Archived

This topic is now archived and is closed to further replies.

dedmonwakeen

SimulationCraft for Shamans

136 posts in this topic

When I run your priority setup in EnhSim it results in about a 1.5k dps loss. Guessing one of the sims is wrong but your money is as good as mine about which one.

From a commit this morning (not released yet):

Fix a pretty massive bug with Lava Burst. Now hard casting it properly resets swing timers, instead of pushing them into the future. This change removes hard casting of Lava Bursts from the melee/melee profile, as they result in a dps loss. With a caster mainhand, hard casting lava bursts still remains a dps upgrade.

There goes one squashed bug to get us closer together. Is it still a DPS loss in EnhSim to hard cast LvB even with the extra spell power from a caster MH?

Share this post


Link to post
Share on other sites

EnhSim isn't modeling the 10% boost yet so those results fit with what I am getting from EnhSim, ie that it could easily be a dps increase to hard cast lava burst as the lowest option on the priority list when running FT/FT with a caster mainhand. Only thing you might want to test out in your action priority list for both WF/FT and FT/FT is that EnhSim is showing that making sure to consume UE with a FS is more overall dps then just prioritizing refreshing FS based on remaining dot duration.

DMC is showing in the CharDev for WF/FT but the item list still shows H. Essence and Lightning Strike doesn't show on the damage breakdown list. DMC for WF/FT is at least a 450 dps upgrade over H Essence so that should help with the spread between them.

Still really not happy to see such a small dps difference between the two conditions. Part of that is that Crul's aren't very well itemized for us while Andoros isn't too shabby since it has mastery on it but even so they are just way too close in value for my liking.

Would you mind updating those things for the sample output, I want to link to it in Cochice's thread over on the DD forums to show that their fix to the caster weapon issue really just wasn't enough.

Share this post


Link to post
Share on other sites
Would you mind updating those things for the sample output, I want to link to it in Cochice's thread over on the DD forums to show that their fix to the caster weapon issue really just wasn't enough.

The profiles were changed recently in SVN, the release just hasn't caught up yet. We should have a new release with the LvB fix and the new profiles out "soon".

Here's the latest output using the new gear setups: Simulationcraft Results

Share this post


Link to post
Share on other sites
The profiles were changed recently in SVN, the release just hasn't caught up yet. We should have a new release with the LvB fix and the new profiles out "soon".

Here's the latest output using the new gear setups: Simulationcraft Results

Wow, so their "fix" really just leveled the playing field. Thanks for linking that. How did FT/FT with the caster mainhand do once you incorporated those fixes?

Share this post


Link to post
Share on other sites
Does that fix involve fixing the uptimes section? We need to know the FS uptime for elemental to see if the rotation logic we're changing works.

Reporting has been quite volatile lately as we are making back end changes. FS uptime and various other outputs should be fixed in the next release, which is the only thing preventing us from releasing a new build currently.

[e] Unless you're referring to how FS is no longer listed under the Uptime section, which has moved to Up% in the damage report.

Share this post


Link to post
Share on other sites

Following Bink's guidelines at Totemspot, I've been trying to prioritize FS cast adding this option (other default options unchanged, ES at 9, etc):

actions+=/flame_shock,if=!ticking|dot.flame_shock.remains<=X

I've run tests with values from X = 2 to 5 (shock CD talented) at intervals of 0.5 using many profiles (mine, BiS in simulator, current BiS in this post and many others) and I'm getting always a dps increase, the best at 3.5s with a ~150 dps gain, maybe due to the GCD and the time between the last two ticks.

Hope this helps, please correct me if I'm wrong.

Share this post


Link to post
Share on other sites

So I was simming some of my characters this evening and in the middle of that process I noticed simcraft finally updated to 4.2 (live) with T12 gear.

Now this is nothing out of the ordinary, and should be expected. What I didn't expect to see we're results like this:

Simulationcraft Results

Now considering that I play resto 90% of the time and play a prot pally the other 10% of the time, I'm no expert on Enhancement. With that said, I do know that enhancement hasn't really ever been flavor of the month (ahem, Frost) or really even above average dps. Dont take that the wrong way, Enh is also one of the most engaging and all-around fun specs in the game. So I have two questions for you super serious enhancement heroes before I go back to bitching about druids over on the resto forums:

1. Is the newest simcraft legit?

2. What does Enhancement_caster mean per se? I know that entails using a caster weapon and hard casting lava burst on cooldown, but is that it?

Share this post


Link to post
Share on other sites
So I was simming some of my characters this evening and in the middle of that process I noticed simcraft finally updated to 4.2 (live) with T12 gear.

Now this is nothing out of the ordinary, and should be expected. What I didn't expect too see we're results like this:

Simulationcraft Results

Now considering that I play resto 90% of the time and play a prot pally the other 10% of the time, I'm no expert on Enhancement. With that said, I do know that enhancement hasn't really ever been flavor of the month (ahem, Frost) or really even above average dps. Dont take that the wrong way, Enh is also one of the most engaging and all-around fun specs in the game. So I have two questions for you super serious enhancement heroes before I go back to bitching about druids over on the resto forums:

1. Is the newest simcraft legit?

2. What does Enhancement_caster mean per se? I know that entails using a caster weapon and hard casting lava burst on cooldown, but is that it?

1. I'm biased in that I'm one of the SimulationCraft developers (and I communicate a lot with our guy who does Shaman stuff both for Ele and Enh) but yes, I believe it is pretty legit. We do have some things that could use a closer examination but some of the things we know we do model correctly. Especially when it comes to how much abilities hit for. We test to the sub-+/- 1 dmg level with those where possible (Fire Elemental melee abilities are an example where it's not possible to be that accurate). We know that compared to EnhSim we model latency a bit differently and our action list includes hardcasting LvB when using a caster weapon but only as the lowest priority.

2. The _caster profile uses the same priority list as the non-caster profile with the following changes:

At the bottom of the priority list we change:


actions+=/lightning_bolt,if=buff.maelstrom_weapon.react=4

to:

actions+=/lightning_bolt,if=buff.maelstrom_weapon.react>1

actions+=/lava_burst,if=dot.flame_shock.remains>cast_time+travel_time

Otherwise we use the same priorities and the only gear difference is that it uses the caster axe (with Landslide and the spirit reforged to expertise rating). Basically, in SimulationCraft we see higher returns from using this caster weapon than seen in EnhSim. We're not really sure why just yet but we've been looking into it. We did find a bug in EnhSim yesterday that was stopping LvB from being cast as often as it could be even if trying to hardcast. Maybe we'll find either more bugs in EnhSim or some in SimulationCraft to explain the differences.

Share this post


Link to post
Share on other sites

I don't know if there is coding for this to ignore one or the other but i saw in the "Constant Buffs" section in the SimCraft that both Mark of the Wild and Blessing of Kings is there. Could be that maybe they are stacking and inflating some numbers?

Constant Buffs

arcane_brilliance

blessing_of_kings

blessing_of_might

blessing_of_might_regen

fortitude

lightning_shield

mark_of_the_wild

replenishment

I also noticed the following stats:

Raid-Buffed Unbuffed Gear Amount

Strength 738 154 20

Agility 8659 5917 5558

Stamina 8712 7039 6890

Intellect 370 353 217

Spirit 267 267 109

Raid-Buffed figures seem a bit off but maybe its my lack of experience on SimCraft that is about to make me look like an idiot.

Share this post


Link to post
Share on other sites
I don't know if there is coding for this to ignore one or the other but i saw in the "Constant Buffs" section in the SimCraft that both Mark of the Wild and Blessing of Kings is there. Could be that maybe they are stacking and inflating some numbers?

Constant Buffs

arcane_brilliance

blessing_of_kings

blessing_of_might

blessing_of_might_regen

fortitude

lightning_shield

mark_of_the_wild

replenishment

I also noticed the following stats:

Raid-Buffed Unbuffed Gear Amount

Strength 738 154 20

Agility 8659 5917 5558

Stamina 8712 7039 6890

Intellect 370 353 217

Spirit 267 267 109

Raid-Buffed figures seem a bit off but maybe its my lack of experience on SimCraft that is about to make me look like an idiot.

The following is the code where things like MotW and BoK gets used:


// player_t::composite_attribute_multiplier ================================


double player_t::composite_attribute_multiplier( int attr ) SC_CONST

{

  double m = attribute_multiplier[ attr ];


  // MotW / BoK

  // ... increasing Strength, Agility, Stamina, and Intellect by 5%

  if ( attr == ATTR_STRENGTH || attr ==  ATTR_AGILITY || attr ==  ATTR_STAMINA || attr ==  ATTR_INTELLECT )

    if ( buffs.blessing_of_kings -> up() || buffs.mark_of_the_wild -> up() )

      m *= 1.05;


  if ( attr == ATTR_SPIRIT )

    m *= 1.0 + buffs.mana_tide -> value();


  return m;

}

As you can see (if you understand C++) you only get to benefit from them once. As for the Raid-Buffed values, they are taken when it first executes the "snapshot_stats" action in the priority list. For some profiles we still have that action happening after they pop their pre-combat potion whereas for others we call snapshot_stats before popping their pre-combat potion. snapshot_stats just takes a record of stats at that given point in time for the purposes of displaying it on the final report. It has no actual effect on the simming itself.

Share this post


Link to post
Share on other sites

Regarding the SimC results, there's been some discussion about the reaction times used in SimC vs EnhSim. Just from looking at some of the spell frequencies in the caster weapon results for SimC (particularly with the use of MWS2/3/4), I find some of the results hard to swallow. That said, that means that there would be implications for more classes than just Enhancement (likely any spec monitoring multiple procs with no concern with resources).

I don't doubt that we will be using the SP axe and hardcasting LvB in Firelands, and putting out competitive numbers. But I think that the discrepancy between the sims is mostly due to a difference in considering reaction times. I don't think that it's realistic for a player to react fast enough to get through MSW2/3/4 and down to LvB at the bottom and still have as many LvB casts as SimC is reporting - I believe one every 13 or 14 seconds. And that's with the ability at the bottom of the priority list.

Share this post


Link to post
Share on other sites
Regarding the SimC results, there's been some discussion about the reaction times used in SimC vs EnhSim. Just from looking at some of the spell frequencies in the caster weapon results for SimC (particularly with the use of MWS2/3/4), I find some of the results hard to swallow. That said, that means that there would be implications for more classes than just Enhancement (likely any spec monitoring multiple procs with no concern with resources).

I don't doubt that we will be using the SP axe and hardcasting LvB in Firelands, and putting out competitive numbers. But I think that the discrepancy between the sims is mostly due to a difference in considering reaction times. I don't think that it's realistic for a player to react fast enough to get through MSW2/3/4 and down to LvB at the bottom and still have as many LvB casts as SimC is reporting - I believe one every 13 or 14 seconds. And that's with the ability at the bottom of the priority list.

With regards to reaction times most (hopefully all by now) of our profiles contain things like the following in their action lists:

actions+=<someability>,if=buff.somebuff.react>=2

or

actions+=<someability>,if=(!ticking|dot.somedot.remains<gcd)&miss_react

The way the first one works is something like this: For every buff it keeps track of a time when you're allowed to "react" to it being that many stacks. This happens like so: When a buff changes to a new stack level we generate a random reaction time based off the formula: react_time = gauss( mean, stddev ) - ln ( random( 0.0 -> 1.0 ) / nu ) * nu. With a minimum reaction time of 0.0 seconds and a maximum of 5 seconds. Our default values are: mean = 0.5 seconds, stddev = mean * 0.25, nu = 0.5. We then add to this react_time the current sim time ( sim -> current_time ) and we store this value in an array for the buff indexed by stack number.

Thus if say we went from 2 stacks of MW to 3 stacks we would generate a react_time, add the current time and store this against the MW buff for 3 stacks. Now any action that depends upon knowing if we have 3 stacks of the buff up will fail its check until such as time as the sim->current_time is >= this stored time.

Thus one cannot simply instantly react to the situation of having a certain number of stacks and changing what you choose to cast based off that. You can't react to such a condition until you've passed the reaction time for it. As such it can be feasible to be casting LvB etc. with our priority. All it says is:

actions+=/lightning_bolt,if=buff.maelstrom_weapon.react>1

actions+=/lava_burst,if=dot.flame_shock.remains>cast_time+travel_time

Which means that if we are able to notice that we have more than 1 stack of MW up we'll cast LB otherwise we'll cast LvB. Because Lighting Bolt has no cooldown passing a test of if=buff.maelstrom_weapon.react>1 for LB is the same as if we added a if=buff.maelstrom_weapon.react<=1 for LvB. We'll only reach that point if we have time to react to the change in MW stacks.

The second example listed above works similar to the buff.react but just says if enough time has passed (again based off that gaussian reaction time) that you've notice that your spell has missed so that you'll cast it again.

What all this means of course is that our reaction modeling means that at times our priority list can make mistakes. It might have you casting a MW5 Lightning Bolt because you gained a new stack of MW but didn't have time to react to it. It might also miss noticing a MW2+ Lightning Bolt and instead cast Lava Burst again because it didn't have time to react to gaining the stacks of MW.

So we're not actually expecting that a player (or the sim) to always cast the perfectly correct spell all the time. Sometimes, because of reaction time they will simply not cast the optimal spell and suffer the DPS loss from doing so.

Share this post


Link to post
Share on other sites

Althor has covered the time needed for a human to react to the fact that the MW stack has just gone from 1 to 2, which we do model. The fancy formula he posted results in an ex-Gaussian distribution with a 650 millisecond expected value. As he mentioned, not noticing that 2nd MW stack immediately actually results in *more* LvB's being cast, not less, so quoting the "high" number of LvBs as some sort of indication that we're not modelling reaction times right is disingenious.

Now, there does seem to be a lot of people who think we need to somehow put more "reaction time" in our model - after all, there are other considerations here, such as whether LvB is off cooldown and whether FS is going to be up for long enough. Don't they need to be "reacted" to? I'm going to quote from a PM conversation I had with Rouncer, where he brought up that notion:

We only model reaction time when there is an actual reaction (to a "surprising" event) involved. I don't see that there is such a reaction involved when GCDs become available. Most players will have timers showing them the duration left on any dots and cooldowns, and for a skilled player there is a lot of thinking ahead to plan the next few abilities used. This means the player will know in advance whether a GCD will be available, and since he's already paying attention to flame shock he should also be able to know in advance whether LvB is a viable option for filling that GCD. Because these things are knowable in advance, it makes no sense to model them with a "human reaction time" using an ex-Gaussian distribution.

It seems to me that the limiting factor in the situation you describe ("reacting" to a free GCD being available and the fact that FS will be up for longer than the cast time of LvB) is not human reaction time, but things like focus, attention level, and the ability to keep track of multiple things at once.

Now, I think there is actually a fundamental disconnect at work here - there's a lot of talk about how there's no way to fit that many LvB/LB casts into a standard enhance rotation, but our Caster profile still has 7% dead time, so obviously there was time. (And if you don't believe that number, you can just go through every ability and add up the multiple of the number of times they are cast and the time they take to execute, and divide by the total average fight time.)

The disconnect is this: The priority list is not considering the length of the LB/LvB casts before starting them. For example, it may start an LvB cast 50 milliseconds before the cooldown on LL is up, thus postponing the LL by more than a second. If you look at the ability details you'll notice the Caster profile is not just sacrificing melee swings to cast LB/LvB, it's sacrificing every other ability as well. This sounds bad, but it really isn't - it's the playstyle of every caster DPS: Never stop casting, because not casting is wasted time. (Of course the melee swings mean this isn't 100% true in the enhance case, but when you wield a caster weapon your melee is sufficiently weakened and your spells are sufficiently boosted to the point where meleeing is close to being "wasted time".)

In short, this particular enhancement shaman is fundamentally more like a caster DPS than a melee DPS, and if you look at the priority list with a caster mindset you'll see how it's really not that hard to pull off. Just think of LvB as your filler and cast it whenever you're not casting something else. If it's on cooldown and you're still not casting something else, just spam the button anyway since you'll want to cast it as soon as it comes off cooldown.

(As always, I should note that this is all assuming a perfectly stand-still fight. Melee swings obviously become more important if fight mechanics force you to move but you are able to stay within range of the target while doing so.)

Share this post


Link to post
Share on other sites

Now, I think there is actually a fundamental disconnect at work here - there's a lot of talk about how there's no way to fit that many LvB/LB casts into a standard enhance rotation, but our Caster profile still has 7% dead time, so obviously there was time. (And if you don't believe that number, you can just go through every ability and add up the multiple of the number of times they are cast and the time they take to execute, and divide by the total average fight time.)

The disconnect is this: The priority list is not considering the length of the LB/LvB casts before starting them. For example, it may start an LvB cast 50 milliseconds before the cooldown on LL is up, thus postponing the LL by more than a second. If you look at the ability details you'll notice the Caster profile is not just sacrificing melee swings to cast LB/LvB, it's sacrificing every other ability as well. This sounds bad, but it really isn't - it's the playstyle of every caster DPS: Never stop casting, because not casting is wasted time. (Of course the melee swings mean this isn't 100% true in the enhance case, but when you wield a caster weapon your melee is sufficiently weakened and your spells are sufficiently boosted to the point where meleeing is close to being "wasted time".)

Actually the disconnect is that your caster is NOT sacrificing the other abilities. It's only losing 0.2 SS, 0.1 LL, 0.03 UE and differences with shocks are negligible. So it's fitting in almost 10 seconds of casting every minute while only sacrificing less then a half a GCD worth of other abilities. That's why I am having an issue with it.

When I run that list in EnhSim with the 250-350ms worth of latency/reaction time that I had confirmed through comparison testing with Saurfang raid parses to give comparable ppm rates for most attacks (it still wastes less MW charges then I saw in the parses) I only get 1 LvB every 34 seconds and the number of LBs is lower then seen in your caster model.

Want to show me something that I feel would be reasonable, then either remove the MW4/3/2 line or remove the LvB line from the caster priority. Keeping both, in my mind, makes it into something not feasible for players. It makes it into something where it's just about seeing how well a computer could do by playing with the priority list not how to make the priority list into something that will make people play better.

Share this post


Link to post
Share on other sites
Actually the disconnect is that your caster is NOT sacrificing the other abilities. It's only losing 0.2 SS, 0.1 LL, 0.03 UE and differences with shocks are negligible. So it's fitting in almost 10 seconds of casting every minute while only sacrificing less then a half a GCD worth of other abilities. That's why I am having an issue with it.

At this point we're chasing each other from forum to forum, but I think this is a better place for the discussion, so I'll repost here:

It's not "less than half a GCD over an entire minute". If you lose 0.2 SS per minute that means in a five minute fight you lose an entire SS. A lost SS means you postponed your SS casts by enough to push the last SS cast off the end of the fight. This is true for all spells that are on a cooldown, and the time you need to "lose" before you lose a full cast of a given spell averages out to the duration of the cooldown on that spell.

So in order to lose an entire SS over the course of a fight you need to lose 8 seconds. Normalize it back to a minute and you've lost 1.6 seconds just on SS. Do the same math for all the other abilities and you'll see that our LvB casting upsets the rest of the rotation by about 4.3 seconds per minute. In other words, each LvB postpones other abilities by about a second on average, which seems entirely realistic to me.

(EDIT - Naively: You might be "unlucky" and start casting LvB a millisecond before one of the other spells come off cooldown, or you might be "lucky" and have room to cast an entire LvB without postponing any other ability. If those are the extremes and the average is in between, you'd expect that average to be about a second, given LvB's cast time of 2 seconds. And that's not even accounting for haste.)

Share this post


Link to post
Share on other sites
At this point we're chasing each other from forum to forum, but I think this is a better place for the discussion, so I'll repost here:

It's not "less than half a GCD over an entire minute". If you lose 0.2 SS per minute that means in a five minute fight you lose an entire SS. A lost SS means you postponed your SS casts by enough to push the last SS cast off the end of the fight. This is true for all spells that are on a cooldown, and the time you need to "lose" before you lose a full cast of a given spell averages out to the duration of the cooldown on that spell.

So in order to lose an entire SS over the course of a fight you need to lose 8 seconds. Normalize it back to a minute and you've lost 1.6 seconds just on SS. Do the same math for all the other abilities and you'll see that our LvB casting upsets the rest of the rotation by about 4.3 seconds per minute. In other words, each LvB postpones other abilities by about a second on average, which seems entirely realistic to me.

(EDIT - Naively: You might be "unlucky" and start casting LvB a millisecond before one of the other spells come off cooldown, or you might be "lucky" and have room to cast an entire LvB without postponing any other ability. If those are the extremes and the average is in between, you'd expect that average to be about a second, given LvB's cast time of 2 seconds. And that's not even accounting for haste.)

This is my big question (Bolded). If the new caster priority rotation is correct and we end up casting more spells in excess of 1.5 seconds, we would need to re-evaluate the benefits of haste under these circumstances.

Share this post


Link to post
Share on other sites

I guess the thing that needs to be added is the reaction time necessary to respond that we have a free GCD. Having a free GCD is not the norm so there should be reaction time to recognize that the GCD is free so might as well cast a LB with less then a 5 stack or hard cast a LvB. There is no cooldown to look at but we still have to react to it's existance and since 85% of the time we have something else we should be hitting there has to be an allowance to react to it.

I always thought the idea of a simulator was to simulate actual conditions and performances from the game in order to help them to play better. We're no longer anywhere near that point. When I hit this point with EnhSim I started comparisons to in-game parses and dummy runs to make sure that it was showing something real. That's what should be done with simulationcraft. To show that the caster priority really can be that complicated. Otherwise all you've shown is how much faster a computer can do things then a player.

Share this post


Link to post
Share on other sites

While I accept that your "feelings" on this issue have some merit because of your experience playing enhance, I haven't really seen any strong argument against my stance on how playable the rotation is either. As I've said before, "reacting" to having free time to cast LvB is a non-issue if you treat LvB as something you should be casting whenever you're not casting something else. Any caster DPS is used to dealing with a filler spell like that - sure, it's a bit more awkward than your usual filler spell due to its cooldown, but spamming the button for a spell that's still on cooldown never hurt anyone.

Share this post


Link to post
Share on other sites
While I accept that your "feelings" on this issue have some merit because of your experience playing enhance, I haven't really seen any strong argument against my stance on how playable the rotation is either. As I've said before, "reacting" to having free time to cast LvB is a non-issue if you treat LvB as something you should be casting whenever you're not casting something else. Any caster DPS is used to dealing with a filler spell like that - sure, it's a bit more awkward than your usual filler spell due to its cooldown, but spamming the button for a spell that's still on cooldown never hurt anyone.

But it's not just waiting for LvB to be off cooldown, it's waiting for all the other cooldowns to be active. It's waiting for there to be no other button to hit, so why would you spam it? 85% of the time there is something else to hit, so 15% of the time you would want to hit that button and you would never want to spam it because if you hit it more then once or twice then there is likely to be something else you should be hitting instead.

I spammed my lava lash button up till now. I will be spamming my lava lash and my stormstrike button tonight if I play. Whenever one of those comes off cooldown, I want to hit it. When they are on cooldown then I hit something else. Only when there is nothing else to hit at all would I want to cast LB with less then a 5 stack or, under your priorities, cast a LvB.

You can't really look at us like a caster because we're not and even when you bring a caster weapon into the picture we still won't play like a caster.

Edit - here's the current top 25H Chimaeron parse - World of Logs - Real Time Raid Analysis

Here is the breakdown of average time between uses in that parse

LB - 9.2

LL - 11.8

SS - 10.54

UE - 17.4

FS - 16.4

ES - 16.4

This is the breakdown for SimulationCraft's melee weapon model. The priority lists of each should be pretty similar.

LB - 7.46

LL - 10.45

SS - 9.36

UE - 15.89

FS - 19.84

ES - 10.94

See the discrepancies. Now realize that with the caster model you are squeezing even more into a situation where we can't even come close to what you were showing even with all the additional "dead time" in the melee model.

Edit2

To be fair, here is EnhSim's breakdown with a generally default config running 250-350ms latency/reaction time and a similar priority list to the melee model. (at work so don't have access to my versions of the HMT12 configs)

LB - 7.74

LL - 10.92

SS - 10.3

UE - 16.5

FS - 16.9

ES - 14.4

Share this post


Link to post
Share on other sites
You can't really look at us like a caster because we're not and even when you bring a caster weapon into the picture we still won't play like a caster.

I've yet to see any explanation for why not. Every caster has a bunch of other things (cooldowns, dots, self buffs, target debuffs, whatever) they should be prioritizing higher than their filler, just like you just described. That doesn't make them sit around for hundreds of milliseconds deciding whether to cast their filler or something else.

I'll accept though, that the number of relatively short cooldowns would make "caster enhance" a bit more hectic to play than virtually any other caster out there at this point in time. But (by default) we simulate low network latency, human-equivalent reaction times, and "perfect" skill at keeping track of cooldowns and other timers. We don't try to emulate some randomly selected skill level as defined by some existing WoL parse which isn't even from a perfect Patchwerk-style encounter.

Share this post


Link to post
Share on other sites
I've yet to see any explanation for why not. Every caster has a bunch of other things (cooldowns, dots, self buffs, target debuffs, whatever) they should be prioritizing higher than their filler, just like you just described. That doesn't make them sit around for hundreds of milliseconds deciding whether to cast their filler or something else.

I'll accept though, that the number of relatively short cooldowns would make "caster enhance" a bit more hectic to play than virtually any other caster out there at this point in time. But (by default) we simulate low network latency, human-equivalent reaction times, and "perfect" skill at keeping track of cooldowns and other timers. We don't try to emulate some randomly selected skill level as defined by some existing WoL parse which isn't even from a perfect Patchwerk-style encounter.

and that's why we are arguing about this.

Show me a single parse from any encounter. Show me a single parse from a dummy. Show me anything that can get anywhere near that performance and I will shut up. You guys choose to give the caster weapon a separate priority list. You choose to make it one that isn't realistic.

Chimaeron is a pretty simple encounter, right? Movement every X number of seconds while staying inside melee range. Incoming damage every Y number of seconds. Bloodlust at a specific health percentage into a hard burn phase. So set up SimulationCraft to emulate the encounter and compare a melee priority to some actual raid parses. If the ability usage results are similar then I'll quiet down until we get some parses from people trying your caster priority list with that caster axe. If they aren't then maybe you guys should take another look at how you are simulating reaction time when it comes to melee specs with their multiple short cooldowns and work on making your modeling more accurate. Especially when thinking about throwing a hard cast spell that precludes movement into the mix.

Edit - and if I am right then do what I asked and make the bottom of the Enh caster priority list for the sample either MW4/3/2 or LvB but not both. (and if you go with LvB then stick a full reaction time response onto it so that it simulates the person noticing that they have an actual free GCD and then responding by casting the LvB).

Share this post


Link to post
Share on other sites

Everyone makes mistakes. Even a really really good player who gets all the stars to align with crits and procs and buffs and lands a #1 WoL parse will have made small mistakes when choosing their next ability at multiple points during that encounter. And there are a lot of specs/classes that are hard to play. If we were to try to match some random "good" player's skill level for every class and spec, all of their reported DPS would go down.

Subtlety is a good example. Everyone who plays a rogue at a high level and sees Subtlety a bit higher than the other two specs in our results will not immediately respec, because they know Subtlety is a lot harder to play perfectly. This "Caster Enhance" is, at least for most players, an entirely new thing which may result in people assuming it's harder to play than it really is, or possibly the other way around, but as more people try it I'm sure we'll see something similar occur there. People who really enjoy the playstyle or feel confident that they can handle the added complexity will still choose to play like that, but most people will not.

That's all fine, and none of that invalidates our results. We simply choose not to go chasing after some arbitrary skill level and instead assume perfect skill. We know nobody actually has perfect skill, but what we simulate is *possible*, and that's what we're aiming for.

The big problem I see with your argument is that you're still trying to get us to model "reaction time" in a scenario that involves no random event to react to. Our modeling doesn't assume super-human reaction time, it simply assumes a human who is very very good at keeping track of timers and planning out ability usage according to a predefined priority list. To better illustrate my point, I would like you to imagine a very clever addon. It watches all your cooldowns as well as the flame shock dot and the searing totem timer. It uses that information to arrange a list of skills along a future timeline for when they will become available for use and shows clearly which skill to use next at all times.

First imagine a world without MW. Because there is no random proc that influences your rotation in that world, the clever addon can show the order of skills to use for quite a long time into the future, so you have plenty of time to prepare, and your job becomes the rote task of hitting the various buttons when the addon tells you to. You even have time to sit ready to spam the relevant button for a split second before each cooldown will come up, so there's no "latency" involved. If you don't fail at your job of staring at the addon and hitting the correct buttons you will have "perfect skill" in the same way our simulated actors do. That this is realistic should be undisputed, and it's important to note that LvB would be one of the abilities handled by the addon. There is no "reaction time" involved in casting LvB - the addon simply correctly predicts that there will be a moment when no other ability is off cooldown and FS will still be ticking at the time you're done casting, and the future timeline will include you casting LvB at that precise moment.

Now add MW back into the mix. This is a random proc, so the clever addon can not predict its occurrence and can thus not place it along the future timeline. It can, however, be helpful and show a big flashing icon (power aura style) when you need to cast LB, and it can do so at the correct times based on other cooldowns and the current stack level of MW, because it knows what your priorities are.

That big flashing icon is definitely something that should cause you to change your ability usage from the perfectly predicted one showed in the future timeline of the addon, and it will show up without warning and thus needs to be reacted to. But we model this reaction time, at an average of 650 milliseconds. If the simulated actor is determined to not have had enough time to react to the MW proc reaching the relevant stack level, it acts as if the proc had not occurred. In other words, it acts as if the player who's sitting there and staring at the addon's future timeline stares at it (and continues to cast as it directs) for a little bit too long. Again, I don't see how the realism of this modeling can be disputed given the scenario, and it's again important to note that LvB is still part of the future timeline displayed by the addon and is not in itself affected by any randomness introduced by MW.

After the big flashy icon pops up and you notice it and hit your LB button, you go back to staring at the future timeline, and you have plenty of time (either the duration of your LB cast or the duration of a GCD if the LB was instant) to adjust to any changes that have occurred in that timeline since last time you looked at it.

Now, if you can present an argument for why such an addon is impossible to create, please do so - who knows, I may have missed some subtle part of enhance gameplay. If it *is* possible to create such an addon, why shouldn't Simcraft simulate a player that uses one?

(I know there have been addons that do this for other classes/specs in the past, and I think Ovale tries to do it for all classes/specs. It may need to be adjusted for a "caster enhance" playstyle before it's usable, but anyway, my argument doesn't really rely on such an addon actually existing.)

Share this post


Link to post
Share on other sites

Wow, so it hard casts LvB without any sort of reaction time or thought involved at all. EDITED (sorry, I'll try to stay civil)

Did you look at the Chimaeron values? Did you try setting up your sim to replicate the fight and see how action usage with the melee priorities (which are the one's I don't have any problem with) compare to the top parses? Those are the things I did when confronted with conflicting data between EnhSim and raid parses. I'm not saying EnhSim models reaction time/latency well or that it's results compared well to those live parses. The 250 - 350ms setting is nothing more then a compromise that gets results that compare favorably enough to what is seen in-game to have relevancy in helping people play the spec better. Although it still doesn't handle MW usage as well as I would like at that setting. I'd love to have a more robust reaction time model incorporated like Althor describes, although in this case you guys really just aren't using it properly to handle our mechanics.

Look, when your simulator's results has no relation to what occurs in what you are trying to simulate then you don't say the thing you are simulating is wrong, you fix the simulator.

Share this post


Link to post
Share on other sites

Look, you don't fix a simulator by tweaking a single crude dial (EnhSim's "latency" setting) until you get the output you want. I submit that Simcraft has a very clear and explicitly defined relation to what we try to simulate, while EnhSim does not, it just tries to tailor certain details about its output to match WoL reports while not actually caring about the mechanics of how it arrives at that output.

Share this post


Link to post
Share on other sites
Look, you don't fix a simulator by tweaking a single crude dial (EnhSim's "latency" setting) until you get the output you want. I submit that Simcraft has a very clear and explicitly defined relation to what we try to simulate, while EnhSim does not, it just tries to tailor certain details about its output to match WoL reports while not actually caring about the mechanics of how it arrives at that output.

and yet EnhSim is far more accurate a portrayal of an enhancement shaman then simcraft.

EnhSim is just as valid a simulator as simcraft. It has just as much, if not more, of a relation to what it's trying to simulate and a lot of what is in simcraft about enhancement mechanics comes straight from EnhSim and the testing we have done. Only difference is we've always tried to make sure that EnhSim is presenting an accurate representation of what happens in game, apparently you can't say the same.

You may not fix it by tweaking a crude dial, but sticking your head in the sand when shown that your simulation is flawed isn't going to fix anything either. Besides, I have tried to do more but I'm not a coder so all I can do is write up tickets about what I'd like to see and hope that the people actually doing the coding take an interest and work with me to make it more accurate.

I really don't care which is the better tool. All I want is the tools to be accurate enough to help me be a better player. Simcraft isn't looking that way to me right now, not when it's output is so far off from what is actually happening in the game.

Share this post


Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.