coderego

Well, after a few weeks of arguments within my guild, I came up with the idea to program this.

There really is no reason why we cannot quantify maximization based on group buffs.

So what I'm thinking is: we can do it one of two ways:

1) define theoretical weights on a point system. For instance: a shadow priest in group with a mage adds 10 value points to that mage. So if its a Shadow priest and 4 mages, that group is worth 40 points.
This system is simplistic and the numbers would probably be a bit arbitrary. However as long as they are relative to each other this should give us an accurate picture.

2) Define actual DPS increase based on level of gear. This is a much larger undertaking. What we can do is work with WWS to determine a particular classes DPS on a specific fight (yes we'd probably want to break this down to changing value based on the fight...) and see how each different buff effects on each different level of gear. Now a lot of this work has been done for us already by many of the great spreadsheet maintainers here at EJ.

I think #1 will work pretty well, but #2 will certainly be more accurate and will take a lot less convincing to get people to use it.

Anyway, before I get underway developing this...I'd like everyones opinion on which of the two they'd prefer, or if they have another option.

I plan to write this in Java... or perhaps python.



The initial commit isn't up yet, will post when it is.

If you are interested in joining me in coding this project send me a forum PM and I will add you as a developer

 

Liriel

I would like such a tool - but I dont think that you actually can do so since most fightes have different requirements for the Raid. Maybe you could build some optimal groups for standard situations (what ever that would be) but most times it would only be optimal for trash where you dont need optimum.

If mages (or other) have a lot to CC they dont do much dmg maybe it would be better to push someone else there.
If its a DPS-race maybe it is a good idea to push the tank instead of one other melee (or caster if tank is a paladin).
If its a bad encounter for melee (there are many) maybe you should push hunters over melee.
If its a realy heal-intensive fight maybe the shadow would do much more good with the healers, if not with the mages.
If you need some people in special spots maybe they would be better in a seperated group for efficient group heal (especialy if they would be too far away from their buffers).
There are fights where casters could profit much for a concentration aura - others where it does not do a difference.
In other fights it is a good idea to have a special totem or aura in every group in others a healer with groupheal possibilities. How to calculate this survivability against the dps-loss?

I dont think that you simply can fetch some WWS-stats from one fight to calculate something for the next week (or even another raid at all). In most times there will be some changes in the availability of people. Than there are fightes where it depends more on luck than on gear who can do much dmg (Archimonde).

And if someone is top of the dmg-meters is he there because of his skill? or his gear? or because you pushed him there with all the buffs which other players lacked? Would there be someone who could bring all the buffs to an even better result? If you dont try it, because your tool tells you this guy was top, you would never know.

I've seen many people with lower gear outperform players with much better gear. If you base your tool on gear you cannot be sure that the best DDler will be bufft to the optimum, even if they can push some little more out of their character. If you calculate everything on last weeks results and dont consider gear you would ignore the gear upgrades from that week.

Maybe you can find a good way to give a hint on how to set up DPS-groups that would be better than a lazy RL. But if it comes to the healers I dont know how to compair them inbetween and with the DPS. You cannot look on a healmeter and say healer A was top healer Z was last so A is the best healer in this fight, Z the worst. Some healers will get special appointments. If they focus on their given job they may push out less healpower then others - but if they wouldnt there could be a wipe. Even if you can say a healer was best and another last - which do you want to push? The best healer so he could do even better? Or the worst healer because the best healers showed that they could do their job without the push? I simply would ask them and switch people around accordingly.

In other situations you have to decide between doing it the fast way (pushing every dps) or the save way (pushing tanks or healers or helping with survivability instead of dps-synergy). Maybe the optimum would be in between....

Even if you could finde an answer to all this questions you cannot forget that players are human beings. If you dont give somebody a special buff every run (and that would happen with such a tool since you would not need the tool if you could give everybody everything) he would be frustrated and perform less or quit the raid. If you do this to too much players you cannot compensate such a loss.

 

coderego

While you make some good points there is one thing I don't think you consider:

There is really no reason why we cannot make the tool value certain classes more on certain fights. We can take DPS time into consideration as well as anything else an encounter might have specific to it.

 

Ragnor

If you want it to actually be accurate it will be a huge undertaking. You basically talking about writing something to model a boss encounter + taking the logic in every dps spreadsheet for every class and moving it online. Of course it can be done but that's a lot of time to spend on something.

Anything less than that and it won't be accurate enough to bother using imo.

The universe is run by the complex interweaving of three elements. Energy, matter, and enlightened self-interest.

www.retpaladin.com

 

Sinborn

I don't think encounter modeling would be required to make something useful. All you really need to report is raid dps , raid hps , tank tps, and tank survivability. Don't forget to report values of each considering single-target and aoe/raid targeting. Each character would have to have full modeling including raid/party buffs and mob debuffs.

I see this happening with less effort than you think. Imagine running Rawr 25 times (imagine Rawr modeled all classes and specs /drool), and having some master control program that pretends it has all of those 25 people in a "perfect" raid, doing "ideal" dps/hps/tps. Set some parameters, push the "optimize" button, and it spits out how to group/gear/spec for maximum raid dps, or max tank survivability, or you make it up from there.

 

Gruntle

Well, Rawr has support for 5 classes/specs or something currently? As you say yourself, you would have to have working models for all valid classes/specs. I don't think all of the current Rawr models are even completely agreed upon. The huge work will not be in writing the actual optimizer, but in modeling the performance of all the different classes. There is no common framework for modeling all classes yet (Rawr might be a starting point, but based on how slow more classes gets added to Rawr we'll likely hit WoW 4.0 before all are modeled), and without that I think a group optimizer will be pretty inaccurate.

I think the best you could do is to calculate scaling for different classes with buffs, then have an input figure for each possible raid member for dps/tps or some measure of healing efficiency (don't know enough about healing to know if healing per sec would be a useful number, I guess probably not). Your optimizer could then calculate the best raid setup using the starting performance numbers and the scaling laws. I have no idea if this would be accurate enough though.

I posted some scaling laws for DW fury warriors in the DPS warrior compendium thread, the exact numbers are of course subject to some changes (because the scaling will change slightly with gear level). It should be perfectly possible to run a few more gear setups to find how scaling varies from starting kara gear to SWP gear. Then you should try and get the same from all classes/specs with working spreadsheets (not sure there are spreadsheets/calculators for all though).

 

Liriel

Maybe you can do so for dps. Since dps everytime tries to put out all they can. It does not depend on the target. Maybe you have to decide between aoe and single-target.

But you cant simply calculate raid hps. Becaus it totaly depends on the encounter and the healing-appointments. It does not help, if raid hps rises, if one healer looses so much that he cannot hold up his healing-target if nobody else can reach the target properly without being able to do their orignial job. (There are fightes where some adds are tanked outside reach. For example at Karatress, Infernals in MH, Council in BT.) So maximising raid hps is not possibly the best option in each situation. You may at least think about maximising tank-healing-hps and raid-healing-hps but then you have to know the healing-appointments as well.

The bigger problem is, most healers dont try to heal the whole time with their best spells. Since that would only be overheal and mana wasted which would be needed later in the fight. With appropriate downranking or regen-breaks you loose hps but get more heal for the overall fight. Maybe you could tolerate a 3-minute-mage. But you definitly do not want to have a 3-minute-healer.

Also the hps of at least priests depends greatly on which spells he picks from his arsenal. That may be the same with dps-casters - but they have some more or less fixed rotations. You could calculate wich spells one would use how often in an optimal situation before the run. But you dont know that for the priests. Since you cannot calculate beforehand the dmg the raid has to take at every moment. But the priest (or maybe every healer) decides about his spell at every moment he can start a heal (or stop a heal). Healing is a great part reactive.

Btw: Healers dont simply think about hps - a lot of their time they think about hpm and manareg.

It's nearly the same with tank tps and tank survivability. And sometimes you simply dont need the best tps or maybe maxed survivability. (If one target is simply tanked but never killed the tank only needs to hold the tps from the healers.)

You HAVE to look at the encounter for deciding where to put which player. And at least if an encounter is not on farm status you will see some group-switches before and after some encounters in every raid. As I said the only "standard situation" is trash. But in most cases you dont need to optimise for trash. (With the exception of ZA-bear-run.)

 

coderego

Perhaps we can get relative comparisons done for now, and work on actuals later.

For instance: the following are DPS increases of one of my druids on specific buffs. We can simply say with these numbers: at tier 5

Unleashed rage: 104.74
Ferocious Inspiration: 96.44
Imp Battle Shout: 94.31
Battle Shout: 75.49
Imp GoA: 67.57
GoA: 58.68
Imp SoE: 55.03
SoE: 48.07
Trueshot Aura: ~30

Unleashed rage > Ferocious Inspiration > Imp Battle Shout > Battle shout > imp goa > goa > imp soe > soe > true short aura.

We can use the actual # of DPS increase per buff and add to the total raid DPS.

This can be done for all classes/specs from a DPS P.O.V....This will then simply become a mutli-variable maximization equation.

Opinions?

 

Liriel

The DPS P.O.V. simply is not the problem. The numbers you gave implicate that you profit well from a shaman. Casters like them, too, since they give manareg and spelldmg/healing and ele-shams give also a crit-totem. Healers do like the manareg and spelldmg/healing-totem, too. And your tank can boost his aggro with them.

So if you have no elemental and noch enhancement shaman where will you place the healing-shamans to optimise the raidthroughput? If you have an enhancement shaman does the tank need him or will he go to you?

Or consider a tree. If he is an the tank group the tank can be healed better. But if the tank is not a paladin, the tree will suffer because he is not in a healing group. You cannot calculate the percantage healing given by the tree to the overall healing, because it helps only on the people in the group of the tree.

Same (or worse) with warlock-imp. The people (tanks) in his group have better survivability if he gives them an imp. But he looses all the group-caster-buffs if he is moved to the tank-group and if he is not an affliction lock (and for raiding he will not be) his dmg suffers extremly for putting out the imp instead of saccrificing the succu.

The problem is, that its not only about dps. Optimizing dps-group-layout could be doable. But at some point you have to consider about your tanks and healers, too. You can ignore their needs in many situations or just give them a cookie (this is done most times in our raid). But there are such encounters where their needs may be more important than those of the DPS (Bloodboil, RoS,...).

And even if you only look on the DPS you have to consider if melee or ranged can do better or if there are some resistances for the caster. If melee have to run around half the time and ranged can stand still, you should buff the hunters, even if they cannot use the buffs as well as the melee. If the boss cannot be hurt by frost maybe you give the buffs to the lock instead of the frostmage. Even if normaly the frostmage can use them better.

 

Licity

I don't think you would need to sweat over assigning values to what person should go where, like the shadow priest + mage = 10 points example above. It would probably work better by just calculating it in direct dps, borrowing from Rawr or spreadsheet formulas.

 

Hegen

In method 1, I see the following advantage:

It's obvious to users that estimates are being done and coins are being flipped. This will encourage users to challenge these estimates and adjust them to their scenario. It makes them think. If the tool says: a shadow priest adds 10 value points to a mage and 8 value points to a warlock, it's obvious that these value points are rounded, artificial ranking units. The user will be encouraged to think: why is this so, and does it really apply to my players and scenario?

If translated into raw dps, the output will look more scientific without actually being more accurate in all situations. We will see raid discussions such as "the tool says we have 7,2 more raid dps if we move that shadowpriest to my group, so I insist on having him". At the same time, it will make it more difficult for the users to see how the estimates are being made and where to adapt them.

Overall, I think method 2 may actually produce more harm than good. Besides being a lot more work.


One suggestion for method 1, if I may:

How about adding a "trust value" to each value point rule? There will be rules where a community discussion will show great trust in a rule - and others where we will say: ok, that's true, the correlation is there, but it's really situational.

Example for a possible trust ranking:

Trust value 1:
Commonly accepted to be true in almost all scenarios

Trust value 2:
Commonly accepted to be true, but only assuming conditions (such as player skill or latency)

Trust value 3:
Commonly accepted to be true, but not applicable to many encounters.


The tool could implement filters based on these trust values. Users might start with just the trust value 1 and 2 rules. Then check up on the trust value 2 rules. As a next step users could rerun the tool to see what happens with taking trust value 3 rules into account - a "look for potential" check. If this last steps shows an interesting boost in raid dps, users could check up on these rules and see which of them apply to their situation. But they would do this only if running with these rules really shows a calculated potential.

 

coderego

Okay lets get this started with system one: lets define the value for each class/spec of having another class/spec in its group. We can also have the application have the following options: group for TPS, group for survivability, group for HPS, group for DPS, group for Optimal Setup.

This system will probably evolve to system 2 as we work it. I will start coding on system 1...once I have something I will post a subversion link and a direct download to the newest stable version. I plan on this being 100% open source. I have fixed on Java to code this with unless there are any strong objections. Please post your class/spec and how you value other class/specs. I will edit this post to keep the numbers current. Please feel free to discuss people's numbers with them if you disagree.

Ultimately I will model all of these myself using rawr/other dps spreadsheets. I just want to get underway with this. All in all, system2 remains inferior to system 1 because it is based on a persons desires and preferences. This will only work for us if every person prefers that the entire raid be bettered, not them individually.


Feral Druid
Enhancement shaman - 22.1 points
Non enhancement shaman - 10.6
BM Hunter = 9.6 points
Fury Warrior - 9.4 points
Warrior with no imp battle shout - 7.5 points
MM Hunter - 3 points

Now, what I plan to do for the alpha of this: user will input all 25 people's class/spec. We will assume they are speced correct (Enhance shamans have Imp GOA for instance). It brute force method calculates each possible group setup (5!^5 possibilities I think? Its early....). Then add the groups up and store the total raid DPS to a linked list, spit the highest setup back. I'm quite aware of how...gross that is. We'll improve it as we improve on how we calculate this. Eventually this will become a multi-var maximization equation...thats a bit down the line though.

Note: if anyone wants access to merge to the SVN and help code this, please send me a private message. I will want to talk to you and possibly see some code samples before giving you access

 

Liriel

I like the idea with the trust leves and the optimization-possibility in different directions. I think the best goal would be to make a semi-automatic-version, where you can (and have) to do somthing by hand before starting it.

Maybe in the end one coud place some players fixed in some groups because one needs them there (because you need a group-heal or because they stand appart from the raid or you need special auras or totems for resi....).

Maybe could set some players as tanks for this encounter with flags for: no thread problem, normal thread, maximised thread needed and somthing alike for survivability. Maybe you could do this for the healers as well (raid-healing/tank-healing/dpsing/tanking). Either you would give a flag (high-mana-issue) to some of them or place the shamans and spriest by hand. You even could make flags for dps-characters who have to tank or to do something else but dps most of the time.

After that you could use the program to find the optimal positioning for the non-fixed people. I think fixing players would only result in less not more variables which have to be optimized, so it should not hurt.

Btw I dont see that the equation would be to complex to find a good result in little time. This is because you can brake the whole thing down to some clusters which do not interfere between each other that much (healers, casters, tanks, melee+hunters). In most cases the optimazing-problem will be inside the clusters not between the clusters and there the number of variables is much less than for the whole raid. If you have found a good guess for your local optimizations you can use that as a starting point for a global opimization.

 

moink

I really don't think this is going to work for you. While you might be able to roughly calculate how various group buffs affect dps classes, I don't think you're going to be able to relate it to tanking or healing groups. I add a lot of dps to every encounter by keeping the raid alive; when I run out of mana, people die and stop doing damage. So how are you going to quantify the value of mana to me compared to the amount of mana for a mage?

This is also going to be highly dependent on the fight. I have never gone anywhere close to out of mana on A'lar, so I certainly don't need a shadow priest there; however I often go oom on Karathress and Tidewalker. How are you going to quantify the increased odds that I will kill my inner demon on Leo because a paladin in the fight has given me concentration aura?

But here are some numbers for you anyway. These are based on Jayde's Holy Priest spreadsheet. The assumptions on which that spreadsheet are based have been heavily debated in the priest thread; there's no accepted way to weight mana regeneration and healing for a holy priest.

The wrath of air totem adds 101 to my healing, and mana spring gives me 50mp5 which according to the spreadsheet is 143 +heal equivalent at my gear level, leading to a total of 244 for the value of a shaman in my group. A shadow priest giving me approximately 120 mp5 is worth 342 +heal equivalent. A point of crit is worth 0.8 +heal.

As you can see from my gear and my examples, I'm not at the level of raiding of most contributors to this forum.

 

Mijae

Even the simple weighting system for DPS alone is going to be much more difficult than you think. Setting something like buff 1 is worth X dps points is not valid because they increase value to other buffs as well. It also depends greatly on both gear and skill.

For example if I remove all buffs from my current druid, Imp GoA alone is worth roughly 55 dps. With BoK, Imp BoM and putting me in a group with 3 BM hunters and 4 Drums of Battle that totem is now worth 72 dps. Adding in Blood Frenzy, Imp SoC and Imp Hunter's Mark puts it up to 76 dps. Add Unleashed Rage and it's worth 83 dps. Add a second feral keeping up Mangle and it's now worth 86 dps. Take all those buffs and give them to a fully max geared druid powershifting once every cycle and Imp GoA is worth over 100 dps.

If you measure each buff individually, the total does not come out the same as if they are all active at the same time.

 

aleyro

Ok.... so, the "boil the ocean" approach, where we accurately model every possible class/spec/group and mathematically optimize the raid for every possible encounter is turning out to be (surprise!) not very viable. What would a simple, "naive" implementation look like? Maybe we can sketch that out, design it for extensibility, and prototype that- then, gradually improve the back-end modeling to improve accuracy.

How does that sound as a meaningful approach to moving forward?

 

Rhaegal

I keep seeing threads like this pop up, and I can never help but think that using such a tool would require a fair bit of suspension of disbelief. I understand what you're trying to do, but there are factors that simply can't be modeled. Imagine a situation where you have two BM hunters in the raid, one of which regularly out-DPSes the other despite inferior gear. You've gone beyond the ability of any tool to predict raid performance.

In my experience (and feel free to correct me if I'm wrong, but I imagine that this is true for 99% of guilds), theorycrafting class synergies and optimal group design is only half the story. It's an important half, but can be completely negated by mediocre players. Let's face it, if you have 4 spots in your caster group and 5 casters to put there, the odd one out is the one who tends to not pay 100% attention, is more likely to die do random avoidable damage ("why didn't anyone tell me I was in the shadow hole??"), or simply doesn't know which buttons to push in the right order compared to the other four. It doesn't matter that he has the +10 Sword of Casting Awesomeness and an app says he should be able to make better use of group buffs.

The job of a raid leader is to be wiser than impersonal models. I don't mean to sound quite as contradictory as I realize this comes across, I just feel like you're going to a whole lot of work for something that isn't going to be as useful as simple personal judgment combined with a reasonable knowledge of class synergies.

Stand back! I'm going to try SCIENCE!

 

Liriel

You have to treat additative and multiplicative buffes seperatly. But if you know the EP for each stat it should be no problem to compute them nearly correctly (since there are so many tools already doing so for single classes). It's hard to do so for a human but its no complex computation for itself.

Btw: I think about a tool which tells you where to put 10/25 given people. Not a tool who has to decide how many pala-buff-sets or curses or healers or else are present.

@Rhaegal: I'm very sceptical, too. I dont think a tool could be better than a good RL. But: There are those situations where we are pushed around in the raid and have to perform in less optimal combinations than possible, because the RL is to lazy to think about it. There may be a place for such a tool (if the RL is not to lazy to use it...)

The other thing is that sometimes you dont know how to place certain people and just put them somewhere. One of our RL does this with hunters. While loading the raidmembers hunters are put in an additional group and after everyone is set they are put where a place is free (most times with tanks or healers mixed whith some casters). A tool may help here to understand more about synergies and to try different strategies.

The tool would not produce the optimal for YOUR raid but it can help you to open your mind for playing around more with your groupsetup and ending with a much better understanding how to optimize your groups.

And it can be helpfull if you have to spread classes over the raid (for groupheal or buffes or because they have to play at different positions in the room) if you have to decide where to put all the people between the players that are fixed to get what synergies are left.

A tool can never counter the knowledge about your own raid and a human would produce much better results for his raid than a tool can - IF he even tries to do the optimization.


The bigger problem I see is that if such a tool would exist and if it would produce some acceptable combinations a RL could have problems to decide against such a tool. Look at your less-optimal hunter from your example. A good RL would not optimize groups for him because overall he would get a better result if he does so for other players. But the hunter may know that such a tool exists and the tool would tell that he would perform much better with all the buffs - even more, that the whole raid-output would be better if he got the buffs. He may than torment the RL with those results. Life would be much worse with such a tool in the hands of everybody else than without it...

 

coderego

You all make very valid points as to the con's of this tool. However I think the task of raid optimization IS possible (perhaps with boiling the ocean...we'll do it if we must!) However I think just forcing people to think about it a little more can help!

By the end of this week I plan to have the system's basics coded with some very fundamental math in it... just going to look at buffs in a black hole (not as they effect each other). Please continue discussions on the best way to do it =)

I'm honestly leaning mostly towards a "boil the ocean" approach. Model every class and buff in one tool. Perhaps we can even add in comparisons of individuals (oust that bad hunter! )

 

dedmonwakeen

Well, that's unfortunate........ because a full raid dps/tps simulator is in the works with a probable delivery in about two months. The sim already handled casters and I've spent the last month re-architecting it for physical dps. A week from now the core engine will be complete and I'll start implementing the classes.

Despite my investment in this effort, I'll still take a RL's intuition over a tool any day. However, a tool to enable synergy experimentation without wasting raid time might still be helpful to a RL.........

Unfortunately, the tool has no notion of a "player-suck" dial. All you can do is give that player a higher average random lag.

http://code.google.com/p/simulationcraft

 

coderego

SVN is up:

The initial commit isn't up yet, will post when it is. Editing first post with this subversion link.

If you are interested in joining me in coding this project send me a forum PM and I will add you as a developer

 

Rhaegal

Well, if you insist on working on this despite the major challenges, I might as well toss out some suggestions.

If you're really doing the "boil the ocean" approach, then you'll want fairly extensive meta data attached to each individual in the raid. Class, spec, and gear level are obvious choices. (This can and should be approximate--for example, I've never stepped foot in MH or BT, but thanks to badges, ZA, etc., I'd place somewhere between T5 and T6 for overall gear level.) The easy solution to the "that guy is bad" problem is putting in a linear scaling factor. Leave it at 1 for most players, but if someone only performs at 70% of what someone at their gear level should be capable of, factor that into their base DPS and any buffs they might be eligible for. You have to factor it into both because of multiplicative buffs (Unleashed Rage, Boomkin Aura) and static ones (Grace of Air, Battle Shout).

For people who have to perform some action to keep up their buffs (totems, Battle Shout), you should also include a way to input uptime of their buffs. In some cases, this effect will be huge. For example, if you have two enhancement shaman, and one's personal DPS is better than the other's but he has terrible totem uptime, the one with worse personal DPS but solid totem uptime belongs in the primary melee group. Fortunately things like WWS can (with some extrapolation) give you rough estimates of uptime, so this may be perfectly quantifiable.

Stand back! I'm going to try SCIENCE!

 

Roefyll

I've been creating a sheet like this myself, at work in my breaks.

Basically - it assumes a certain gearing (T6 + badge loot), uses an existing spreadsheet to determine the base DPS and buffed group DPS for every possible combination of party members.

This is all done from one mega table of classes and buffs.

The only tricky part is getting the raw data for classes that don't have awesome spreadsheets to work with.

I don't see any insurmountable issues preventing such a tool from functioning - it's not trying to calculate class dynamics, it's just referencing a table of precalculated data. There's absolutely no need to merge all spreadsheets into one monstrosity - that level of detail is irrelevant. You just want to get each character's DPS, their grouped DPS, and the raid's DPS.

The purpose of such a tool (for me at least) is to get a quantifiable idea of how various group makeups function. No, the theory will not always match the practice. That's true of every piece of theorycraft. I have to ask what someone is doing in this forum if they think trying to optimise raiding in a mathematical way is not worthwhile.

 

Liriel

Roefyll: You need more as only DPS if you want to get the best raid-setup not only the best dps-setup. Most here belive that you can create something that puts out a good estimated guess which combination could be best for overall DPS. But that is worth nothing if the tank cannot hold the aggro or everybody cannot outperform their buffs becaus they are aggro-starved. And it is even worth less, if the healers cannot hold the tank or raid alive.

Another problem I stumbled about is, when do you want to use such a tool? It has to be an addon if it could be usefull in a raid-situation. If it isn't you would have to do your calculations before the raid starts. That would be no problem if you simply want to do some theorycrafting. But if you want to bring it to use that would be a bad drawback...

 

aleyro

A couple of weeks ago, I started work on a naive implementation- it was basically a group "composer", instead of a true group optimizer. Over the next week or two, i'll look at cleaning it up and presenting it as an alternative approach- hopefully, addressing the challenge from two different directions will lead to a better overall solution.