Cloudburst Question re: Mastery

[Lowpro 0]

Not sure if this has been asked before but I asked on the WoW forums and got no answer so I'll ask here.

 

When Cloudburst totem is dropped it essentially stores up heals done and releases them as a smart heal to allies.  My question is whether the heals it stores are the heals attenuated with mastery (ie Heal received instead of the general SP heal).  The other part is whether the Burst part of the Cloudburst heal itself attenuates with Mastery as well.  

[Stoove 347]

Cloudburst stores healing based on all the healing you do, Mastery included. Cloudburst's heal does not scale with Mastery (i.e. it doesn't double-dip).

 

I hope that helps!

 

ps: I think you mean "scales" rather than "attenuates" - "attenuate" means "to get smaller".

[Lowpro 0]

First off: The editor for this site...not a fan

 

Anyways CB versus High Tides, from my general playstyle I much prefer CB but I'm guessing once you can really support the mana cost of Chain Heal High Tides would have to pull ahead.  

 

A serious question: How do people model healing for fights.  Recount and Skada do not give reliable representations of efficacy in a fight because no one cares how high your HPS is, only whether the healing keeps in tempo with a fight (ie the burst healing comes when needed, the raid healing comes when needed) normalizing that in an HPS makes little sense.

Edited January 12, 2015 by Lowpro

[Pandacho 286]

 How do people model healing for fights.

What did you mean here?

How theorycrafters are modeling the healing for their healing efficiency theories?

How RLs / healing coordinators know to divide raid CDs or healing priorities?

How healers know what spells to use?

 

About theorycrafters Stoove can explain much better than myself

 

RLs / Healing coordinators (who cares about team healing) have to know strengths and weaknesses of healing classes that are present in their raid group - who is better in AoE healing / tank healing / direct targeting / shielding / raid CDs - and to apply this knowledge to specific fights and bosses mechanics.

 

Raid healers - well, you have to know your class, bosses mechanics and damaging abilities timers and to plan in advance what you should do in every situation.

In addition to this, I personally go through my logs rankings for every boss and pick there fights that have 90+ percentile (or at least 80+ if it was my maximum). I go through talents, glyphs, spell usage and amount of casts for every spell in these fights, write this down and check deliberately in the next raid. Then play with the results until I get stable percentile - if it remains on 80+ - it's right set of talents and spells for me.

Edited January 12, 2015 by Pandacho

[Lowpro 0]

I mean how do they model a scenario that's not conducive to normalizing.  If there's a fight that lasts 6 minutes that ends if the tanks die but the tanks risk of dying is event based on damage taken then you'd want some sort of survival model.  Depending on the fight, an overall HPS readout and in fact "healing loss" in a rotation is nonsensical.  If you wanted to make judgments on a healer's performance in a fight to see how they could improve has a lot of nuances.  

 

Maybe on this forum this is recognized but I've seen a lot of people on the WoW forums who are concerned with their overall throughput but what should really matter is their throughput relative to the tank's risk of dying (thus a wipe).  Resto Shaman's mastery literally normalizes that though; it increases as the tank's risk of dying increases.  I'm just wondering if I wanted to start up R right now an make a survival simulation to measure healing performance, how would I make that model accurately.  How could I use that to compare a fight to that model to see where I diverged from that model to make a wipe happen basically.  

Edited January 12, 2015 by Lowpro

[Stoove 347]

So, I find it interesting that you're talking about the tank healing here. I'm quite well convinced that tank survival is most strongly affected by the tank behaviour (appropriate use of active mitigation) than on healer behaviour.

 

If you wanted to make judgments on a healer's performance in a fight to see how they could improve has a lot of nuances.

 

It sure does! That's one of the interesting sources of depth that the raiding game is all about. It's also why this forum is so popular for discussing healing combat logs. There is a lot of context required when interpreting how any healer is doing, and you're right that raw HPS isn't as helpful as many people think. However it's definitely a clear signal if something's wrong. Additionally, in the current Heroic/Mythic healing environment, there's not much room for niche healers (i.e. someone who cares solely about the tanks).

 

I don't think it's really reasonable to expect that a simulation of tank survival will give you a good idea of overall healing performance, because it's a little too niche. 

 

Having said that, it *is* interesting and I'd like to see some quantitative results from it. If I were you, I'd start by trying to learn how to model tank damage and mitigation first. If you can make a semi-realistic tank work for your simulation, then you can implement a simple set of rules which can dictate the rest of your simulation.

 

Here is how I would make it work; your tank is taking significant damage periodically and is well modeled, and the rest of your raid is spread out and moving around a bit. There is moderate AoE damage happening on them throughout the fight in short bursts (like Imperator Mar'gok). You have two "wipe" conditions; 1) more than four non-tank players die during the fight, 2) either tank dies. For a 20-man raid, put 4 healers, 2 tanks, 6 melee and 6 ranged DPS and simulate the whole fight for different healer priorities. You then have two measures of success; 1) did the raid survive? and 2) which strategy did the highest HPS?

 

The missing piece is then to implement a priority system for your raid healers and follow it up with a believable model of your healer's mechanics. A good study would look at the performance of different strategies with different healer comps (or at the very least, your own healer comp for reference).

 

If you want to see some examples of the implementation which I made to study Chain Heal behaviour in the WoD Alpha, see my series of blog posts "To Craft A Sim" (parts 1 2 3 4 5, prelim results, theory of distribution, results post). You might find some useful notes on implementation of the raid and healing parts there, especially the mechanics of Chain Heal.

[Karanir 15]

Every "model" is an answer to a question. They are usually carefully designed to be as simple as possible, but not simpler, to provide specific information on a particular scenario. Typically, healing simulations aren't "realistic" in the sense that they model a real fight from start to finish. This is because the real world scenario is just way too complex, and there are so many different combinations and variables that you get a huge fluctuation of varying responses (which is not useful). 

 

You have stated that you are interested in tank survival, but this question is too broad and vague. I've read all your posts but I can't really pinpoint what you are interested in investigating. 

 

You would need to refine your query a bit better for us to provide a good answer to you. 

 

Some factors you may wish to consider: 

 

• Are you talking about healers working together as a team to ensure tank survival? Or individually? Different combinations of healers would completely change the outcome. 
  
• Are you talking about short term survival or long term? Is it a short spike of damage that lasts seconds? Or is it steady consistent damage taken by the tank throughout the fight? 
  

 

What are you interested in finding more about? 

 

• Is it the strongest burst "rotation" that gives you maximum healing throughput? 
  
• Are you trying to figure out how much burst healing you can do before running out of mana? 
  
• Are you trying to compare which is the best spell to use in various situations to help a tank survive? 
  
• Are you trying to figure out which healer is the most well-rounded? Example: they can handle a variety of widely different damage patterns? 
  
• Are you trying to investigate the impact of Shaman's mastery on the survival of tanks? Perhaps to show that shamans are more reliable in doing "clutch heals" to save the tank? 
  

[Lowpro 0]

I guess I should try to give a rudimentary idea.  The reason I am using the tank as an example is that if they die we can call that an endpoint; a wipe.  That's not always true but it's useful.  If I was making a survival model of that tank an each event affects its survival (the event discrete 0,1 where 0 means no damage taken because it was mitigated, and 1 meaning they've taken the hit. The amount of damage might as well be Markov Chains for this example).  Left alone the tank will always die because there's no healer right.  So we have maybe 4 actors who are independent healers that can "heal" our tank and keep them surviving each event.  

 

When you modify the damage events to resemble a raid fight, that would inform the heuristics of each 4 healing actors.  Because each actor is independent but they feedback on each other this model would necessarily find a heuristic that would keep the tank up.  If each 4 actors had different healing capabilities as we see in WoW we could see how the model adapts the four healers to keep the tank up.  Rather than being forward processing it's driven by feedback and thus integrative. 

[Pandacho 286]

Still can't understand what do you want to figure out 

The only thought that came through my mind was that you try to write a healing programm, because the healing process can not be simplified to 0 and 1.

Healing of tank is always a combination of tank's and healer's class abilities as a reaction OR prevention of specific boss mechanics.

 

For example:

- DK tank has strong self healing

- Paladin tank has shields

- Monk tank has mitigation

 

- Paladin healer has Beacon and Hands

- Disc Priest has shields and Clarity of Will

- Shaman healer has Earth Shield that reacts to damage

 

Boss mechanics may include (examples):

- physical

- magic

- DoT

- immobilization

- any combination of mentioned above 

 

So every specific case is different combination of boss abilities+tank class abilities+healer class abilities

Any damage can be prevented, healed or mitigated.

 

I'm not saying that it can't be counted and written down, but why to do it? In vast majority of situations any standard combination of tanks and healers is working, in the really specific cases you have to check all three components to base the decision on them: tank class, healer classes, kind of incoming damage.

 

But it's really not that complicated and does not need a theoretical base. If you know main CDs and abilities of tank and healing classes - and you do know this if you are a good enough RL, healing coordinator or even just good tank/healer, normally it's enough to jump to the right decision.

Edited January 13, 2015 by Pandacho

[Stoove 347]

But it's really not that complicated and does not need a theoretical base.

 

Sorry Panda, but I disagree on this. I think the more sophisticated the theoretical underpinning we have for raid healing, the better we can understand what's going on in real raids. Actually, I think the main challenge is discussing clearly how real theoretical concepts (e.g. Markov Chains) apply to raids.

 

I'm not saying that it can't be counted and written down, but why to do it?

 

This, I think, is precisely the right question to ask. What is the substantial aim of the potential simulation or model?

 

The problem is that experts tend to discuss things without giving a thought to those people who would benefit but don't speak the jargon, for example;

 

I guess I should try to give a rudimentary idea.  The reason I am using the tank as an example is that if they die we can call that an endpoint; a wipe.  That's not always true but it's useful.  If I was making a survival model of that tank an each event affects its survival (the event discrete 0,1 where 0 means no damage taken because it was mitigated, and 1 meaning they've taken the hit. The amount of damage might as well be Markov Chains for this example).  Left alone the tank will always die because there's no healer right.  So we have maybe 4 actors who are independent healers that can "heal" our tank and keep them surviving each event.  

 

When you modify the damage events to resemble a raid fight, that would inform the heuristics of each 4 healing actors.  Because each actor is independent but they feedback on each other this model would necessarily find a heuristic that would keep the tank up.  If each 4 actors had different healing capabilities as we see in WoW we could see how the model adapts the four healers to keep the tank up.  Rather than being forward processing it's driven by feedback and thus integrative. 

 

Lowpro writes like an academic of some kind (statistician, perhaps?) and from what I understand of the things they say, they have a good idea. The main problem is that the approach has to be spelled out from the point of view of a non-expert, without losing any of the technical accuracy.

 

Here is how I understand Lowpro's suggestion;

 

The aim is to make a model that looks at tank healing strategies and picks out "winning" co-operative strategies using minimal computational effort.

 

The method is to model a fight as a series of "potential tank-death events", with the tank having to survive each event by EITHER mitigation OR healing.

 

Once the pattern of events is decided, the simulation effectively works backwards to determine the best strategy for co-operative tank healing.

 

Somehow Markov Chains are important to this, but I lack the expertise to say exactly what their relevance is. I think that the biggest challenge in developing the idea further is to get a straightforward explanation of what that phrase means and where Lowpro thinks it's relevant. Note that this does not mean a wikipedia link or a textbook page*.

 

I think that this is overall a good idea for two reasons;

1. It brings around a conversation about tank healing that isn't often brought up. What are the optimal strategies, and what are the optimal stats for those strategies? We might find out something interesting.
2. The Theorycrafting community gets exposure to some of the more advanced tools of statisticians, which can undoubtedly be applied in other areas if we only have the expertise to do so.

__________________________

* - I know academia far too well to allow this. A wall of text is not an explanation, so much as it is a declaration that the person linking it knows better than the linkee.

[Karanir 15]

I think that this topic is highly interesting. It is relatively unexplored because most healer blogs speak about healing in isolation and focus only on a specific class. The concept of synergy between healers and synergy between different tanks and healers has been largely ignored. 

[Lowpro 0]

Sorry Panda, but I disagree on this. I think the more sophisticated the theoretical underpinning we have for raid healing, the better we can understand what's going on in real raids. Actually, I think the main challenge is discussing clearly how real theoretical concepts (e.g. Markov Chains) apply to raids.

 

 

This, I think, is precisely the right question to ask. What is the substantial aim of the potential simulation or model?

 

The problem is that experts tend to discuss things without giving a thought to those people who would benefit but don't speak the jargon, for example;

 

 

Lowpro writes like an academic of some kind (statistician, perhaps?) and from what I understand of the things they say, they have a good idea. The main problem is that the approach has to be spelled out from the point of view of a non-expert, without losing any of the technical accuracy.

 

Here is how I understand Lowpro's suggestion;

 

The aim is to make a model that looks at tank healing strategies and picks out "winning" co-operative strategies using minimal computational effort.

 

The method is to model a fight as a series of "potential tank-death events", with the tank having to survive each event by EITHER mitigation OR healing.

 

Once the pattern of events is decided, the simulation effectively works backwards to determine the best strategy for co-operative tank healing.

 

Somehow Markov Chains are important to this, but I lack the expertise to say exactly what their relevance is. I think that the biggest challenge in developing the idea further is to get a straightforward explanation of what that phrase means and where Lowpro thinks it's relevant. Note that this does not mean a wikipedia link or a textbook page*.

 

I think that this is overall a good idea for two reasons;

1. It brings around a conversation about tank healing that isn't often brought up. What are the optimal strategies, and what are the optimal stats for those strategies? We might find out something interesting.
2. The Theorycrafting community gets exposure to some of the more advanced tools of statisticians, which can undoubtedly be applied in other areas if we only have the expertise to do so.

__________________________

* - I know academia far too well to allow this. A wall of text is not an explanation, so much as it is a declaration that the person linking it knows better than the linkee.

 

Yes I am a biostatistician however my focus is on study design, not so much in theoretical statistics.  It's difficult for me to describe this model because it's stochastic and that is much easier to describe using math, not words.

 

The reason we are using Markov Chains is because each event of damage is independent to each other ( ie: each boss strike on the tank occurs independently.  The second hit didn't happen because the first hit happened) but because they do occur in a series the tank's position in space (that is, their healthbar's position) changes.  That change is directly responsible for the heuristics of our healing actors; they heal based on the damage the tank takes.

 

That's a complicated way of saying that the reason a healer heals the tank is to bring their healthbar up.  It seems silly to model but if you think about it, the reason we heal people and more importantly choose which heal, how we heal with that heal, are meant to reduce their risk of dying.  When four healers are trying to throw out heals but they don't know how the effect of each other's heal until it's cast, they're constantly making new decisions from each event of damage (because as I said, that damage is independent so each damage event means each healer has to reevaluate their strategy to the damage and necessarily how the damage is handled by the other healers.)

 

This is really just a "sort of for funsies" idea but one I feel could be made.  If that model reflects a fight and we make accurate tank and healing actors (we could even make DPS actors that take damage; we could do a LOT in this model) it'd be really cool to find a way to say take logs of a person's fight, find the success/fail simulation space that resembles their events and see where they diverge.