SimulationCraft Model Development - Page 15

Septhaka · 06/05/09, 6:53 AM

I searched the thread for this question but was not able to locate it. I noticed that Rogue results incorporate Tricks of the Trade (TotT) buffs while other classes do not. I was curious the results are computed with only Rogues receiving TotT? To my knowledge, a Rogue is unable to cast TotT on himself. Therefore, the simulation results assume the Rogue being tested has another Rogue in the raid and that other Rogue is casting TotT on the tested Rogue. Other classes do not incorporate this assumption. Yet, a Fury Warrior, a Ret Paladin, or any other DPS class could also just as easily be the target of a Rogue's TotT. The results seem to skew results in favor of Rogues by making available a buff to them that is not made available to other classes in other simulation results even though the buff in question, TotT, is equally available to all classes.

dedmonwakeen · 06/05/09, 8:01 AM

Originally Posted by Septhaka

I searched the thread for this question but was not able to locate it. I noticed that Rogue results incorporate Tricks of the Trade (TotT) buffs while other classes do not. I was curious the results are computed with only Rogues receiving TotT? To my knowledge, a Rogue is unable to cast TotT on himself. Therefore, the simulation results assume the Rogue being tested has another Rogue in the raid and that other Rogue is casting TotT on the tested Rogue. Other classes do not incorporate this assumption. Yet, a Fury Warrior, a Ret Paladin, or any other DPS class could also just as easily be the target of a Rogue's TotT. The results seem to skew results in favor of Rogues by making available a buff to them that is not made available to other classes in other simulation results even though the buff in question, TotT, is equally available to all classes.

I waffle back and forth on issues like this......

When you run a single Rogue through the sim, he does not (by default) get the benefit of TotT. I do allow the Rogue to cast it on himself if desired because I want to model both the benefit and the energy cost.

In the Raid_T8 config the Rogues perform a "circle jerk" of TotT casting. I chose to keep the buffs to just Rogues in this example because I'm trying to measure the full value that they bring. I do the same with the Focus Magic buff: Only those mages with the talent get to be in the "circle jerk".

Of course, the sim allows you to select ANY target for TotT or to turn if off entirely. The Raid_T8 is just an example...... one in which people sometimes put a bit too much weight....

nuoHep · 06/08/09, 2:56 AM

In reality though, on most of Ulduar fights rogues (well, for my guild that is) actually have to TotT the tank.

minx69 · 06/08/09, 8:24 AM

Would love to see some numbers for a 20/51/0 mage that hasnt got to keep up scorch/have the scorch glyph for comparison.

am having the "Lock V Mage" crit debuff argument and without comparable data its hard to work out who should be doing it.

dedmonwakeen · 06/09/09, 7:21 AM

Originally Posted by minx69

Would love to see some numbers for a 20/51/0 mage that hasnt got to keep up scorch/have the scorch glyph for comparison.

am having the "Lock V Mage" crit debuff argument and without comparable data its hard to work out who should be doing it.

I would love to be able to upload wiki files!!! It has been a week now and the GoogleCode folks still have not resolved it.

Kalle · 06/09/09, 2:59 PM

I wrote a small addon that queries mana and time with UnitPower() and GetTime() API functions to generate a timeline for the pet's mana. I have a value for mana each 0.1s (approximately). I emptied the pet's manapool and recorded from there until its mana was full again.

I tested mana regeneration of imp, felhunter, succubus and felguard in this way. The three melee pets seem to use the same formula, the imp uses a different one. All pets use the five second rule mechanic; I derived different formulas for the mana regeneration in the first five seconds after a cast and for the time after five seconds. There is no simple relationship between the regeneration in and out of the 5SR, the regeneration out of the 5SR depends on the pet's spirit, the in 5SR regeneration does not.

I first tested the scaling of mana regeneration with the pet's intellect by swapping out different items. I found that their mana regeneration increases linearly with their intellect. Since pets don't benefit from their master's spirit, I had to use spirit increasing buffs to test spirit scaling. This was the point where it started to get complicated, because some buffs were taken into account for computing mana regeneration but others were not.

I will show the intellect sclaing with different buffs on the pets. If you're only interested in the results, scroll down to the bottom. MPS = mana gained per second. Uncertainties are in the last digit for all values.

Imp, in 5SR

No buffs:		MPS = 0.2001 * int - 73.7  = 0.2001 * (int - 368)
3/3 Fel Vitality:	MPS = 0.1753 * int - 74.4  = 0.1753 * (int - 424)
Scroll of Spirit VII:	MPS = 0.2002 * int - 73.9  = 0.2002 * (int - 369)
Scroll of Spirit VIII:	MPS = 0.2006 * int - 74.0  = 0.2006 * (int - 369)
Blessing of Kings:	MPS = 0.1800 * int - 72.5  = 0.1800 * (int - 403)
Mark of the Wild:	MPS = 0.2002 * int - 84.15 = 0.2002 * (int - 420)
Divine Spirit:		MPS = 0.2010 * int - 74.2  = 0.2010 * (int - 369)
MotW+Divine Spirit:	MPS = 0.1997 * int - 83.7  = 0.1997 * (int - 419)

The increase per intellect is more or less constant except for fel vitality and BoK. 369 intellect is the base value for the imp, so we most likely have 0.2*(int-369). Let's take a closer look at Fel Vitality: 0.1753 ~ 0.2/1.15 and 424 ~ 369*1.15 - the talent seems to be neglected in the calculation. BoK and MotW too: 0.1800 ~ 0.2/1.1, 403 ~ 369*1.1; 420/419 ~ 369+51 (improved MotW).

Succubus, in 5SR

No buffs:		MPS = 0.0676  * int - 10.3  = 0.0676  * (int - 152)
3/3 Fel Vitality:	MPS = 0.0583  * int - 10.10 = 0.0583  * (int - 173)
Arcane Intellect:	MPS = 0.06649 * int - 13.89 = 0.06649 * (int - 209)
Scroll of Spirit VII:	MPS = 0.0667  * int - 10.0  = 0.0667  * (int - 150)
Scroll of Spirit VIII:	MPS = 0.0669  * int - 10.04 = 0.0669  * (int - 150)
Divine Spirit:		MPS = 0.06672 * int - 10.01 = 0.06672 * (int - 150)

Base intellect for the succubus is 150. Arcane Intellect is not taken into account (209 ~ 150+60), neither is Fel Vitality (0.0583 ~ 0.0667/1.15, 173 ~ 150*1.15). MPS per int may be 1/15.

Felhunter, in 5SR

No buffs:		MPS = 0.0667 * int - 9.97  = 0.0667 * (int - 150)
3/3 Fel Vitality:	MPS = 0.0581 * int - 10.00 = 0.0581 * (int - 172)
Scroll of Spirit VII:	MPS = 0.0669 * int - 10.08 = 0.0669 * (int - 151)
Scroll of Spirit VIII:	MPS = 0.0667 * int - 9.97  = 0.0667 * (int - 149)
Fel intelligence:	MPS = 0.0667 * int - 13.17 = 0.0667 * (int - 197)

Base intellect is again 150. Fel Vitality as above. At least the intellect portion of fel intelligence is neglected (197 ~ 150+48). Looks like the same formula used for succubus.

Felguard, in 5SR

No buffs:		MPS = 0.0669 * int - 10.04 = 0.0669 * (int - 150)
3/3 Fel Vitality:	MPS = 0.0582 * int - 10.00 = 0.0582 * (int - 172)
Scroll of Spirit VII:	MPS = 0.0668 * int - 10.02 = 0.0668 * (int - 150)
Scroll of Spirit VIII:	MPS = 0.0667 * int - 9.98  = 0.0667 * (int - 150)

Base intellect is again 150. Looks like the same formula used for succubus and felhunter.

Imp, out of 5SR

No buffs:		MPS = 0.2282 * int - 60.4 = 0.2282 * (int - 265)
3/3 Fel Vitality:	MPS = 0.201  * int - 59.6 = 0.201  * (int - 297)
Scroll of Spirit VII:	MPS = 0.2310 * int - 58.9 = 0.2310 * (int - 255)
Scroll of Spirit VIII:	MPS = 0.2329 * int - 58.0 = 0.2329 * (int - 249)
Blessing of Kings:	MPS = 0.209  * int - 56.9 = 0.209  * (int - 272)
Mark of the Wild:	MPS = 0.2294 * int - 67.4 = 0.2294 * (int - 294)
Divine Spirit:		MPS = 0.2337 * int - 57.3 = 0.2337 * (int - 245)
MotW+Divine Spirit:	MPS = 0.235  * int - 64.2 = 0.235  * (int - 273)

Fel Vitality, BoK and MotW look strange again. Both parameters seem to be influenced by spirit. The spirit from BoK and MotW may be taken into account, the MPS per intellect is slightly higher for MotW+Divine Spirit than for just Divine Spirit.

Succubus, out of 5SR

No buffs:		MPS = 0.088  * int - 4.6 = 0.088  * (int - 52)
3/3 Fel Vitality:	MPS = 0.0755 * int - 3.0 = 0.0755 * (int - 40)
Arcane Intellect:	MPS = 0.0858 * int - 7.7 = 0.0858 * (int - 90)
Scroll of Spirit VII:	MPS = 0.092  * int - 3.2 = 0.092  * (int - 35)
Scroll of Spirit VIII:	MPS = 0.0940 * int - 2.5 = 0.0940 * (int - 26)
Divine Spirit:		MPS = 0.096  * int - 2.1 = 0.096  * (int - 22)

Fel Vitality and Arcane intellect look strange. Dependence on spirit.

Felhunter, out of 5SR

No buffs:		MPS = 0.0877 * int - 4.2 = 0.0877 * (int - 48)
3/3 Fel Fitality:	MPS = 0.0779 * int - 3.9 = 0.0779 * (int - 50)
Scroll of Spirit VII:	MPS = 0.092  * int - 3.2 = 0.092  * (int - 35)
Scroll of Spirit VIII:	MPS = 0.0940 * int - 2.5 = 0.0940 * (int - 26)
Fel Intelligence:	MPS = 0.0911 * int - 4.8 = 0.0911 * (int - 52)

Values with fel intelligence are not what they should be. As MPS per int is higher than without buffs, the spirit increase is probably used in the formula. Fel Vitality looks strange again.

Felgueard, out of 5SR

No buffs:		MPS = 0.0877 * int - 4.3 = 0.0877 * (int - 49)
3/3 Fel Vitality:	MPS = 0.071  * int - 1.2 = 0.071  * (int - 17)
Scroll of Spirit VII:	MPS = 0.092  * int - 3.2 = 0.092  * (int - 35)
Scroll of Spirit VIII:	MPS = 0.0939 * int - 2.5 = 0.0939 * (int - 26)

Fel Vitality. Rest looks like the formula used for succubus and felhunter.

I think this is enough to say, that mana regeneration out of the five second rule depends on spirit. When looking at the values for the regeneration out of the 5SR (leaving out those from neglected buffs), I see a pretty much perfect plane, so thats what I'm going to assume. There seems to be some curvature, but the deviation from the plane is very small (< 1 MPS for the range I tested). I wasn't able to recognize any parameters that can be derived from the pet's stats in this case.

To summarize:
Mana regeneration in the 5SR:
MPS = 0.2 * (int - 369) for imps and
MPS = 1/15 * (int - 150) for melee pets
(where "int" is the pet's intellect without Fel Vitality, Blessing of Kings, Mark of the Wild, Fel Intelligence and Arcane Intellect).

Mana regeneration out of the 5SR:
MPS = 0.2308 * int + 0.070 * spi - 87.3 for imps and
MPS = 0.0911 * int + 0.053 * spi - 16.3 for melee pets
((where "int" is the pet's intellect without Fel Vitality, Blessing of Kings, Mark of the Wild, Fel Intelligence and Arcane Intellect and "spi" is its spirit including all buffs).

Note: The above formula doesn't really explain the values I got with Fel Vitality, Arcane Intellect, Fel Intelligence, Mark of the Wild or Blessing of Kings. It also doesn't explain them if you only count the increase of spirit and neglect the increase of intellect. The difference in MPS isn't that large though.

dedmonwakeen · 06/09/09, 5:04 PM

Kalle, thanks for the meticulous detail! We'll work on getting this implemented before next release.

• Zakalwe · 06/13/09, 3:55 PM

When you did these tests with buffs, were the buffs present only on the pet, or on the master as well?

I'm asking because it seems like the discrepancies with the buffs and with fel vitality might be explained by the MP5 scaling directly with the master's intellect, not with the pet's own intellect.

Kalle · 06/14/09, 5:48 AM

I didn't pay attention. I instructed my buffers to buff the pet, but I remember that they sometimes buffed me, too.

I assumed that the regeneration was derived from the pet's stats because I did some tests with spi/int scrolls that I didn't mention. A scroll of spirit on the master didn't influence the pet's regeneration, a scroll of intellect on the pet did. For the scroll of spirit values in the post above I only buffed the pet.

Fizzl · 06/15/09, 8:58 AM

Does anyone have a good way to model Aura of Celerity? (+20% haste).

dedmonwakeen · 06/21/09, 11:54 PM

WARNING! WARNING! WARNING!

Any folks out there living on the ragged edge building simcaft execs straight from the latest commits: There has been a fairly substantial incompatible change to profiles. Please examine the default configs to catch the changes. Mage_T8_20_51_00.simcraft is a very good example.

A new release is imminent, including some info on Armory downloads among several other new interesting features.....

Erdluf · 07/09/09, 2:20 PM

I'd like to bring up the old topic of the latency model (discussed on first couple of pages of this thread).

From one of the globals.simcraft files:

# Lag is a random value (gaussian distribution) centered on the value below with a stddev of 25%
# The type of lag is dependent upon the previous action and whether or not the player is
# able to take advantage of the spell queuing system, has to time a multi-tap on GCD, or
# has to wait for a channel to finish.
queue_lag=0.075
gcd_lag=0.150
channel_lag=0.250

The code in sc_player.cpp seems to be consistent with this quote. Ignoring channeled spells, SC uses an average lag of .075 for casts that are >= to the GCD and .150 for casts that are < GCD.

For a nominal 1.5s cast (= GCD), which has been hasted to 1.01s, the average SC spam-cast time will be 1.085s. Add a bit more haste so that the theoretical cast time is 0.99s, and the SC average spam-cast time will actually jump to 1.14s. In this neighborhood, SC Haste scaling is negative.

I believe the actual mechanics are closer to:

1) You can't queue a spell during the first second of the GCD. This gives you a Not Ready message.

2) You can't queue a spell while your previous spell has more than about 0.3s left to complete. This gives you an "Another action is in progress" message.

3) You can queue a spell during the GCD as long as you satisfy both (1) and (2) above.

If I'm correct you should not see a jump in cast times as haste increases. Rather, you will reach a point where the value of haste drops to zero, but it won't have a neighborhood of negative value.

Moonkin formulation tools typically implement this as something like

delta_time = max(tooltip_cast_time+queue_lag, gcd+queue_lag, 1.0+gcd_lag)

Here is some limited testing that suggests that instants can be queued, when the GCD is > 1s, and that queueing breaks down when the gcd is 1s and cast time is < 1s. (Note, at the time of this test, Nature's Grace was 0.5s off of cast times, rather than the current 20% spell haste, and Wrath's GCD under Nature's Grace was 1s).

Old moonkin tests

Some more recent tests by Skjaven seem to show that for a spell at the GCD, queueing is less effective when the GCD is down to 1s.

dedmonwakeen · 07/09/09, 3:12 PM

When generating scale factors I had to change the latency code to allow queuing on non-zero cast times that went below the min gcd. It seems as if this should be the default behavior.....

dedmonwakeen · 07/09/09, 4:19 PM

SimulationCraft r2808 available for download.

There is considerable amount of new functionality that has been unannounced (and undocumented ofc). I suppose I should let more people in on the fun. A big thanks goes out to Nenad for doing the initial exploratory work in most of these areas:

(1) combat_log=filename This will generate a Blizzard style combat log that can be loaded up into WWS-like tools. While there is the remote possibility for abuse, the generated combat logs will never withstand any real scrutiny.

(2) smooth_scale_factors=1 An enormous amount of research was done in the area of not-so-random numbers. This option is on by default. It does not effect the normal dps reporting, but it does affect scale factor calculation by creating a new "smooth" baseline for comparison with stat-delta runs. The end result are reliable scale factors that can be calculated in 10x to 100x fewer iterations. (Essentially, we allow our "absolute" accuracy to drop while vastly improving our "relative" accuracy given stat changes.)

(3) raid_events=... It is now possible to simulate raid events such as movement, stuns, (in)vulnerability, splash damage. The splash damage will trigger effects like Incanters Absorbtion. When moving, the player is still able to perform instant actions, etc. You can see opt_raid_mov.simcraft for some examples. More will be forthcoming.

(4) armory=region,server,name It is now possible to download your character directly from the armory. Since the sim defaults to max_time=300 and optimal_raid=1 AND determines optimal spell priorities based upon talents, to simulate your char you just need: simcraft.exe armory=us,Llane,Segv

(5) wowhead=region,server,name OR wowhead=id It is now possible to download your character from wowhead in a manner similar to the armory. You can also download a virtual profile via wowhead=id. This allows you to copy your char, change talents/gear/glyphs/etc, and then see how it affects DPS: simcraft.exe wowhead=123456

(6) save=filename OR save_profiles=1 It is now possible to save a character to a config file after download. This config file will contain everything necessary to run. It will also include an "origin=" link to show where it came from. We will be using this feature extensively: We plan to put all our default configs into the wowhead profiler. Our "golden" data will be online and we will just re-generate the scripts for distribution.

Of course, all the usual modeling stuff should be up-to-date, including PTR mechanics which can be accessed via patch=3.2.0 on the cmd line.

I'll re-generate the wiki pages sometime tonight when I can steal some server time.

Enjoy!

PS: Erdluf: This recent version has the change you requested.

Erdluf · 07/09/09, 10:48 PM

Originally Posted by dedmonwakeen

(4) armory=region,server,name It is now possible to download your character directly from the armory. Since the sim defaults to max_time=300 and optimal_raid=1 AND determines optimal spell priorities based upon talents, to simulate your char you just need: simcraft.exe armory=us,Llane,Segv

This worked well for me (except not wearing PvE gear in Armory). If running from a windows cmd window, use a '+' to replace a space in the server name: armory=us,Echo+Isles,Erdluf

I'll take a closer look at the numbers.

Thanks.