Jump to content


Photo

SimulationCraft for Hunters (Panda edition)


  • Please log in to reply
63 replies to this topic

#1 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 22 July 2012 - 12:37 PM

This thread is for discussion of the hunter module for SimulationCraft, including issues, features, action lists to describe particular dps strategies, etc. This post automatically updates to reflect the latest results at simulationcraft.org. Click on the charts below for details:

http://www.simulatio...hart.php?Hunter

Watch the changelog for updates to the SimC hunter model. It may take a few hours for new changes to be reflected in the results.

#2 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 22 July 2012 - 12:48 PM

The hunter module in simc has undergone a LOT of work due to the large amount of mechanics changes in MoP. In the initial release, all level 85 abilities are (re)implemented, and the default profiles have been adjusted for the expertise cap. T14 and several level 86-90 abilities are implemented; the remaining ones are in progress.

We incorporate changes as we hear about them; the ability characterization, bug reporting, and action-list development from the community has been invaluable. Please keep it up and let us know about any inaccuracies.

Edit: no Lynx Rush. That's going to be after Barrage.

#3 Adarant

Adarant

    Glass Joe

  • Members
  • 4 posts

Posted 24 July 2012 - 02:10 PM

If I am reading your results correctly, all three specs are using Thrill of the Hunt. I have two questions about that choice.

First question, how many free Arcane Shots are you providing when Thrill procs (I assume that you won't use Multi-Shot in a single target sim)? I believe that the talent provides just two free Arcane Shots, but your results show that it can stack to 3. Also, I see that there is some uptime for the buff thrill_of_the_hunt_3, which indicates to me that you might be stacking the buff too high.

Second question, why did you choose Thrill of the Hunt for that talent tier instead of Dire Beast or Fervor? I fully understand if the answer is, "that's all that I have implemented," but I am curious about the comparison between all three talents. After this latest round of changes to the tier, I feel that the talents are well balanced but small differences will surely arise. For example, I think the Focus provided to our pets with Fervor will cause that talent to be best for BM, but your sim uses Thrill of the Hunt. Is there a way to customize this selection, or do you report just the best talent?

Similarly, I am curious about the Blink Strike, Murder of Crows, Lynx Rush tier. I realize that Lynx Rush is not yet implemented, but is it possible to see a comparison between Blink Strike and Murder of Crows?

#4 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 24 July 2012 - 06:26 PM

If I am reading your results correctly, all three specs are using Thrill of the Hunt. I have two questions about that choice.

First question, how many free Arcane Shots are you providing when Thrill procs (I assume that you won't use Multi-Shot in a single target sim)? ...

Second question, why did you choose Thrill of the Hunt for that talent tier instead of Dire Beast or Fervor? I fully understand if the answer is, "that's all that I have implemented," but I am curious about the comparison between all three talents. After this latest round of changes to the tier, I feel that the talents are well balanced but small differences will surely arise. For example, I think the Focus provided to our pets with Fervor will cause that talent to be best for BM, but your sim uses Thrill of the Hunt. Is there a way to customize this selection, or do you report just the best talent?

Similarly, I am curious about the Blink Strike, Murder of Crows, Lynx Rush tier. I realize that Lynx Rush is not yet implemented, but is it possible to see a comparison between Blink Strike and Murder of Crows?


I'm so glad you asked! Little if any optimization has been done yet, and some of the default action list configuration right now is to enable testing. I originally had TotH as the default talent for that tier because it was the only one in that tier that was substantially different than in Cata (e.g., recall that Readiness used to be there instead of Dire Beast). It was also quite good before the recent change. Now it's still the default because it's the most problematic from a priority point of view: the action list needs to decide where to prioritize taking a free arcane shot. For example, even a free arcane shot is not worth delaying Murder of Crows, Chimera Shot, etc. for. My impression is that the current TotH is extremely weak, but actually verifying that will indeed require comparison and playing with the action list. I think the reason for the high uptime on the TotH in the current sims is that the current default priority list simply has other things to do than spend a precious GCD on a fairly weak ability.

If you look at the TotH ability on wowhead, it is supposed to provide 3 stacks currently. They've already indicated they intend to change TotH further (to reduce the cost by 20), again changing it's priority.

Your latter questions are the important ones, and are best answered by downloading the tool at the above link and trying it out. The character files contain a spec, a talent set, and the default action list. We rely on people in the community to download the tool and try different combinations. As people post improved talent and action list strategies, I'll incorporate them as the defaults. To do comparisons, copy the file, change the name, and then run a single raid with both parties (that's how the current hunter comparison is generated).

Note that the T13 hunter is level 85, so the last talent selection will be irrelevant. To actually play with strategy details, copy an existing file and modify it (make sure not to use "default_action=1" or your new action list will be ignored). My level 90 example is still in progress; feel free to start from it for your own experiments:
Spoiler
The action_list entries are evaluated in priority order, so you can change them and try different strategies. To compare the barrage profile (which isn't actually very smart about when to use barrage) with powershot:
  • copy it
  • give the hunter a new name (e.g. hunter="Hunter_BM_T14_powershot")
  • make the last talent digit 2 (since powershot is the second entry on the 6th tier of talents)
  • delete the "action_list+=/barrage" line
  • add a line with "/powershot" wherever in the priority you want it to be
  • run simc with all files for everyone you want to compare
There's of course much more flexibility in simc than I can explain here; more information is available at the links above. Once I get to optimization, I'll also post those here, of course.

#5 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 24 July 2012 - 11:59 PM

In comparing Zeherah's numbers for explosive shot, I realized that I hadn't moved Explosive Shot to the "ignite" mechanics as described in Frostheim's post. I've made an initial fix to Explosive Shot that does partially close the gap between SV and MM/BM. Currently it is modeled as an initial impact with two dot ticks (that use ignite mechanics). As currently implemented, the ticks individually crit. Three additional mechanics questions need to be answered from logs:
  • Does an impact just extend the dot with 3 ticks or is there a tick on impact as currently modeled?
  • Do the ticks crit individually or is the crit computed on impact and distributed?
  • If Curse of Elements or some such is applied to the target between ticks, do later ticks deal more damage?
While the answers won't have a significant dps impact in the simple case, having a trinket proc from the first tick could have a non-trivial impact on later ticks.

Edit: Zakalwe and I did some testing. For item #1, the jury is still out. The event sequences look a little like overlapping dots. This may just be vagaries of event reporting, however.

For item #2, all the damage comes out as SPELL_PERIODIC_DAMAGE events. Each tick can individually crit. The most telling sequence was: 8286, refresh, 4118, 4117, 4117, refresh, 4108, 4108, refresh 5438, 5438, 5437, 10875. The 8k and 10k events are tick crits.

For #3, Zak dropped a CoE on the dummy after one tick of an ES, and we got 690, 725, and 725. Therefore, the target debuffs are considered at each tick.

Finally, in trying to verify that it followed ignite semantics, we found an anomaly. 8181, refresh, 8150, 4075, 4075, refresh, 4101, 4101, 4101, 4102, simply ignoring crits to determine the ES damage leads to shots of 4090, 2706, and 4109. 2706 isn't possible given the relatively small variation in the ES damage formula. But if the entire value of the crit is subtracted from the remaining damage when the refresh happens, then you end up with 4090, 4070, and 4109, all of which are plausible damage values. We still need more logs to characterize this, but it looks like it doesn't recompute the base damage until there's a refresh (thus a crit on the first tick won't reduce later ticks, but will reduce the carry-over to overlapping shots). If this is the case, then it would reduce the actual damage of ES, and reduce the value of crit for ES.

Edit 2: Blizzard said that the current misbehavior is a bug. I'll save further changes for explosive shot till after the next build.

#6 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 29 July 2012 - 07:20 PM

The results presented in the first post of this thread now are for T14H profiles. Most of the new build changes are in, with still-naive integration of new talents and spells into the profiles. There are still some inconsistencies with female dwarf and expected results that we are working through. The "BIS" gear profile still has Cata trinkets; that will be updated once the new trinkets are implemented in both simc and FD. The default equipment profile is:
Spoiler


#7 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 19 August 2012 - 12:15 AM

Simc (in svn) should now be up to date with build 16004. The Dire Beast change is just a tooltip bug. After another round of comparison, Zeherah spotted bugs in Lock and Load's ICD and the 4T14H bonus. Those are now fixed. The resulting dps (improved gearsets, legendary gems, etc.) is:

109154 Hunter_BM_T14H
104740 Hunter_SV_T14H
100619 Hunter_MM_T14H

Bug reports and profiles welcome! (Lord Blastington's scope still needs updating).

#8 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 21 August 2012 - 06:22 AM

The ranged vs. melee haste logic is not correct, which affects all specs. Among other things, ranged-only haste being inherited by the pet, and focus fire is not granting the pet stacks*6 extra focus. I'm still straightening it out, but it will be at least another day. Thus take the hunter sim numbers with an extra grain of salt. This was provoked when someone reported to Zeherah that dire beast is now affected by haste. I think the net effect will be to bring BM down and the other specs up, but that intuition is guided by possibly buggy interim results.

Edit: The issue is now fixed. It indeed brings the three specs (at least with their current profiles) closer together:

108,271 BM
105,392 SV
102,745 MM (without incorporating Whytfysts suggestions).

#9 Nerec

Nerec

    Von Kaiser

  • Members
  • 39 posts

Posted 22 August 2012 - 06:22 AM

Nothing great, but can you get rid of the gcd used to switch into aspect of the hawk in the beginning of every sim? Warrior as Hunter will start in there main "stance/aspect" and dont use a gcd infight to change into it...

#10 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 23 August 2012 - 07:50 PM

Nothing great, but can you get rid of the gcd used to switch into aspect of the hawk in the beginning of every sim? Warrior as Hunter will start in there main "stance/aspect" and dont use a gcd infight to change into it...


In a debug log for just hunters, the activation doesn't take any time. Thus, you can do hunter-only testing (e.g., different CA sequences) without an issue. I believe that the simulation is implemented so that combat only begins at the first harmful action. I'll check later whether it's getting caught with a GCD in a raid with other classes or some such, and make sure the GCD doesn't get spent for it at the beginning of combat. Thanks for the report.

#11 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 03 September 2012 - 07:20 AM

Currently, beta support requires setting the "ptr" flag. Thus, the T14 normal gear run can be invoked in the appropriate profile directory with:
>[B] simc ptr=1 Hunter_T14N.simc[/B]

 84610  34.0%  Hunter_BM_T14N
 82273  33.1%  Hunter_SV_T14N
 81920  32.9%  Hunter_MM_T14N

As an example for analyzing the careful aim period, here is an invocation to run the sim and produce results just for that period. This also outputs the result into the local file ca.html for more in depth review.
> [B]simc ptr=1 target_death_pct=90 max_time=50 Hunter_T14N.simc html=ca.html[/B]

139834  35.3%  Hunter_BM_T14N
137498  34.7%  Hunter_MM_T14N
119350  30.1%  Hunter_SV_T14N

The "Action Priority List" section for each spec entry shows both the action list used and an example action sequence. Numerous further details can be explored with the log. To focus down onto just MM, we use the MM profile only:

> [B]simc ptr=1 target_death_pct=90 max_time=50 log=1 Hunter_MM_T14N.simc > ca.log[/B]
The log itself has lots of information, including focus gains, pet activity, buffs applying and dropping, etc. A surprising amount of analysis can be performed with simple text searches:
> [B]findstr /c:"Hunter_MM_T14N performs" ca.log[/B]

0.000 Hunter_MM_T14N performs summon_pet (100)
0.000 Hunter_MM_T14N performs trueshot_aura (100)
0.000 Hunter_MM_T14N performs snapshot_stats
0.000 Hunter_MM_T14N performs aspect_of_the_hawk (100)
0.000 Hunter_MM_T14N performs glaive_toss (100)
0.000 Hunter_MM_T14N performs wild_quiver_shot (85)
1.040 Hunter_MM_T14N performs lynx_rush (90)
2.080 Hunter_MM_T14N performs dire_beast (94)
2.423 Hunter_MM_T14N performs ranged (100)
2.423 Hunter_MM_T14N performs wild_quiver_shot (100)
3.106 Hunter_MM_T14N performs rapid_fire (100)
3.106 Hunter_MM_T14N performs stampede (100)
4.144 Hunter_MM_T14N performs readiness (100)
4.144 Hunter_MM_T14N performs glaive_toss (100)
4.846 Hunter_MM_T14N performs ranged (91)
4.846 Hunter_MM_T14N performs wild_quiver_shot (91)
5.185 Hunter_MM_T14N performs lynx_rush (92)
6.229 Hunter_MM_T14N performs dire_beast (100)
6.443 Hunter_MM_T14N performs ranged (100)
8.527 Hunter_MM_T14N performs aimed_shot (100)
8.527 Hunter_MM_T14N performs wild_quiver_shot (50)...


The number at the beginning is the time of the action (time 0 is special in that any number of non-harmful actions can be packed into there, on the assumption that you do them before combat). The number in parentheses is the amount of focus at the time the ability is performed. The wild_quiver and ranged entries are not explicitly fired by the hunter, of course. Even the little snippet above shows that the default action list is sitting at focus cap for much of the first 10 seconds of the fight. The spoiler below filters the ranged and wild_quivers so that it can show the entire sequence of actions for the CA phase.
Spoiler


When I first did this to start writing this message, rapid fire was after stampede, resulting in initially lower pet focus regeneration. The additional regerneration can make the difference in whether each pet can afford one more basic attack during its stampede. That's provides a noticeable dps improvement. I also incidentally noticed that simc was still implementing the old form of the master marksman buff in which an aimed shot in progress woudl also be free. That's fixed in game and now fixed in simc. Those changes and fixes led to the above action sequence. There are clearly remaining issues: There are a surprising number of unpaired steady shots. There are also several places during which focus caps. I think these are related. The unpaired seady shots are to add enough focus for an aimed shot. However because it does not then complete the pair, it requires a full pair of shots later to keep teh buff up. What other issues might this indicate. The action priority list is below, whee the talents are dire beast, lync rush, and glaive; all talents comments (e.g., /powershot) for unavailable talents will be ignored:
Spoiler


#12 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 06 September 2012 - 07:10 AM

There's been some discussion on Wow Hunter Hall and thegrumpyelf about the number of new buttons to press for dps. I was noting a similar thing and looking at optimizing the "DPEB" with appropriate macroing (e.g., Rabid goes well with Lynx Rush, Readiness could be sequenced with bestial wrath, etc.). But to motivate actually skipping some abilities altogether, we need more information. Inspired by the above posts, here's the simc results of turnign off some individual abilities. I didn't do any other action list changes other than suppressing the particular abilities, but I'm surprised by the results. The results below are made using the corresponding T14N profiles; these have constraints like "no professions, neutral races". I used simc's "modify_action to suppress a single line of the profile's action list; e.g., "copy=BM_no_BW modify_action=bestial_wrath,if=0" makes a copy of the default BM profile with bestial wrath turned off (because 0 is never true). The command line to invoke the profile was:
simc ptr=1 compare.simc html=compare.html
using the profile:
Spoiler
The results are:

Name | DPS | Delta
Hunter_BM_T14N | 84,800 | 0.0%
BM_no_Re | 82,625 | -2.6%
BM_no_KS | 82,624 | -2.6%
BM_no_ST | 82,623 | -2.6%
Hunter_SV_T14N | 82,210 | 0.0%
Hunter_MM_T14N | 81,914 | 0.0%
BM_no_GT | 81,717 | -3.6%
BM_no_BW | 81,125 | -4.3%
SV_no_ST | 80,652 | -1.9%
SV_no_Re | 80,493 | -2.1%
BM_no_LR | 80,481 | -5.1%
SV_no_KS | 80,307 | -2.3%
MM_no_ST | 80,200 | -2.1%
MM_no_KS | 80,015 | -2.3%
MM_no_Re | 79,921 | -2.4%
SV_no_GT | 79,561 | -3.2%
SV_no_LR | 79,557 | -3.2%
MM_no_SrS | 79,541 | -2.9%
MM_no_GT | 79,399 | -3.1%
MM_no_LR | 79,263 | -3.2%
BM_no_KC | 78,821 | -7.1%
BM_no_DB | 78,583 | -7.3%
SV_no_AS | 77,550 | -5.7%
MM_no_AS | 77,511 | -5.4%
MM_no_DB | 77,203 | -5.8%
SV_no_DB | 76,933 | -6.4%
SV_no_ES | 76,088 | -7.4%
BM_no_AS | 75,714 | -10.7%
BM_no_SrS | 73,223 | -13.7%
SV_no_SrS | 71,582 | -12.9%

I included the profile above so you can locally generate the html file to verify any particular version. Zeherah is graciously hosting the result of this run on her beta site.

Since these do not otherwise include copying and modifying the action lists, I didn't bother with removing aimed (e.g., since the action list is designed around using aimed in the CA phase). I spot checked the various scenarios (e.g., the no ST really doesn't trot out Stampede). I'm particularly surprised at how small the impact of readiness is ("Re"). That may be an artifact of poor Readiness usage, of course (run an individual profile with log=1 to look at the precise timing of it).

Edit: I since learned that this kind of test is more directly supported than I thought. I updated the profile above to use "copy=SV_no_ES skip_actions=explosive_shot" to suppress a "/" separated list of actions, which woudl make testing combos easy (e.g., no Glaive or Lynx Rush).

#13 Whitefyst

Whitefyst

    Great Tiger

  • Members
  • 771 posts

Posted 06 September 2012 - 03:31 PM

CA Phase post...


Concerning your CA phase post and the shot order in the spoiler, I suggest disabling GT during the CA phase to increase the MM DPS you are simming (with realizing that I have never used SimC myself to test these out but am working from theorycrafting and FD observations):

Skip GT during the CA phase for the following reasons:
- It is not a worthy replacement for AI from a damage standpoint. AIs with 100% crit, 100% PS, and highly hasted does a lot more damage than GT. Each AI does about twice the damage as GT on a single target.
- With AI being hasted over the CA phase, AI cast time should be around 1.45s with RF only and without the T14 4P.
Thus, every 3 GTs cost you about 2 AIs from a GCD usage.
- Although GT only costs 15 focus, that is 15 focus taken away from AI casts. About every 3 GTs cost you an AI in focus.
- Removing GT means that less RF hasted GCDs are used on instants instead of cast costs. You can cast your first AI in the sim about 2 GCDs earlier with only using 2 GCDs after the RF before the first AI instead of 3 GCDs.

At my request, Zeherah has added an option to FD to disable GT during CA. Doing so is about a 400 DPS improvement in her sim.

Another item to check is how the second use of Readiness if being used. In the FD sims, Readiness becomes available about 2s before the 4th RF. It is better DPS to delay that Readiness use by a few seconds until after the RF is cast so that you can get an extra RF. Also, this Readiness in FD is being used a few seconds before MoC and a DB. Its is better DPS to delay it a few more seconds until after their uses so their CDs are reset too. If your second use of Readiness is being used too early without resetting key CDs, then your sim is definitely undervaluing it.

EDIT on Readiness use. Silly me, the reason some of the abilities were used shortly after Readiness is because Readiness reset their CDs. Relooking at the data, when Readiness is available the second time, there is still about 80s left on RF CD where it makes no sense to wait on it to use Readiness in most situations. FD is still not ideal in its used about 8s into CS 9s CD, where delaying it a couple seconds for back to back CSs would be an improvement. It was also used about 27s into DB's 30s CD. Finally, it was used about 7s into GT's 15s CD, but I am not as worried about that one.

Concerning combining keys, the obvious one for me is Stampede and RF. In the vast majority of cases, you do not want to use Stampede unless it is being hasted, ideally by RF, for maximum DPS. If you are using RF off CD with a little bit of a delay in the initial cast due to casting instants first so that they will reset with Readiness, the second Stampede becomes available about 4 to 5s before the 4th RF. Waiting 4 to 5s to cast Stampede due to combining the buttons is not a big DPS deal. Since the next Stampede is not available uintil over 10 mins into the fight, you will not get to it for most fights.

If you combine ST with RF, then the only additional keys left for MM (and the other specs) are DB, MoC/LR, and GT. These do complicate things a little, but this is alleviated some by their uses ideally just replacing an AS in your "rotation/priority".

Concerning the single SSs during the CA phase with dropping SF sometimes and overcapping focus, you may want to check out how FD implements the situation. In the sim I do during the CA phase, single SSs casts are never repeated and focus does not overcaps a lot less. Then again that scenario was using BL at the start too since that is currently the maximum DPS case for MMs - stacking both RF and BL during the CA phase. Even with removing the BL at the start, FD has a much cleaner usage of AI and SS. Part of this is that I require for an SS pair to be performed if there is less than 4s left on the SF. Hence, if 6 seconds have past since the last SS pair, a new one will be cast. It looks like in both cases 11 SSs are cast before the second RF expires, but the FD case never loses the SF focus buff. (Note that in my FD sim I was using MoC instead of LR.)

One question for my edification since I have not yet played BM in 5.0.4 and really checked it out, but what makes not casting SrS such as big DPS loss for BM? I am not aware of anything that makes SrS so important to BM. It is important to SV, but I am surprised how much of a loss it is for SV.

#14 Rivkah

Rivkah

    Great Tiger

  • Members
  • 838 posts

Posted 06 September 2012 - 05:11 PM

One question for my edification since I have not yet played BM in 5.0.4 and really checked it out, but what makes not casting SrS such as big DPS loss for BM? I am not aware of anything that makes SrS so important to BM. It is important to SV, but I am surprised how much of a loss it is for SV.


I suspect there is something else going on with sting that having it missing in the profiles is screwing up the rotation. When I have more time later I'll try to see if I can spot it. The damage loss isn't even close to that amount when simmed on my site, and although simc and my site work a bit differently, I don't think it makes any sense for both BM and SV to lose the same large amount when serpent sting is removed either. Sting is worth a sizeable amount for SV given improved serpent sting's multiplier, the focus regen, and the mastery bonus, but it still shouldn't be that large.

#15 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 07 September 2012 - 12:40 AM

I suspect there is something else going on with sting that having it missing in the profiles is screwing up the rotation...


We do end up with less uptime on Lord Blastington's scope, but the real problem is that arcane shot is also suppressed. The recently added line to ensure SrS stays up gets triggered if serpent sting is not up at all, and so it was always trumping Arcane. So the BM_no_SRS number is really "without arcane and serpent sting". SrS is 2600dps in all the other sims, so that's approximately the real loss for that scenario.

#16 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 14 October 2012 - 04:39 AM

TL;DR is that the change to Rabid does seem to only affect burst and not sustained dps. "Rabid has been changed and now increases a pet's attack speed by 70%. Pet attack power is now unaffected by this ability". I hacked in the change to test it; it's added to the code base, but will need to be updated once the client (and therefore the spell data) is updated. The current profiles are all using Lynx Rush, which had a nice synergy with Rabid. On the theory that it loses such a synergy, I also compared with Murder of Crows. The basic T14N results are:

Spec|AP + LR|Spd + LR|Spd + MoC
BM|86,453|86108|85,233
SV|84,207|84,454|84,648
MM|83,545|83,812|83,884


So basically, the differences for T14N are in the noise. In another fit of tuning I'll probably switch SV to use MoC, but it remains a fight and style-specific question which talent would actually be better in practice.

#17 AnotherTrollHunter

AnotherTrollHunter

    Glass Joe

  • Members
  • 9 posts

Posted 14 October 2012 - 12:18 PM

TL;DR is that the change to Rabid does seem to only affect burst and not sustained dps. "Rabid has been changed and now increases a pet's attack speed by 70%. Pet attack power is now unaffected by this ability". I hacked in the change to test it; it's added to the code base, but will need to be updated once the client (and therefore the spell data) is updated. The current profiles are all using Lynx Rush, which had a nice synergy with Rabid. On the theory that it loses such a synergy, I also compared with Murder of Crows. The basic T14N results are:

Spec|AP + LR|Spd + LR|Spd + MoC
BM|86,453|86108|85,233
SV|84,207|84,454|84,648
MM|83,545|83,812|83,884


So basically, the differences for T14N are in the noise. In another fit of tuning I'll probably switch SV to use MoC, but it remains a fight and style-specific question which talent would actually be better in practice.


How does Simcraft simulate Rabid? Ist it averaged over the course of a single fight like Femaledwarf does or is it used like ingame where it gets the full benefit for the duration of the buff? And did you have priorities to delay the old version of Rabid so it lined up with trinket procs or stuff like agility potion?

#18 Lokrick

Lokrick

    Piston Honda

  • Members
  • 121 posts

Posted 14 October 2012 - 06:07 PM

How does Simcraft simulate Rabid? Ist it averaged over the course of a single fight like Femaledwarf does or is it used like ingame where it gets the full benefit for the duration of the buff? And did you have priorities to delay the old version of Rabid so it lined up with trinket procs or stuff like agility potion?


SimulationCraft is an event simulator, so it simulates Rabid directly; i.e., my change replaces the code that applied an AP buff to the pet with code that adds an attack speed buff. The existing action list does not make any special effort to line up rabid with procs. When I played with a few variations during beta, it was a loss because: the 2min cd didn't line up well with the ~110 second interval of Terror in the Mists. Similarly, lining it up with the ~70 second interval of Relic of Xuen is possible, but has a high probability of losing a rabid, and stacking AP buff with an Agi buff isn't very high synergy. The intervals I mention here come from the buff results in the automated simulation run. Syncing up controlled buffs with longer CDs to those procs (e.g., delaying Rapid Fire for a Relic of Xuen proc) should have better synergy.

The current action list (see the above link) has a few entries that coordinate between abilities, but it does not contain any gear-specific ones; the default action list is generic across a wide range of gear. Proposing profiles with specific gear/talent/action combinations (with simulation results to back them up) is however eminently suitable for this thread. I'll integrate gear-specific actions into the default action list generation if they are very common or significant enough.

Edit: as an example, a quick comparison of changing the rapid_fire line for SV to "actions+=/rapid_fire,if=!buff.rapid_fire.up&buff.relic_of_xuen.up" results in a 5% drop in rapid fire uptime and so loses 500dps.

#19 AnotherTrollHunter

AnotherTrollHunter

    Glass Joe

  • Members
  • 9 posts

Posted 14 October 2012 - 07:42 PM

SimulationCraft is an event simulator, so it simulates Rabid directly; i.e., my change replaces the code that applied an AP buff to the pet with code that adds an attack speed buff. The existing action list does not make any special effort to line up rabid with procs. When I played with a few variations during beta, it was a loss because: the 2min cd didn't line up well with the ~110 second interval of Terror in the Mists. Similarly, lining it up with the ~70 second interval of Relic of Xuen is possible, but has a high probability of losing a rabid, and stacking AP buff with an Agi buff isn't very high synergy. The intervals I mention here come from the buff results in the automated simulation run. Syncing up controlled buffs with longer CDs to those procs (e.g., delaying Rapid Fire for a Relic of Xuen proc) should have better synergy.

The current action list (see the above link) has a few entries that coordinate between abilities, but it does not contain any gear-specific ones; the default action list is generic across a wide range of gear. Proposing profiles with specific gear/talent/action combinations (with simulation results to back them up) is however eminently suitable for this thread. I'll integrate gear-specific actions into the default action list generation if they are very common or significant enough.

Edit: as an example, a quick comparison of changing the rapid_fire line for SV to "actions+=/rapid_fire,if=!buff.rapid_fire.up&buff.relic_of_xuen.up" results in a 5% drop in rapid fire uptime and so loses 500dps.

I don't really know how to read your string but if it means that you only use Rapid Fire if you have Relic of Xuen buff up, then that is an awfully unoptimised condition. I was thinking about only waiting for cooldowns if the internal cooldowns of trinkets and stuff is almost expired and you would not lose another cooldown by waiting. That's a little more complicated.

#20 NoGoal

NoGoal

    Von Kaiser

  • Members
  • 61 posts

Posted 15 October 2012 - 09:55 AM

TL;DR is that the change to Rabid does seem to only affect burst and not sustained dps. "Rabid has been changed and now increases a pet's attack speed by 70%. Pet attack power is now unaffected by this ability". I hacked in the change to test it; it's added to the code base, but will need to be updated once the client (and therefore the spell data) is updated. The current profiles are all using Lynx Rush, which had a nice synergy with Rabid. On the theory that it loses such a synergy, I also compared with Murder of Crows. The basic T14N results are:

Spec|AP + LR|Spd + LR|Spd + MoC
BM|86,453|86108|85,233
SV|84,207|84,454|84,648
MM|83,545|83,812|83,884


So basically, the differences for T14N are in the noise. In another fit of tuning I'll probably switch SV to use MoC, but it remains a fight and style-specific question which talent would actually be better in practice.


Is there any reason as to why simulationcraft doesn't apply BM mastery to MoC?

FemaleDwarf does and Ghostcrawler said it should.

The DPS I get from it in game is also higher than the DPS simulated while my gear is lower.




0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users