[Shaman] Experimental combat simulator - Page 8

Malan · 08/09/07, 2:38 PM

draghkar would consider renaming this thread? Slow vs Quick weapons is not what your end-state is. How to Model Windfury or some such would best represent what you're doing. The current title is extremely misleading.

Originally Posted by draghkar

Question, I introduced a second weapon (offhand) and now I have a problem I need to solve.
I roll the dice at each swing, what happens if the two weapons shall do a swing at same time,
and both have a WF proc? Wich of the two I have to consider? MH or OH?

The combat log in most cases reports the MainHand first, followed by the Offhand even if they both get the same timestamp. You could therefore consider that the Mainhand would always get priority (assuming that the weapon speeds are identical). However we know from in-game observation that once lag and other factors get introduced things get wonky, and people have noted that they'll have long streaks where the off hand is hitting first and always gets the windfury proc.

draghkar · 08/09/07, 3:10 PM

Originally Posted by Malan

draghkar would consider renaming this thread? Slow vs Quick weapons is not what your end-state is. How to Model Windfury or some such would best represent what you're doing. The current title is extremely misleading.

The combat log in most cases reports the MainHand first, followed by the Offhand even if they both get the same timestamp. You could therefore consider that the Mainhand would always get priority (assuming that the weapon speeds are identical). However we know from in-game observation that once lag and other factors get introduced things get wonky, and people have noted that they'll have long streaks where the off hand is hitting first and always gets the windfury proc.

Changed the name, not sure if that is the correct one, but for sure it will not give wrong information to new players. Ok, at the moment I will give MH events an high priority over OH events.

songster · 08/09/07, 3:26 PM

Hang on, for at least the 1H case it should be possible to do it analytically.

At 3s and up, you get the full benefit of WF.
From 1.5s - 3s, you get 5/6 of the benefit of WF (proc is 1/6 instead of 1/5)
From 1s to 1.5s, you get 5/7 of the benefit (proc is 1/7)

General case: n = 1 to infinity.

From 3/n to 3(n-1) seconds weaponspeed, you get 5/(n+4) times the benefit of WF. As n tends to infinity, the added WF DPS tends to zero.

Adding flurry in, I suspect but can't prove that it's simply down to flurry uptime. The location of each jump is shifted by a factor of 1.3, and the depth of each jump multiplied by the uptime.

So for a flurry uptime of 20%, there's a jump at 3.9 seconds that's 20% of (5/5-5/6) and another jump at 3 seconds that's 80% of (5/5-5/6). There's a jump at 1.95s that's 20% of (5/6 - 5/7) and another at 1.5s that's 80% of (5/6-5/7). And so on.

Shouldn't be beyond the combined wit of the EJ boards to write a more formal description of the general case. Then you just plug in the estimated flurry uptime for a given critrate, and we should have an analytical model we can compare to the simulation.

draghkar · 08/10/07, 11:45 AM

Update.

Code is now hosted by google subversion, so if you want take a look at the code:

wowcombatsim - Google Code

To check out it use:

svn checkout http://wowcombatsim.googlecode.com/svn/trunk/ wowcombatsim-read-only

- Fixed the wrong haste calculation
- I'm debuggin atm DW simulation capability

The last version generates this graph:

Weapon DPS: 100
X-Axis: Weapon Speed
Y-AXis: DPS obtained
Simulation Lenght: 1 milion seconds

the discontinuities suggested in post #171 by sogster seems are in the right place now.

draghkar · 08/10/07, 10:23 PM

Update.

First DW simulation seems working. This is what I obtained.

X-axis: Main Hand Speed
Y-axis: Off Hand Speed
Each Color represent the DPS obtained
Both weapons are 100 DPS
Crit% = 10

During the night I will run the simulation for 0%, 5%, 10%, 15%, 20%, 25%, 30%, 33% crit for a time longer
than previous graph ( I hope some ripples go away ).

Does that graph makes sense to you ?

songster · 08/29/07, 6:33 PM

Nice graph - can you do a simple two-dimensional graph plotting the (average/2) of the two weapon speeds against WF benefit, with lines for different crit rates?

I think you'll end up with a situation exactly mirroring the discontinuities in the single-weapon situation. That is, two weapons at 3s speed is effectively the same as one weapon at 1.5s speed. Two weapons at 2.4s speed is the same as one weapon at 1.2s speed. A 2.6 weapon and a 1.4 weapon is the same as one 1s weapon.

Reasoning is quite simple: with two 3s weapons you get 2 hits every 3 second - one every 1.5 seconds on average. Unless your weapons are exactly the same speed, they'll be going in and out of phase with each other, so the effect of syncronisation should be minimal. Even if they're exactly the same speed, I don't think it'll have much bearing on the final outcome. All that matters is how many hits are ineligible due to the internal cooldown.

So actually, I think the first "stepped" graph will be a more or less accurate rendition of WF effects.

songster · 08/29/07, 7:05 PM

Originally Posted by songster

To aid debugging your code, here's where you expect to see the discontinuities. I've listed all the "jumps" that fall within the range [0.5, 4]. Obviously you'd have to run the simulator for many hours to see them all clearly. As it stands, everything below about 1.5s is getting blurred out due to statistical fluctuations and/or the resolution of your ticker.

WITH NO CRIT/FLURRY
3s
1.5s
1s
0.75s
0.6s
0.5s

UNDER THE INFLUENCE OF FLURRY
3.9s
1.95s
1.3s
0.975s
0.78s
0.65s
0.557s

WHEN 0 < FLURRY UPTIME < 100%
3.9s
3s
1.95s
1.5s
1.3s
1s
0.975s
0.78s
0.75s
0.65s
0.6s
0.557s
0.5s

I'm taking this set of discontinuities as correct within the parameters of the model, since Draghkar says they accurately match what he gets.

If I'm right in the above post, and that you can model dual-wielding as a single weapon of half the speed, then it actually looks as though the effect of weapon speed on DW shaman itemisation should be less than people fear.

Two 2.9 speed weapons gives you 2 attacks every 2.9 seconds, or 1 every 1.45 seconds on average. Two 2.0 speed weapons gives you 2 attacks every 2 seconds, or 1 every 1 second on average.

If you look at the list of transition points above, there's only one "bump" in between 1s and 1.45s. So the difference in WF benefit when going from two 2.9s weapons to two 2.0s weapons is not huge. The relevant transition is the one at 1.3 seconds. Note also that this is one of the "secondary" transitions caused by flurry uptime, and thus will be a comparatively small bump unless your crit rate is high.

What this says is that if your combined attack speed is slower than 1.3 seconds, you'll get slightly more benefit than if your combined attack speed is below 1.3 seconds. Note that this is the attack speed before considerations of flurry etc. - that's accounted for later in the model.

Dual wielding 2.6 speed weapons should give you approximately the same WF benefit as dual-wielding 2.9 speed weapons. Going faster than 2.6s will lower the benefit of WF, unless the other weapon is correspondingly slower: for instance a 2.8s and a 2.4s weapon should work out OK.

We can see this beautifully on that colour graph Draghkar made. The only part of that we're really interested in is the part from (2.0 , 2.0) to (2.9 , 2.9). That graph was made with only 10% crit rate, so the transition at a net attack speed of 1.3s is slight. As you can see, the colour for dual wielding 2.9s weapons is very similar to the colour for dual wielding 2.6 speed weapons, or even dual 2.0 speed weapons.

What would really help is some empirical data to compare this against. Do we have any parses from DW shammies autoattacking against BL mobs with different speed weapons?

songster · 08/30/07, 5:33 AM

Oof, ignore the above two posts - I thought WF had the same mechanics as sword spec in that OH WF procs would trigger MH hits. Since that's not the case, you'll of course get much worse results if you have a fast OH constantly resetting the cooldown and preventing MH procs.