculated for each of their effective ratings you calculated above to determine the final overall weight of that stat. So, for instance, the value of a point of int is:
Int ESP = Any Value From Talents + (Effective Crit Rating) * (Crit Rating ESP) + (Effective mp5 Rating) * (mp5 ESP)
I used Int on purpose, because an observant reader will note that it provides a benefit we haven’t modeled: an increase to your base mana pool. You’ll have to determine what weight (if any) you give to that similarly to how you chose to calculate mp5; I didn’t mention it there mostly so that I could point out that at this stage you should be on the look out for stat effects that you hadn’t considered yet!
Once you’ve done this for everything, you’ll have your final weights for stats as calculated against your baseline! Now you can use them to help consider gear . . . until you move too far away from your baseline and have to recalculate them.
I’ve always been fairly skeptical of stat weights as anything other than very general comparisons of how “strong” a given stat is. Having seen this demonstration of what hot dogs are made up, you can probably see why: there’s a lot of assumptions in the process, and the end result loses value considerably as you move away from your baseline. I tend to agree with Ghostcrawler’s assessment that if you’re just pulling “best” gear from somewhere based on this stuff, you’re not necessarily doing yourself a service as it won’t necessarily actually be “best” for you if your gear is different than the assumptions.
One thing this process will teach you, though, is the underlying mechanics of your class, including what scales with what and how things interact, which is probably the most important thing to learn for any form of theorycraft. The numbers generated may be of limited value, but the process of generating them has some fairly valuable lessons to be learned along the way.
Aaand done! You’ve made it! Special huge thanks to Yuki for putting this together and trying to learn us some math. Did it work for you?
Also, there’s not gonna be a Tuesday post for me next week. I’m currently in California visiting my family with the Tarsus; and my mom doesn’t really live in this century as far as internet goes. But I’ll catch you all the next week!