Rawr (v2.2.27 released on 11/9) - Page 41

Bashui · 08/05/08, 5:14 PM

Originally Posted by Promethius

I think the calculation for mangle uptime should be changed as well. Proc Mechanics shows trinket proc modeling and I think it would be more accurate to model Mangle as if it were a trinket with a 12 sec duration, 6 sec internal cooldown, with a chance to proc equal to the chance to land the mangle. The current source code uses the trinket model of assuming Mangle has no cooldown.

As I mentioned over there, the formula was not chosen arbitrarily, it was determined using a simulator. If you CAREFULLY read the thread you link to you will notice that the equation you are trying to use is for a proc that CANNOT refresh itself, which is clearly not the case for mangle. If you read further you will see the discussion for the derivation of an exact result. It is pretty simple though, when you think about it. The chance that mangle is down at any random time is EXACLY equal to the chance that you missed the last two mangles (assuming 0 latency etc.)

Originally Posted by Promethius

float averageWhiteDamage = 1 - chanceAvoided - chanceCrit + critMultiplier * chanceCrit;
float averageYellowDamage = (1 - chanceAvoided) * (1 + critMultiplier * chanceCrit);

I don't know what you are trying to get at with averageYellowDamage, but it makes no sense whatsoever. Let us try to plug in some numbers:
1-.2-.4+.4*2.33 = 0.866 vs
(1-.2)*(1-2.33*.4) = .0544
I hope there is a typo in what you are trying to get across, because I know my average yellow hit does more than 5% damage

Yellow attacks are not able to glance. I suppose that there is technically an error with extreme crit for white attacks, since I am converting hits to glances.

Originally Posted by Promethius

(.077f * stats.AttackPower + 92) looks like it should be (.0714f * stats.AttackPower + 84) for Swipe

You could be right for the values for swipe, wowwiki does say 84 and I find it interesting that 92 is ~ 84*1.1 (though I compute the naturalist bonus separately). I have no problem changing it, but I don't want to change something based on someone's feelings, but rather on someone's research.

Gruntle · 08/05/08, 5:32 PM

Feedback on Rawr DPS(Arms)warr.

It's looking pretty good I must say, I haven't discovered any major bugs. It gives reasonable results as well. Some things to consider for the next version:
- Weapon specializations. It seems they are not included in the current version. Should be pretty easy to implement.
- Some more information on how the dps cycle looks like. I.e. if there is not enough rage to do white-slam-instant, what is the priority? Or how is the specials damage modeled? What happens when swing speed go below 2.5 sec (making a normal slam cycle impossible)?

WOWTigra · 08/05/08, 5:35 PM

good stuff! I've been playing with the Moonkin and Tree models and noticed some things...

Moonkin:

1. Gemming choices per the optimizer are a bit strange in places. i.e. with t6 shoulders, chest and legs equipped Rawr chose to enforce the gemming socket colors on the chest and ignore them on the shoulders and legs (putting +12 spell dmg in the shoulders and legs instead but matching colors on the chest for the +5 spell damage bonus. Now, this is a really small thing, and probably explained by what order Rawr parses the gemming choices in, but if it had reversed that, ignoring the socket colors in the chest, but honoring them in the shoulders and legs, it would have used the exact same gems over all 6 slots, but earned 1 more +dmg from the shoulder and legs socket bonuses.

2. Haste, Gemming & Ashtongue Trinket - I'm well over the theorycrafted breakpoint for gemming haste (around 1450 spell damage raid buffed), but Rawr was optimizing for a lot of +12 spell damage gems over haste/spell dmg and not choosing items with haste. i.e. selecting a Brute Cloak of the Ogre Magi (crit) over Shadowcaster's Drape (haste) when both were marked available. The setup it made put me at 129 spell haste. Then, it chose the Ashtongue Trinket (which I though really required a great deal of haste to make even a possibility) over Sorc's alchemist stone, or over other highly regarded trinkets - DMC, ISC, Timbal's. I'm kind of confused as to why it selected Ashtongue Trinket - or is 129 spell haste above the point at which an Ashtongue trinket becomes useful?

Tree:

1. Same problem as others have been having, you really have to do a lot of manipulation to get it to stop valuing mp5 quite so heavily - it was even suggesting t5 gloves over t6 (no set bonuses in play), I think because if the 2 more mp5 on t5.

2. Head slot items - Powerheal 9000 (and Powerheal 4000) are not able to be crafted by druids, as far as I know. I think we only have the option of creating the leather engineering hats.

Darknebula · 08/05/08, 6:12 PM

Originally Posted by Darknebula

Just downloaded the latest version, and i'm not sure if i'm doing it wrong, but when i select that i have the Mana Attuned Band for Finger 1, it automatically assumes i can use it for Finger 2 as well ie: that i have 2 of this item where i only have the 1 of them, is there anything i can do to specify that i only have one of this item.

It would appear that i'm an idiot.. apologies i was referring to the Ring of Ancient Knowledge, not a unique item, but rawr assumes i have two of them if i select it for either Finger for the optimiser.

• Astrylian · 08/05/08, 6:38 PM

Re Powerheal 4000/9000 showing up for druids... Right click on them and hit Refresh Item Data. Do they disappear? I think Dwin fixed it to automatically add those class requirements when loading those items from the armory, but I may not have refreshed them in the default item cache, sorry about that. Let me know if that fixes it.

Re Ring of Ancient Knowledge... Yeah, we don't really have a way to specify that an item is non-unique, but I only have one of them. This is another situation where the easy fix is just to make it so users can set uniqueness in the item editor. In the mean time, just edit the itemcache.xml in notepad, and find that item and make it unique (search for 'unique' to find the line you need to add for unique items, it should be pretty obvious). Be sure to make a backup in case you screw up your item cache, and let me know if you don't figure that out easily enough. Thanks!

halmmar · 08/05/08, 6:44 PM

Originally Posted by Astrylian

3- We'd really love to provide filtering/sorting by item source, but there just isn't the data available to support it yet. Wowhead doesn't expose their data, so we only have the Armory to work with, whose item source data is flaky, and imposing filtering based on that would open up tons of inconsistency, bugs, issues, etc.

Parsing the item name and checking for Vindicator's, Vengeful Gladiator's, etc would be simple enough (but only for pvp items).

--

@LineNoise, WOWTigra: Check the relative stat values for more accurate results regarding gems. 10 spirit will not always help your healing rotation since it will probably still use the exact same one. Averaging over several gems provides better results (Rawr kinda makes an assumption that the values are linear, at least for small deviations of a stat, but this is not at all the case for Rawr.Tree, where stat increases cause jumps in final values).
(For my gear, [Quick Lionseye] is valued at 5 HPS, but 9 gems mean each is valued at 14 HPS)
The optimizer does a good job of finding decent gems, looking at Rawr's gem view does not.

T5 gloves is valued at 60 HPS less than T6 for me, so I guess your rotations or fight length is too high and it seems to favor T5. Would love to see the xml-file causing this behaviour.

Scurn · 08/05/08, 6:46 PM

Under the buff section there should be a checkbox that allows for the Onslaught 2pc prot bonus of 170 more health to commanding shout. I believe the same addition should be made for Solarian's Sapphire and it's AP bonus to Battle Shout.

• Astrylian · 08/05/08, 6:58 PM

Originally Posted by Scurn

Under the buff section there should be a checkbox that allows for the Onslaught 2pc prot bonus of 170 more health to commanding shout. I believe the same addition should be made for Solarian's Sapphire and it's AP bonus to Battle Shout.

Unfortunately, we don't have the framework in place for these buffs which are improved, other than by talents. I hope to improve the buff framework to allow for cases like this, in a future version, but I expect we won't see it until Rawr 2.0 (the version of Rawr we'll be branching off soon, to work on WotLK).

In the mean time, those buffs do at least work for the users of them in Rawr. (ie, you can't mark that you're getting someone elses' better commanding shout, but the protwarr will get it for themselves.)

[DRF]Solmyr · 08/05/08, 7:15 PM

Kavan,

Is there a F.A.Q. somewhere where I can learn what each checkbox in your options for rawr.Mage do in more detail? I am slow having trouble reading the output to the "sequencer." I assumed it would spit out ideal cooldown management for maximum DPS, but I'm getting weird errors, such as the HexShrunkenHead buff lasting for more than 30 seconds, and Molten Fury being in effect for the middle 65% of the fight.

Thanks again for all the hard work.

p.s. BTW, will you be continuing to maintain this through WOTLK? and do you already have BETA access?

Kavan · 08/05/08, 7:59 PM

Originally Posted by [DRF]Solmyr

Kavan,

Is there a F.A.Q. somewhere where I can learn what each checkbox in your options for rawr.Mage do in more detail? I am slow having trouble reading the output to the "sequencer." I assumed it would spit out ideal cooldown management for maximum DPS, but I'm getting weird errors, such as the HexShrunkenHead buff lasting for more than 30 seconds, and Molten Fury being in effect for the middle 65% of the fight.

Thanks again for all the hard work.

p.s. BTW, will you be continuing to maintain this through WOTLK? and do you already have BETA access?

You can read about the advanced options in Rawr.Mage here:

Rawr - Home

For sequence reconstruction it is strongly suggested to enable both cooldown fragmentation and integral mana consumables for display, especially for fire builds. If when mousing over sequence reconstruction you see a score at the very bottom (lower than 100) that is an indication that without the additional information it can get from those features it is unable to find a meaningful sequence that would correspond to the output of the optimizer.

If after you enable those two options the sequence you get is still bad, then that's a bug and you can open a new issue on codeplex and attach the problematic character and I'll take a look at it.

Also note that the advanced solver is a lot more complex and in many cases won't be fast enough to have it enabled while making changes to the character. So you should view the sequence reconstruction as something you would use once you've optimized your gear to see how to stack the spells/cooldowns in that situation.

PS: I just got into beta recently and I have Arcane changes mostly implemented. I haven't decided yet whether I'll go ahead with the other two trees or wait till the talents stabilize a bit.

Orsier · 08/06/08, 6:35 AM

Originally Posted by Astrylian

In the mean time, those buffs do at least work for the users of them in Rawr. (ie, you can't mark that you're getting someone elses' better commanding shout, but the protwarr will get it for themselves.)

This isn't true for Solarian's Saphire on DPSWarr model, the attack power stays unchanged when selecting it. It would be nice to actually model it.

Squishypants · 08/06/08, 11:09 AM

Originally Posted by Astrylian

Squishypants, right now we don't have any way for the optimizer to correctly use only one of a unique gem, so don't mark the 24healing gems as available, just the 22s. That may reduce how much is suggests using healing gems. Also, make sure you've completely filled out the buffs/enchants, and tweak the fight parameters as needed. Let me know if you aren't able to get the results you're looking for, and please send me the character file in question if so. Thanks for your feedback!

I tried again last night with 22 Healing gems available for optimizing as well as the other standard priest gems, nothing higher, and it still puts way too much emphasis on +healing, and puts them in every slot. You can tell that the weights are a little off just by looking at the gem list, it puts [Teardrop Crimson Spinel] well above [Purified Shadowsong Amethyst] and in most priests opinion, afaik, they are about even depending on if you are light or heavy on regen in the rest of your stats. It also sometimes puts [Royal Shadowsong Amethyst] in slots and those are never better for priests.

Also Haste is a very valuable stat if you're at a Sunwell level and have an acceptable level of regen/healing already. I would rate it a bit higher than it does but other priests may not, especially in T4/T5. Perhaps there should be an option to check which Tier of content you're gearing for and the weights adjusted accordingly? Or be able to set those stat values yourself?

Also, when I do an optimization it appears to equip an OH with a Staff but the OH is slightly greyed out, I assume that means it's not really counting the stats on that OH?

I really do like the tool so far though, I used to check gearsets with a website and it was very tedious to set up so this makes testing tweaks in my gear/gems much easier! I just think the optimization tool/stat weights need some work, can't wait to use future versions. =)

charriu · 08/06/08, 11:33 AM

Originally Posted by Squishypants

Also Haste is a very valuable stat if you're at a Sunwell level and have an acceptable level of regen/healing already. I would rate it a bit higher than it does but other priests may not, especially in T4/T5. Perhaps there should be an option to check which Tier of content you're gearing for and the weights adjusted accordingly? Or be able to set those stat values yourself?

Rawr doesn't use any kind of Equivalence Point system like you seem to be expecting. Instead, it calculates total throughput/damage/mitigation/avoidance, depending on the chosen model, and gives points for items according to their contribution to that total value (it rates the whole character without that item, then with it, and the difference is the value that item gets).

Drashian · 08/06/08, 12:14 PM

Originally Posted by Scurn

Under the buff section there should be a checkbox that allows for the Onslaught 2pc prot bonus of 170 more health to commanding shout. I believe the same addition should be made for Solarian's Sapphire and it's AP bonus to Battle Shout.

When I was looking to model the effects of having an arms warrior with the Sapphire in my group, I entered it into my BuffCache.xml as a separate buff that conflicted with Battle Shout. I counted on Commanding Presence also applying to the Sapphire bonus, which a quick search implied was the case. The XML:

  <Buff>
    <Name>Battle Shout</Name>
    <Category>ClassBuffs</Category>
    <Stats>
      <AttackPower>305</AttackPower>
    </Stats>
    <Type>LongDurationNoDW</Type>
    <ConflictingBuffs>
      <string>Battle Shout</string>
    </ConflictingBuffs>
    <SetThreshold>0</SetThreshold>
  </Buff>
  <Buff>
    <Name>Improved Battle Shout</Name>
    <Category>ClassBuffs</Category>
    <Stats>
      <AttackPower>76</AttackPower>
    </Stats>
    <Type>LongDurationNoDW</Type>
    <RequiredBuff>Battle Shout</RequiredBuff>
    <ConflictingBuffs />
    <SetThreshold>0</SetThreshold>
  </Buff>
  <Buff>
    <Name>Battle Shout (Sapphire)</Name>
    <Category>ClassBuffs</Category>
    <Stats>
      <AttackPower>375</AttackPower>
    </Stats>
    <Type>LongDurationNoDW</Type>
    <ConflictingBuffs>
      <string>Battle Shout</string>
    </ConflictingBuffs>
    <SetThreshold>0</SetThreshold>
  </Buff>
  <Buff>
    <Name>Improved Battle Shout (Sapphire)</Name>
    <Category>ClassBuffs</Category>
    <Stats>
      <AttackPower>93</AttackPower>
    </Stats>
    <Type>LongDurationNoDW</Type>
    <RequiredBuff>Battle Shout (Sapphire)</RequiredBuff>
    <ConflictingBuffs />
    <SetThreshold>0</SetThreshold>
  </Buff>

Oh, and since this is my first post here I'll take this opportunity to heap the appropriate amount of praise onto Astrylian and the other Rawr developers. I've been using the software since it was just for bears, and it's been basically the best thing to happen to my gear since 2T4. Thanks for all of this. And I downloaded Visual C# and the Rawr source recently, hopefully I'll be able to find some small place where I can leverage my own programming skills to help out too.

• Astrylian · 08/06/08, 12:20 PM

Squishypants, please post the character file where you see results you don't expect attached to a new Issue on our codeplex.com/Rawr site.. Regarding the staff+Offhand... Yeah, you can (and the optimizer does) select an offhand, but its stats aren't counted; instead they're counted when comparing 1H weapons (with the selected offhand) to 2H weapons. That way you don't just see all the staves as above all the swords/daggers, just cause 2Hs have tons more stats than 1Hs.

EDIT: Drashian, thanks for the xml for everybody, and for those kind words.

Dwin · 08/06/08, 1:51 PM

Originally Posted by Astrylian

Re Powerheal 4000/9000 showing up for druids... Right click on them and hit Refresh Item Data. Do they disappear? I think Dwin fixed it to automatically add those class requirements when loading those items from the armory, but I may not have refreshed them in the default item cache, sorry about that. Let me know if that fixes it.

I did add the ability to add a class requirement based on an itemId on an item update; however, I only added those for the druid/rogue/hunter goggles. That can be attributed to stupidity and I'll add the rest of the goggles tonight.

thedopefishlives · 08/06/08, 4:24 PM

Originally Posted by WOWTigra

good stuff! I've been playing with the Moonkin and Tree models and noticed some things...

Moonkin:

1. Gemming choices per the optimizer are a bit strange in places. i.e. with t6 shoulders, chest and legs equipped Rawr chose to enforce the gemming socket colors on the chest and ignore them on the shoulders and legs (putting +12 spell dmg in the shoulders and legs instead but matching colors on the chest for the +5 spell damage bonus. Now, this is a really small thing, and probably explained by what order Rawr parses the gemming choices in, but if it had reversed that, ignoring the socket colors in the chest, but honoring them in the shoulders and legs, it would have used the exact same gems over all 6 slots, but earned 1 more +dmg from the shoulder and legs socket bonuses.

2. Haste, Gemming & Ashtongue Trinket - I'm well over the theorycrafted breakpoint for gemming haste (around 1450 spell damage raid buffed), but Rawr was optimizing for a lot of +12 spell damage gems over haste/spell dmg and not choosing items with haste. i.e. selecting a Brute Cloak of the Ogre Magi (crit) over Shadowcaster's Drape (haste) when both were marked available. The setup it made put me at 129 spell haste. Then, it chose the Ashtongue Trinket (which I though really required a great deal of haste to make even a possibility) over Sorc's alchemist stone, or over other highly regarded trinkets - DMC, ISC, Timbal's. I'm kind of confused as to why it selected Ashtongue Trinket - or is 129 spell haste above the point at which an Ashtongue trinket becomes useful?

As for the optimizer, I'm not entirely certain how it works, so it's very possible that it didn't just see that combination. As for haste - this is something I'm looking at fixing _once and for all_ once I get the time. The current spell calculation algorithm is good, but it's far from perfect and I'll admit that. Haste is a real sore spot in my algorithm. The Ashtongue trinket shouldn't be that good. I'm going to re-evaluate trinket math when I get some more free time to work on Rawr.Moonkin, with an eye to improving the accuracy of proc-based effects like the Ashtongue. I've already laid some groundwork to that effect; now I just have to finish the job.

• Adoriele · 08/06/08, 7:51 PM

Originally Posted by thedopefishlives

As for the optimizer, I'm not entirely certain how it works, so it's very possible that it didn't just see that combination. As for haste - this is something I'm looking at fixing _once and for all_ once I get the time. The current spell calculation algorithm is good, but it's far from perfect and I'll admit that. Haste is a real sore spot in my algorithm. The Ashtongue trinket shouldn't be that good. I'm going to re-evaluate trinket math when I get some more free time to work on Rawr.Moonkin, with an eye to improving the accuracy of proc-based effects like the Ashtongue. I've already laid some groundwork to that effect; now I just have to finish the job.

I get the same issue with the optimizer some times. I have [Amani Divining Staff] and [Hammer of Judgement], and the optimizer insists on putting a blue gem in the Staff, then tells me it's far above the Hammer + [Fetish of the Primal Gods] because losing the blue gem loses the meta bonus, even though I have two other blue slots on other pieces of gear (one of which is mis-socketed). I've gotten in the habit of, once the Optimizer runs, checking that errors like that didn't get made with respect to where I put my blue gems in order to get the meta and socket bonuses.

As to why it's socketing so "little" haste for WOWTigra, he may be above the haste goodness threshold, but that's an equivalence point. Past that he'd want to gem equally for both Haste and SD. With 1450/129, it's actually more haste than it should be, not less (though 1450 may not be accurate. Parity for 129 haste would be around 1520 spell damage).

• Astrylian · 08/06/08, 8:03 PM

There is indeed a certain amount of randomness in the Optimizer. It tries quite alot of different gear sets and it gets smart about choosing things based on how well those things do in other gear sets, to try to find combos that work best. So when there's a slight gain from making a bunch of specific gem changes, it can miss those sometimes. That's what the Thoroughness slider is for. That controls how long it keeps looking for better combos without finding any, before giving up and returning the current best. With that slider maxed, it should be pretty rare that it misses things, but it's still possible. With a low Thoroughness, it'll miss subtle things like that a significant amount of the time. So if you can stand to wait, use a high Thoroughness.

• Adoriele · 08/06/08, 8:09 PM

Originally Posted by Astrylian

There is indeed a certain amount of randomness in the Optimizer. It tries quite alot of different gear sets and it gets smart about choosing things based on how well those things do in other gear sets, to try to find combos that work best. So when there's a slight gain from making a bunch of specific gem changes, it can miss those sometimes. That's what the Thoroughness slider is for. That controls how long it keeps looking for better combos without finding any, before giving up and returning the current best. With that slider maxed, it should be pretty rare that it misses things, but it's still possible. With a low Thoroughness, it'll miss subtle things like that a significant amount of the time. So if you can stand to wait, use a high Thoroughness.

I never run the optimizer without maxing thoroughness. It has also, in the past, consistently missed that 4T5 is about a 100 DPS upgrade from anything pre-hyjal plus badges. So I double-check set bonuses as well.

• Astrylian · 08/06/08, 8:13 PM

Can you post a character file that it doesn't find the optimal gear commonly, in an Issue on our Rawr - Home site? There's alot of tweaking we can do, so we can probably make it smarter about these situation.

Docjowles · 08/06/08, 11:11 PM

This is a silly detail, but the Mage model still allows you to say whether Curse of Shadow is active on your target. As of patch 2.4.3, that spell no longer exists (it was finally merged with Curse of the Elements).

Originally Posted by WoWWiki

Warlocks

* Curse of Shadows: This spell has been removed.

Patch 2.4.3 - WoWWiki - Your guide to the World of Warcraft

• Astrylian · 08/07/08, 12:11 AM

Yeah, someone else noticed that today too. *blush* Oops.

Promethius · 08/07/08, 6:03 AM

Originally Posted by Bashui

I don't know what you are trying to get at with averageYellowDamage, but it makes no sense whatsoever. Let us try to plug in some numbers:
1-.2-.4+.4*2.33 = 0.866 vs
(1-.2)*(1-2.33*.4) = .0544
I hope there is a typo in what you are trying to get across, because I know my average yellow hit does more than 5% damage

Yellow attacks are not able to glance. I suppose that there is technically an error with extreme crit for white attacks, since I am converting hits to glances.

There is a typo actually; it should be:
float averageWhiteDamage = 1 - chanceAvoided - chanceCrit + critMultiplier * chanceCrit;
float averageYellowDamage = (1 - chanceAvoided) * (1 - chanceCrit + critMultiplier * chanceCrit);

so with the numbers provided you'd get:
1-.2-.4+.4*2.33 = 1.332
(1-.2)*(1-.4+.4*2.33) = 1.2256

The point of this is due to how white attacks use a 1 roll system whereas yellow damage uses a 2 roll system.
White table would look like 20% miss/dodge/parry + 40% crit + 40% hit to fill the rest of the table
Yellow table would look like 20% miss/dodge/parry + 80%(40% crit + 60% hit) = 32% crit + 48% hit

Thus, having more avoided attacks would reduce the overall effectiveness of your crits with respect to specials. I believe I first found that information in the rogue theorycraft section, and it appears to be consistent at least empirically with Brutallus WWS parses in a difference of avoided and critical yellow attacks from bear yellows (facing) and cat yellows (behind). I'm a bit taken back by that internal cooldown clause I seemed to have missed so I'm hoping this isn't incorrect either. I also hope I didn't come across as offensive in the prior post as your work on modeling bear threat is greatly appreciated. I found the refresh attempts of lacerate in a 1m,1l,2s cycle and the uptime calculation for 2m,1l,5s pretty innovative.

Edit: I'm unsure of whether or not bear special attacks use a 2-roll system so it may be something that would have to be researched more in depth before actually implementing, but this was my train of thought.

thedopefishlives · 08/07/08, 10:57 AM

Originally Posted by Adoriele

As to why it's socketing so "little" haste for WOWTigra, he may be above the haste goodness threshold, but that's an equivalence point. Past that he'd want to gem equally for both Haste and SD. With 1450/129, it's actually more haste than it should be, not less (though 1450 may not be accurate. Parity for 129 haste would be around 1520 spell damage).

This is good to know. Maybe I did something right after all

Regardless, I do know that there are shortcomings to the way I handle Nature's Grace in the current calculation algorithm. It's kinda kludgy at the moment, and I'm hoping for a more elegant solution.