[BrM/All Tanks] Theck-Meloree Index (TMI)

[Geodew]

Only partly. His own post calls it a smoothness metric. Smoothing out damage is the whole point of EH decisions. EH is not the same as actual damage reduction, it's a way to add a buffer to account for reaction time and RNG events. Unless the whole point of this is to provide a performance metric for tanks from log parses, which could be interesting to see, accounting for the ability rotation isn't relevant. Tank rotations are a priority based on two things, planned spikes and unexpected emergencies. If all he wants to do is model smoothing of spikes and assign an index value, then he's really not doing much more than over-complicating an EH calculation. He even mentioned near the end of needing assign a standard boss model and that modeling one type verse another type can greatly skew this value.

As I said, I don't know much about EH. You should bring this up with him. You don't need to register with anything even; you can comment as a Guest.

[Felade]

Basically, what you're dealing with here is that Tanks and Healers don't like to have things boiled down to a stat weight, and would rather be touchy-feely about their stats a little more than DPS are. Which, for the most part works, because every tank and healer combo is a little different (the players, if not the classes) so its a broad stat weight isn't as useful to tanks and healers because there are a lot of variables.

It is helpful to be able to say which stat is better. It isn't quite as helpful always to say how MUCH better it is (for tanks and healers, anyway), or only in narrow (like Ascension and Haste) cases.

The issue here is that for DPS, its a general model (which is what a stat weight really is) is fine, because the DPS community has accepted (or may not realize) that their stat weights change from fight to fight and on factors like movement, because the metric they care the most about (raw single target DPS over a long period of time without interruption) is actually pretty easily modeled. The Tank and Healer communities haven't really accepted stat weights because there's no single, easily modeled case. It is always going to be "it depends" somewhat with tanks and healers.

That doesn't mean that Theck doesn't know what he's talking about, or that he doesn't have good conclusions. Just that stat weights might not be the best way to implement his ideas.

[Theck]

Only partly. His own post calls it a smoothness metric. Smoothing out damage is the whole point of EH decisions. EH is not the same as actual damage reduction, it's a way to add a buffer to account for reaction time and RNG events. Unless the whole point of this is to provide a performance metric for tanks from log parses, which could be interesting to see, accounting for the ability rotation isn't relevant. Tank rotations are a priority based on two things, planned spikes and unexpected emergencies. If all he wants to do is model smoothing of spikes and assign an index value, then he's really not doing much more than over-complicating an EH calculation. He even mentioned near the end of needing assign a standard boss model and that modeling one type verse another type can greatly skew this value.

I'd argue that smoothing damage was one goal of EH decisions, but EH itself is a very limited metric. By its definition, it can only account for static, guaranteed factors: stamina, armor, etc. It doesn't account for changes in rotation or random events (i.e. avoidance). And more importantly, it is not truly a "smoothing metric" because it does not actually measure smoothness.

It measures how much damage you can take before dying without external heals. It can give you a rough approximation of TTL against a given boss, and it can give you a rough estimate of how long your healers have to react. And you can infer that the tank with higher EH can survive larger spikes. But none of those give you a good, quantifiable measurement of how smooth your damage intake is.

TMI is an attempt to do exactly that. The point you raised about a "performance metric for tanks from log parses" is exactly what I was aiming for with this idea. The concept is that having a high static EH value is far less important than the smoothness/spikiness of your incoming damage. In other words, it's better to be the tank that has 600k HP but takes damage in very consistent 300k chunks than the tank that has 1M HP but takes damage in chunks that vary from 100k to 500k.

It's also sensitive to pretty much every real-world factor that EH simply isn't. Running a poor rotation? Now we can quantify that effect. Avoidance and blocks? We can quantify that too. Want to know whether haste or mastery has a larger effect on damage smoothness? Now you can calculate it. You really can't do any of those things to any degree of accuracy with EH.

[Dirtdogs]

@Theck... Math overrated Play for the funs >>

[Geodew]

@Theck... Math overrated Play for the funs >>

Kicking butt is fun. Numbers facilitate that.

[Deleted]

Sounds interesting, hope you find more competent support for the BM module

[Aceshigh]

@Theck... Math overrated Play for the funs >>

Not sure if familiar with Theck, but MATH IS HIS DRUG!!!! haha

[Deleted]

Math is a very important part of WoW. If everybody thought that tanks and healers just try to find some feel-good-state then tank and healer theorycrafting would be obsolete. But it isn't. That is because even for healers and tanks there are certain things that can be determined with pure numbers, ie. haste breakpoints, resource generation...

TMI can also be a very helpful tool for brewmasters. Not to show some obvious things that theorycrafting already brought to light, but to find the smaller differences in a simulation based on the one char that you want to see. The brewmaster module needs a lot of work until we reach that point, but I am sure that this is just a matter of time.

[Hinalover]

TMI can also be a very helpful tool for brewmasters. Not to show some obvious things that theorycrafting already brought to light, but to find the smaller differences in a simulation based on the one char that you want to see. The brewmaster module needs a lot of work until we reach that point, but I am sure that this is just a matter of time.

Well Alysmera just started updating SimC with 5.4 Windwalker information and I've been trying to help him out whenever possible. I'm sure once we have the Windwalker system set up we can take a look at Brewmaster and get that working. I know from looking over the code that the Brewmaster information is in there, just needs to be cleaned up and added to the TMI system properly. But that could still take a while.

[Mesima]

Somebody buy Theck a beer or two! Great work!

[adalovelace]

I'm copying this from what I wrote in response to someone in the main thread:

Off the top of my head, there are a few big things that need to be fixed:

Elusive Brew: It looks like the module tracks the procced stacks correctly. I had it set up to to use EB any time you're at 6 or more stacks and it seemed to do this correctly. The problem is that it appears to only give you one second of the buff no matter how many stacks you have when you use EB. This craters the value of crit and other stats for TMI. (and gives you a low active EB up-time)
Self-healing: Expel Harm and other sources of self healing don't work. I don't remember if Chi Wave did or not. Also, no healing from orbs, though that's a little trickier on how you'd want to handle it.

There were also issues with different set bonuses not being implemented. There are probably some other things that are off, but those were the ones I noticed.

Theck can correct me, but my understanding is that the TMI calculation has nothing to do with the class modules directly, just the the damage and healing events. As long as those are calculated and reported correctly, then the TMI calculations should be correct, too.

[lairpie]

brm tmi


Other tanks are trying to determine how to best smooth their damage so they don't die. We know how, like, 0 argument. If your goal is to never take bursts of damage, you cap hit, cap exp, stack mastery to infinity. End of discussion, I didn't even have to do any math. Gearing for brms isn't about deciding how to reduce your burst damage taken the most, it's deciding how much you care about damage done and overall damage mitigation compared to your TMI. Better modeling of that would be good for some things like, given a fixed TMI i want, how does dual wielding, using the multi-strike trinket, or shifting mastery to crit affect my damage done and my overall healing needed?

I'm pretty sure I could build this all easier not in simcraft than in simcraft, the issue is mostly that it will be more useful to more people if I do it inside of simcraft, and more likely to continue functioning when they change spells and/or I quit the game someday. One of the big issues is that BrM actions are very dependent on your current health, which isn't really modeled in the TMI stuff i've seen. You have to have a realistic model of healing and damage taken by the monk to model the monk's actions. In simcraft right now, even if you fix up the action list, and even if the spells were all working properly, it shows i only jab like 2 times per fight and keg smash once or twice because I'm below 35% hp for all but the first couple seconds and at negative 948209584295 hp I use every bit of energy on expel harm (I put basically to replace jab with expel harm if i'm below 70% hp, replace anything other than keg smash with EH if i'm below 50, and prioritize all energy for EH if i'm below 30% hp) because i'm not ever above my hp thresholds.

So barring the desire to setup a huge complicated healing model with multiple actors and such, I was going to basically create fake Monk spells that have no cost, and are executed completely separately, but simultaneously to the acutal monk rotations. Things like "lifebloom" which would simply cast a small heal on me every 1s regardless of anything. 2 druids in your raid? put the line to execute that twice. Then "renew/rejuv" a spell that gets cast on you for a slightly bigger heal every 3 s. Probably "beacon of light" which would be a fairly random amount of healing that doesn't react at all to your hp. Then "lay on hands" which is essentially coded as a monk ability that gets executed separately from any others but whenever your hp drops below 25%, if you don't have guard up, it hits you with huge heal 1s later (reaction time). Basically, I'd try to model a variety of healing spells, some that intelligently get cast on you, some that don't, and see if I can build a model that remotely resembles the actual healing i get.



Much more than TMI between classes information, I would want to use simcraft to answer things like:

• if i drop 10k mastery for 10k crit, what is the real difference in outcome there? Do we kill the boss 1 second faster but i have a 5% greater chance to die? that's a pretty crappy outcome for my guild and I wouldn't be suprised if that's the answer.
• Does some world first guild pick up an enrage beating amount of damage by accepting a 50% chance the tank will die before the enrage? yes, the boss died, they can pull a few times until the tank gets enough CD/ ExpHarm using skill/ luck to not die that time.
• raid dps difference of giving me a weapon vs that weapon going to a rogue? Survival difference, healing output difference?
• On a fight where it doesn't make guard line up with some specific boss ability, is that CD Reduction trinket worth much of anything? Like sure, guard, fb, and zen med more often are nice, but what % of the time do they actually matter?
• I'd also like to be able to model in specific boss ability timing and specific player CD timing to be able to compare especially DH and DM.

These are the sorts of things dps specs can answer with simcraft and we usually guess and since no one can prove us wrong, we all get to be happy with our own guesses as being awesome and we pretend all the differences in answers are because of differences in our raid or strat or playstyle. Sometimes that's true, and sometimes there's just no way to show it's BS.

(Plus I'd actually know how stagger calculates, not just how the ability text says it calculates)

[Geodew]

Other tanks are trying to determine how to best smooth their damage so they don't die. We know how, like, 0 argument. If your goal is to never take bursts of damage, you cap hit, cap exp, stack mastery to infinity. End of discussion, I didn't even have to do any math.

How do you decide whether to match socket bonuses for stamina or not without mathematics? Guessing?

I'm pretty sure I could build this all easier not in simcraft than in simcraft, the issue is mostly that it will be more useful to more people if I do it inside of simcraft, and more likely to continue functioning when they change spells and/or I quit the game someday. One of the big issues is that BrM actions are very dependent on your current health, which isn't really modeled in the TMI stuff i've seen. You have to have a realistic model of healing and damage taken by the monk to model the monk's actions. In simcraft right now, even if you fix up the action list, and even if the spells were all working properly, it shows i only jab like 2 times per fight and keg smash once or twice because I'm below 35% hp for all but the first couple seconds and at negative 948209584295 hp I use every bit of energy on expel harm (I put basically to replace jab with expel harm if i'm below 70% hp, replace anything other than keg smash with EH if i'm below 50, and prioritize all energy for EH if i'm below 30% hp) because i'm not ever above my hp thresholds.

That's because you need to use the "damage taken in last x milliseconds" argument, not current HP. Current HP just returns garbage without healer actors, as you saw.

The Blood DK module is in a bit better shape. If you want to model EH's cooldown removal effectively (enough to simulate real scenarios of how much damage you'd have to take in a certain time span to drop that low), I'd find out how they're modeling Will of the Necropolis.

[kaiadam]

How do you decide whether to match socket bonuses for stamina or not without mathematics? Guessing?

A) don't have gear with stam socket bonuses.
B) get the bonus with hybrid accuracy gems.

[lairpie]

How do you decide whether to match socket bonuses for stamina or not without mathematics? Guessing?

They're always worth matching unless someone types spirit or int into an agi item as the bonus by mistake.

There's been extensive math done on brm stat weights focusing on reducing damage taken by venyasure, mostly on the EJ forums. The general conclusion is that for a balance of damage done, damage taken, burst damage reduction (loosely TMI) that there isn't a socket bonus imagineable that isn't worth matching. If you're not matching every bonus, it's because you so strictly are pushing for damage that you want pure dps and are going to ignore any bonus that isn't damage. Deciding bonuses isn't done without math. it's that the math is so simple it doesn't matter. The best estimates I've seen put haste, crit, dodge, parry all around the .6 agi range, with master basically anywhere between .1 or so if you're doing 10 man normal in heroic gear, to nearly infinite depending how you value burst damage reduction compared to damage reduction compared to damage done. How you value TMI compared to dtps compared to dps is pretty subjective, varied from fight to fight, raid to raid, and that's why people didn't have too much interest in a brm tmi value. Also, the fact that knowing might get us nerfed.

Why I care is that regardless of the validity of cross class tmi with weird standards on cd usage and reactive ability usage and stuff, is that for a given set of gear, i should be able to know, if i play perfectly, or if i play with x amount of lag, how does changing 1 thing affect my dps, dtps, and ~TMI. I feel this would be useful to me, and likely useful to others. The biggest thing is i would like to know what different ways to trade TMI and or dtps result in the biggest increases in dps, along with having a way to model the benefits of monks getting gear vs other classes. That's not really directly related to TMI, but the work needed to figure them out is pretty overlapping. Plus i would get to respond to the 809242094234 "should i use a 2her or 2 1hers questions" and similar threads with a nice link to how to find your own answer out, rather than a lot of fuzzy inputs, followed by specific math, followed by fuzzy interpretation of the results.

That's because you need to use the "damage taken in last x milliseconds" argument, not current HP. Current HP just returns garbage without healer actors, as you saw.

The Blood DK module is in a bit better shape. If you want to model EH's cooldown removal effectively (enough to simulate real scenarios of how much damage you'd have to take in a certain time span to drop that low), I'd find out how they're modeling Will of the Necropolis.

I'll look into that, but to get the sort of "chance of death" difference vs "damage and healing output" differences that i want, and that i think other monks would find useful, i'm going to need to actually model the heals. I may just come up with a wa to match TMI vaguely accurately as a byproduct of that.

[SurrealNight]

[*]raid dps difference of giving me a weapon vs that weapon going to a rogue? Survival difference, healing output difference?

This is actually something I struggle a bit with deciding. Have a bear OT/rogues/enh shaman/kitty DPS on the raid team. I typically skip out on the 1st 2H drops for the droods but what about the fist weapon that drops earlier in the raid than the maces/swords? Part of the reason I've always been showing with DW setup (though I like it more anyways).