SimulationCraft for Rogues - Page 17

dedmonwakeen · 03/28/09, 5:34 PM

Originally Posted by Thargood

I just noticed a bug in the implementation of the new 3.1 PPM poison increase from the envenom buff. I was comparing the output of equivalent rogue specs between T7 and T8, and noticed that the IP was applying far more often in the 3.1 model than 3.0.9.

The PPM is being multiplied by the envenom modifier of +75%, instead of adding it. This results in IP having a 52.5% chance per hit for a 1.4 speed weapon instead of the correct 45% chance.

Thanks for the catch. This is fixed in version r1976. I'll put out another release later tonight.

EDIT: New binary available for download. SampleOutput and SampleOutputPTR wiki pages have been updated.

Sarah · 03/31/09, 7:05 AM

I guess this concerns scaling factors and hard caps in general, but I'm using expertise rating as an example here.

I'm not sure that I interpreted the code correctly, but I think that the scaling factor for expertise rating is calculated in the following way:

scaling_factor_for_expertise_rating = (dps(exp_p) - dps(exp_x))/(exp_p - exp_x)
where:
exp_p denotes the actual expertise rating listed in the config file.
exp_x is some other value, with exp_x < exp_p.
dps(y) is the dps from a simulation using y expertise rating.
I think that the scaling factors are also normalized as a group in some way, but that doesn't really matter here.

My point is that exp_x should be close enough to exp_p, so that when exp_p is clearly over the expertise rating cap, the scaling factor for expertise rating ends up being ~0. Right now this isn't the case, which leads me to believe that currently, exp_x is a very low number (I'm not sure what the current implementation actually uses as value for exp_x, maybe zero).

While that isn't wrong in general, it can give some confusing results. For example, when going from 217 to 317 expertise rating, simulations show that the DPS gain is 0, but the expertise rating scaling factor is still far from zero (~0.76 at 317 expertise rating). In other words, the effect of the expertise rating cap is currently not accurately represented in the expertise rating scaling factor.

To fix this, I think exp_x should be chosen close enough to exp_p (within 5-10 rating?). That way, not only will the effect of the expertise rating cap be more accurately represented, but also the calculated scaling factor would become more useful as a whole, since it would better represent the actual effect that adding/removing expertise rating has, assuming a certain amount of expertise rating from gear.

dedmonwakeen · 04/02/09, 2:26 PM

Originally Posted by Sarah

To fix this, I think exp_x should be chosen close enough to exp_p (within 5-10 rating?). That way, not only will the effect of the expertise rating cap be more accurately represented, but also the calculated scaling factor would become more useful as a whole, since it would better represent the actual effect that adding/removing expertise rating has, assuming a certain amount of expertise rating from gear.

Unfortunately, you run into the #1 drawback of using RNG simulation: Variance

Run the sim once, and it will give you a "right" result.
Run the sim again, and it will give you a different "right" result.

The smaller the delta, the larger the number of iterations....... To use a rating delta of 5-10 rating, it would require an immense amount of cpu-time before the variance in results drops below the benefit/loss of such a small delta.

This is one of the strongest reasons for using formulation to determine weights for various stats.

dedmonwakeen · 04/02/09, 4:24 PM

SimulationCraft now uses a 2-roll system for all special attacks.

While performing this conversion, I noticed that the HAT watcher was triggering off of "may_glance" to determine if an attack was from a "special". The problem with this is that Hunter ranged auto-shot has may_glance=false....... which resulted in too many HAT procs in the Raid_T7 and Raid_T8 configs in which there are two hunters in the party with the HAT Rogues.

To atone for my mistake, here is some analysis that I have done.....

I became curious about the effectiveness of each class in "donating" CPs to HAT Rogues so I added a "honor_among_thieves_donor" proc. Classes with pets have both player and pet data rolled into the owner's report.

I created a party including everyone in the Raid, filtered out the "honor_among_thieves_donor" proc, and then sorted them by frequency. The data in the following list is the number of procs over the course of the fight and the average time between procs. (Both numbers are themselves averages of many simulation iterations.)

honor_among_thieves_donor=151.9|1.89sec    Hunter_T8_53_13_05
honor_among_thieves_donor=148.2|1.94sec    Hunter_T8_06_14_51
honor_among_thieves_donor=114.6|2.51sec    Mage_T8_57_03_11
honor_among_thieves_donor=112.2|2.56sec    Mage_T8_53_18_00
honor_among_thieves_donor=110.8|2.60sec    Mage_T8_18_00_53
honor_among_thieves_donor=109.6|2.63sec    Warlock_T8_00_40_31
honor_among_thieves_donor=109.0|2.64sec    Hunter_T8_15_51_05
honor_among_thieves_donor=104.7|2.75sec    Shaman_T8_57_14_00
honor_among_thieves_donor=098.1|2.94sec    Warlock_T8_03_13_55
honor_among_thieves_donor=097.6|2.95sec    Shaman_T8_16_55_00
honor_among_thieves_donor=090.8|3.17sec    Druid_T8_58_00_13_SF
honor_among_thieves_donor=090.5|3.18sec    Druid_T8_58_00_13_W
honor_among_thieves_donor=087.2|3.30sec    Mage_T8_20_51_00
honor_among_thieves_donor=083.3|3.46sec    Mage_T8_00_53_18
honor_among_thieves_donor=078.1|3.69sec    Priest_T8_13_00_58
honor_among_thieves_donor=076.6|3.70sec    Rogue_T8_08_20_43
honor_among_thieves_donor=073.0|3.95sec    Warlock_T8_00_41_30
honor_among_thieves_donor=071.8|4.01sec    Rogue_T8_07_51_13
honor_among_thieves_donor=070.1|4.10sec    Warlock_T8_03_52_16
honor_among_thieves_donor=068.8|4.18sec    Warlock_T8_00_56_15
honor_among_thieves_donor=068.2|4.22sec    Rogue_T8_18_51_02_FD
honor_among_thieves_donor=068.2|4.16sec    Rogue_T8_23_05_43
honor_among_thieves_donor=067.3|4.28sec    Rogue_T8_51_13_07
honor_among_thieves_donor=059.2|4.87sec    Rogue_T8_18_51_02_S
honor_among_thieves_donor=059.0|4.88sec    Druid_T8_00_55_16_M
honor_among_thieves_donor=056.2|5.13sec    Druid_T8_00_55_16
honor_among_thieves_donor=052.8|5.46sec    Rogue_T8_51_18_02
honor_among_thieves_donor=041.9|6.88sec    Warlock_T8_53_00_18

Note that in order to properly calculate this, I needed to make sure there was only one HAT Rogue in the raid, and his "donor" rate was ignored. The values for the two HAT Rogues were taken from a generic Raid_T8 profile so they would not be skewed by the massive influx of HAT procs given the 30-player party.

For reference, the T8 raid profile uses the following party setups for the HAT Rogues:

# Since honor_among_thieves_interval=0 the HAT Rogues must be in a party
party=Rogue_T8_08_20_43/Hunter_T8_06_14_51/Hunter_T8_15_51_05/Mage_T8_57_03_11/Mage_T8_18_00_53
party=Rogue_T8_23_05_43/Hunter_T8_53_13_05/Warlock_T8_00_40_31/Mage_T8_53_18_00/Shaman_T8_57_14_00

Mavanas · 04/02/09, 6:06 PM

Originally Posted by dedmonwakeen

SimulationCraft now uses a 2-roll system for all special attacks.

While performing this conversion, I noticed that the HAT watcher was triggering off of "may_glance" to determine if an attack was from a "special". The problem with this is that Hunter ranged auto-shot has may_glance=false....... which resulted in too many HAT procs in the Raid_T7 and Raid_T8 configs in which there are two hunters in the party with the HAT Rogues.

To atone for my mistake, here is some analysis that I have done.....

I became curious about the effectiveness of each class in "donating" CPs to HAT Rogues so I added a "honor_among_thieves_donor" proc. Classes with pets have both player and pet data rolled into the owner's report.

I created a party including everyone in the Raid, filtered out the "honor_among_thieves_donor" proc, and then sorted them by frequency. The data in the following list is the number of procs over the course of the fight and the average time between procs. (Both numbers are themselves averages of many simulation iterations.)

honor_among_thieves_donor=151.9|1.89sec    Hunter_T8_53_13_05
honor_among_thieves_donor=148.2|1.94sec    Hunter_T8_06_14_51
honor_among_thieves_donor=114.6|2.51sec    Mage_T8_57_03_11
honor_among_thieves_donor=112.2|2.56sec    Mage_T8_53_18_00
honor_among_thieves_donor=110.8|2.60sec    Mage_T8_18_00_53
honor_among_thieves_donor=109.6|2.63sec    Warlock_T8_00_40_31
honor_among_thieves_donor=109.0|2.64sec    Hunter_T8_15_51_05
honor_among_thieves_donor=104.7|2.75sec    Shaman_T8_57_14_00
honor_among_thieves_donor=098.1|2.94sec    Warlock_T8_03_13_55
honor_among_thieves_donor=097.6|2.95sec    Shaman_T8_16_55_00
honor_among_thieves_donor=090.8|3.17sec    Druid_T8_58_00_13_SF
honor_among_thieves_donor=090.5|3.18sec    Druid_T8_58_00_13_W
honor_among_thieves_donor=087.2|3.30sec    Mage_T8_20_51_00
honor_among_thieves_donor=083.3|3.46sec    Mage_T8_00_53_18
honor_among_thieves_donor=078.1|3.69sec    Priest_T8_13_00_58
honor_among_thieves_donor=076.6|3.70sec    Rogue_T8_08_20_43
honor_among_thieves_donor=073.0|3.95sec    Warlock_T8_00_41_30
honor_among_thieves_donor=071.8|4.01sec    Rogue_T8_07_51_13
honor_among_thieves_donor=070.1|4.10sec    Warlock_T8_03_52_16
honor_among_thieves_donor=068.8|4.18sec    Warlock_T8_00_56_15
honor_among_thieves_donor=068.2|4.22sec    Rogue_T8_18_51_02_FD
honor_among_thieves_donor=068.2|4.16sec    Rogue_T8_23_05_43
honor_among_thieves_donor=067.3|4.28sec    Rogue_T8_51_13_07
honor_among_thieves_donor=059.2|4.87sec    Rogue_T8_18_51_02_S
honor_among_thieves_donor=059.0|4.88sec    Druid_T8_00_55_16_M
honor_among_thieves_donor=056.2|5.13sec    Druid_T8_00_55_16
honor_among_thieves_donor=052.8|5.46sec    Rogue_T8_51_18_02
honor_among_thieves_donor=041.9|6.88sec    Warlock_T8_53_00_18

Note that in order to properly calculate this, I needed to make sure there was only one HAT Rogue in the raid, and his "donor" rate was ignored. The values for the two HAT Rogues were taken from a generic Raid_T8 profile so they would not be skewed by the massive influx of HAT procs given the 30-player party.

For reference, the T8 raid profile uses the following party setups for the HAT Rogues:

# Since honor_among_thieves_interval=0 the HAT Rogues must be in a party
party=Rogue_T8_08_20_43/Hunter_T8_06_14_51/Hunter_T8_15_51_05/Mage_T8_57_03_11/Mage_T8_18_00_53
party=Rogue_T8_23_05_43/Hunter_T8_53_13_05/Warlock_T8_00_40_31/Mage_T8_53_18_00/Shaman_T8_57_14_00

Most of these results line up with empirically tested rates posted in the HaT thread except for shadow priest. 0.27 crit per second is about twice the HaT rate of a shadow priest. Make sure mind flay crits do not contribute to HaT procs.

dedmonwakeen · 04/02/09, 6:49 PM

Originally Posted by Mavanas

Most of these results line up with empirically tested rates posted in the HaT thread except for shadow priest. 0.27 crit per second is about twice the HaT rate of a shadow priest. Make sure mind flay crits do not contribute to HaT procs.

Ah, I didn't know that about Shadow Priests.

Mind Flay has much in common with Arcane Missiles. Can AM ticks proc HAT?

Mavanas · 04/02/09, 11:47 PM

Yeah AM can. Post 775 on this page has a nice summary of abilities that were confirmed to proc HaT. You can add Starfall for druids to it.

Edit: Sorry forgot the link. Here it is http://elitistjerks.com/f78/t38187-f...tlety_you/p31/

tedoubledy · 04/06/09, 9:02 AM

Anyone have any new numbers off the SIM?

Loot · 04/06/09, 9:07 AM

SampleOutputPTR - simulationcraft - Google Code ?

• Maestroquark · 04/06/09, 10:00 AM

Originally Posted by Loot

SampleOutputPTR - simulationcraft - Google Code ?

Note that Sample Output must be run separately to be updated. In this case, it hasn't been updated since 4/02 (Simcraft build 1998) and many changes have gone in since then (build 2052 is current).

dedmonwakeen · 04/06/09, 10:08 AM

Originally Posted by Maestroquark

Note that Sample Output must be run separately to be updated. In this case, it hasn't been updated since 4/02 (Simcraft build 1998) and many changes have gone in since then (build 2052 is current).

The bulk of those changes have been from Starfox who has been working (apparently without eating or sleeping or even blinking) on the Warrior module.

Still.... there -have- been some key changes. I was holding off until I cleaned up the HAT mechanics (which requires some non-trivial work on Searing Totem and Arcane Missiles). My real coding job has been keeping me up past 1am every night, so that puts a cramp in my play-time.

Expect an update in the next couple of days.

dedmonwakeen · 11/22/09, 11:14 PM

While we manage to keep the core Rogue module current with recent changes, the default profiles tend to lag significantly. If someone is interested in maintaining (or just contributing to) the default profiles, please let me know. This is not just a BiS list, but rather the set of interesting talent/glyph/ability combinations.

Akaron · 03/08/10, 5:29 AM

how to add the rupture ito the combat cycle?
(need that for 3.3.3)

dedmonwakeen · 03/08/10, 9:30 AM

Originally Posted by Akaron

how to add the rupture ito the combat cycle?
(need that for 3.3.3)

By default, the code will only include Rupture if you have a talent/glyph that explicitly improves Rupture. Otherwise, it will assume that you want to skip it.

For now you have to add a line like the following in an appropriate spot in the priority list.

actions+=/rupture,if=buff.combo_points.stack=5&target.time_to_die>10

Of course, you will likely have to change the logic on some of the other lines to tune your energy/cp management.

I'll be putting out a release shortly that will turn Rupture back on always if patch=3.3.3.

Malkom72 · 06/20/10, 3:17 PM

How I can delete Rupture into the combat cycle ??

Where I can find the line : "rupture,if=buff.combo_points.stack>=4&target.time_to_die>15&time>10&buff.slice_and_d ice.remains>11" to delete it ??

I can't find the file....

Thx