Rogue Simple Questions Thread: Cataclysm Edition - Page 27

• Andeh · 01/24/11, 6:07 PM

Yeah, I had a discussion with Zakalwe (one of the SC devs) about a month or so ago and went over the Assassination logic with him to make sure it was using the same priorities, etc. as what we're used to. What sparked that discussion was SC's high value of Exp at the time, but it turned out that it wasn't using the new rule about all dodged/missed abilities getting 60% (or whatever it is) refunded.

I'll ask him about the Hit issue, and if it can break down Hit's valuation. But I am pretty sure that the reason why it's showing it rating so high is that for all of our normal gear setups, we're either just barely below the Spell Hit cap, or just barely over it. So, I'd think it would always end up valuing it higher than a method where Hit is added incrementally.

• Zakalwe · 01/24/11, 6:58 PM

I'll repost my response to Andeh here:

Simcraft just simulates the fight using a given amount of hit rating. You can [separate different "types" of hit] by manually specifying hit rating to the various caps before calculating scale factors and being careful about choosing scale deltas, but the sim doesn't do anything like that by default - it just simulates once with current stats and then once with 300 hit rating removed and compares the results.

An example of where it can go wrong when trying to find the "spell hit value" is if your current hit rating is less than 300 above the yellow hit cap. The second simulation will then occur below the yellow hit cap and your scale factor result will be influenced by those extra yellow misses, and thus be inflated.

On the topic of how reliable our numbers are, it should be noted that there are two ways numbers from simcraft can be "wildly wrong":

One is if there are actual bugs, either in the rogue module or somewhere in the fundamental simulation framework, which can obviously result in incorrect results. There have definitely been some of those bugs, which is a natural result of us not having anyone on the team who actually plays a rogue. We've had various rogues looking at our results lately though, and finally got a dedicated rogue developer on the team this week, so the situation should be very much improved at this point.

The other potential issue is our default profiles, which a lot of people like to run and compare and publish, but which will always end up being imbalanced if the gear hasn't been optimized or if the wrong talents have been picked, etc. Because we're working with a very well-designed simulation framework it's actually quite easy for me to implement all sorts of rogue mechanics even having no clue how to play a rogue, but if I fail completely at setting up a default "BiS" profile, those big simulations comparing all the classes will show rogues much lower than they should be. I suspect that's what Aldriana is referring to - we had some default rogue profiles with bad gear choices and worse talent choices, and someone went and simulated them alongside things like the default warlock profiles, which I keep very carefully optimized.

dedmonwakeen · 01/24/11, 7:56 PM

Originally Posted by Aldriana

1) I personally don't have a lot of confidence in SimCraft. The last set of numbers I saw before Cata launched from SimCraft were *wildly* wrong. It had all 3 rogue specs as the bottom 3 rogue DPS specs, down around 16k in BIS. Obviously this is incorrect.

Any cross-class DPS comparison using SimC that is not actually present on our wiki should be completely disregarded. This is especially true while we retain the "beta" appellation.

The SimC team has a great deal of respect for the individual class modeling tools and we often use them to help ensure we are moving in the right direction. When discrepancies arise players should feel free to open an Issue containing the necessary data to exercise both SimC and whatever it is being compared against. The devs jump on modeling issues with alacrity.

• Aldriana · 01/24/11, 8:07 PM

Well, I'm specifically referring to things like the following chart, posted December 1, with numbers pulled from SimCraft:

1.Shaman (elemental) - 24868dps
2.Warrior (arms) - 24810dps
3.Spriest - 24223dps
4.Warlock (aff) - 23927dps
5.DK (unholy) - 23674dps
6.Druid (balance) - 23322dps
7.Mage (fire) - 23164dps
8.Warlock (destro) - 23120dps
9.Warlock (demo) - 22554dps
10.Shaman (enhancement) - 21577dps
11.DK (frost) - 20803dps
12.Warrior (Fury) - 20770dps
13.Druid (Feral) - 19703dps
14.Mage (arcane) - 18856dps
15.Mage (frost) - 18783dps
16.Ret - 17892dps
17.Rogue ( mutilate) 17649dps
18.Rogue (combat) - 16885dps
19.Rogue (sub) - 14565dps
...Hunters - N/A dps

Note how these numbers have basically nothing to do with reality. This does not inspire a great deal of confidence in me as to its general accuracy. If the massive discrepancy here is known to just be because of a stupid gear set, that may improve my opinion of it somewhat.

That said: the fact that it takes stats in chunks of 300 is going to make really good EP values tricky for capped stats, so I guess I'm not totally surprised that the numbers are somewhat different than one you compute them one point at a time.

Edit: Having read Dedmonwakeen's post, its entirely possible that these are invalid based on not being "offiicial" numbers, but I think the overall criticism - both the specific point that computing EP for capped stats is tricky with that step side, and the larger point that the SimCraft team seems somewhat lacking in specific rogue expertise - remains largely fair. Again, not claiming ShadowCraft is perfect either, just saying I have a bit more confidence in it, particularly when it comes to EP values.

Shalcker · 01/24/11, 8:41 PM

Originally Posted by Glanzick

Again good agreement for everything except the capped hits. I'm aware that SimC calculates scaling factors by lowering expertise and hit instead of raising them, could that be the source of difference?

You can set "Options/Scaling/Use Positive Deltas only", that will drop hit value for your gear considering that you're spellhit capped and simcraft estimates deltas in fairly wide swings.

Should I just disregard SimC's results?

Looking ShadowCraft code... Umm, isn't nothing but white swings are affected by expertise in Shadowcraft for Assassination cycle? Maybe i'm missing something, but there seems to be no other connection there...

Then problem is most likely here in SimCraft output for your gear:
Gains -> energy_refund 251.8 0.0%
With 80% energy returned in refund we can see that 62.95 energy were "wasted" in dodged finishers and combo builders, which is a tiny bit more then two backstabs or one mutilate.
Dodged garottes also lead to lower energy regen (as well as lower damage), and unlike other moves they can only be tried once per stealth/vanish, that should lead to a bit of dps difference too.

• Aldriana · 01/24/11, 9:24 PM

Originally Posted by Shalcker

Looking ShadowCraft code... Umm, isn't nothing but white swings are affected by expertise in Shadowcraft for Assassination cycle? Maybe i'm missing something, but there seems to be no other connection there...

Ability costs are divided by one handed strike chance, to account for the energy you lose by reapplying. This is why Expertise has a different EP value than white hit.

shadowboy813 · 01/25/11, 1:45 AM

Another note on Simulationcraft's results:

I did a little testing with combat for simulationcraft and the priority list it offers as a default option is not optimal. I simply changed 4 characters in the priority list for my own rogue (not BiS by a long shot) and went from 16.1k to 16.6k dps. I did not save them but I could attempt to reproduce the issue with a BiS combat list if anyone would like to see.

Point being that the problem may very well go beyond the modeling and even into their own suggested priority queues.

I do share general concern and distrust over simcraft's scaling values. One of the biggest things that had me scratching my head is that the PTR version has mastery's scaling for combat even above expertise. It's a buff, no doubt but not that much of one. Even the direct damage from PTR vs how it is on live won't be as much as one would think because it is normalized on live and still normalized on PTR (I did some testing myself to confirm whether or not it remained so, I can post screencaps if anyone wants).

Shalcker · 01/25/11, 3:09 AM

Originally Posted by Aldriana

Ability costs are divided by one handed strike chance, to account for the energy you lose by reapplying. This is why Expertise has a different EP value than white hit.

Hmm, yes, sorry.

Then there can be second-order effects (like chance of double dodge), as well as possibility of longer rupture downtime, or losing garrote. Just adding (1.0 - 1.0 * self.strike_hit_chance) to rupture_downtime seems to boost expertise slightly ahead in value, adding garrote no-retry-on-dodge-possibility should boost it even more.

This is damage breakdown for simcraft vs shadowcraft (Using Rostana gear, 5m fight with 2 potions, flask, heroism in simcraft, defaults in shadowcraft, second cycle is current default for assassination in simcraft):

attack	(1+cp rupt, garrote)	(4+cp rupt, gar)	(4+cp rupt, no gar)	shadowcraft
backstab	1914	1902	1887	2164.41224806
deadly_poison	2692	2690	2695	2587.30661479
envenom	2777	2646	2630	2825.40812458
garrote	161	157	-	-
instant_poison	6028	5804	5785	5416.1241499
autoattack	3895	3811	3803	3902.94136581
mutilate	2435	2444	2480	2237.67094306
rupture	605	627	633	588.659144725
venomous_wound	1883	1707	1506	1746.5467185
total	22389	21791	21423	21469

Looking at it like that higher value of spellhit probably comes from higher expected contribution of instant poison and venomous wounds.
Simcraft also seems to consider execute phase shorter then ShadowCraft (where it is set to exactly 0.35 by default, which is probably a bit optimistic considering execute abilities of other classes).

Originally Posted by shadowboy813

I do share general concern and distrust over simcraft's scaling values. One of the biggest things that had me scratching my head is that the PTR version has mastery's scaling for combat even above expertise. It's a buff, no doubt but not that much of one. Even the direct damage from PTR vs how it is on live won't be as much as one would think because it is normalized on live and still normalized on PTR (I did some testing myself to confirm whether or not it remained so, I can post screencaps if anyone wants).

Main gauche is marked as normalized in simcraft.

I cannot seem to reproduce high combat mastery scaling value, because both bis and your character provide scaling values as exp > hit > haste/mastery > crit (since most combat characters should be close to capped on expertise, "use positive deltas only" shouldn't be used when looking at relative expertise scaling).

shadowboy813 · 01/25/11, 3:45 AM

Originally Posted by Shalcker

Main gauche is marked as normalized in simcraft.

I cannot seem to reproduce high combat mastery scaling value, because both bis and your character provide scaling values as exp > hit > haste/mastery > crit (since most combat characters should be close to capped on expertise, "use positive deltas only" shouldn't be used when looking at relative expertise scaling).

I never changed the default scaling values when I got the high scaling for mastery, so the hit and expertise delta values were negative. Maybe Simulationcraft has been updated with a newer build since I downloaded it 10 days or so ago. I honestly haven't checked.

Edit: I just checked, I'm using an older build of Simulationcraft. I now see the same results you do. I ran my rogue through simulationcraft with default and modified priority queue, and I am still seeing the dps increase that I saw before. Whether it means anything or how significant or important it is, I have not considered yet. The priority queue it uses by default still seems to aim for a high rupture cycle and leads to more energy capping, presumably waiting to reapply rupture. The changes I made cut the energy cap uptime in half and smoothed the energy timeline considerably.

All I did was change:
actions+=/eviscerate,if=combo_points=5&buff.slice_and_dice.remains>7&dot.rupture.remains>6
actions+=/eviscerate,if=combo_points>=4&buff.slice_and_dice.remains>4&energy>40&dot.rupture.rem ains>5

to:
actions+=/eviscerate,if=combo_points=5&buff.slice_and_dice.remains>7&dot.rupture.remains>0
actions+=/eviscerate,if=combo_points>=4&buff.slice_and_dice.remains>4&energy>40&dot.rupture.rem ains>0

• Aldriana · 01/25/11, 2:04 PM

Originally Posted by Shalcker

Hmm, yes, sorry.

Then there can be second-order effects (like chance of double dodge), as well as possibility of longer rupture downtime, or losing garrote. Just adding (1.0 - 1.0 * self.strike_hit_chance) to rupture_downtime seems to boost expertise slightly ahead in value, adding garrote no-retry-on-dodge-possibility should boost it even more.

<snip>

Simcraft also seems to consider execute phase shorter then ShadowCraft (where it is set to exactly 0.35 by default, which is probably a bit optimistic considering execute abilities of other classes).

The multiple dodges effect is accounted for; consider, for instance, Envenom which costs 35 energy when it lands and 7 each failed attempt. Assume your dodge/miss rate is p, q=1-p, and you reapply until the envenom lands. Well, you'll always land an envenom eventually, so you're paying 35 regardless; the only variance is how many times you pay 7 energy first. With probability q you don't pay it at all, with probability pq you pay it once, p^2 q you pay it twice, and so on. So the total cost is

35 + 7 (pq + 2p^2 q + 3p^3 q + ..)
 = 35 + 7q(p+p^2+...)+7pq*(p+p^2+..)+7p^2 q(p+p^2+...)+...
 = 35 + 7pq(1+p+p^2+...)(1+p+p^2+...)
 = 35 + 7pq*(1/q)*(1/q)
 = 35+7p/q
 = 28+7(p+q)/q
 = 28+7/q

Which is exactly the formula that appears in ShadowCraft. The cycle disruption is also modeled to some extent IIRC, but you're right, once tight rupture cycles and garrote are added its value may creep up somewhat. Its conceivable to me that it will pass haste, though I'd be a bit surprised if it went much higher than that.

Re: Backstab uptime. .35 might be a bit optimistic, but not wildly so. If you assume that all classes have an execute that's exactly as good as Murderous Intent - kicks in at 35% and gives about as good as Backstab (say, a 15% damage increase), and that no one is dead by the time the execute phase rolls around (not always a good assumption for progression content), the expected time spent in execute range is (.35/1.15)/(.65+.35/1.15) = 32%. And in practice, many DPS don't have an execute, and/or its not as good as Murderous Intent, and/or it starts at 20% not 35, and/or some people die before you get to execute phase... I mean, if you want to use 32 instead, you're welcome too, but 35 doesn't strike me as completely unreasonable either.

Toan · 01/25/11, 3:54 PM

To be fair, the Assassination Guide doesn't provide a specific answer for parry/miss/block. Parries behave the same way as dodges, in that parried envenoms generate the envenom buff. Blocked attacks still count as landed attacks and, therefore, envenom behaves as if it had landed normally. Misses I am not certain of, but realistically speaking, you should never have an envenom miss as the hit cap for finishers is extremely low.

ieatpaperbag · 01/25/11, 4:14 PM

On the beta at least, missed Envenoms still provided the Envenom buff.

Killme888 · 01/25/11, 4:52 PM

Originally Posted by Toan

To be fair, the Assassination Guide doesn't provide a specific answer for parry/miss/block. Parries behave the same way as dodges, in that parried envenoms generate the envenom buff. Blocked attacks still count as landed attacks and, therefore, envenom behaves as if it had landed normally. Misses I am not certain of, but realistically speaking, you should never have an envenom miss as the hit cap for finishers is extremely low.

Missed envenom definitely applies the buff as it regularly happens on Chimaeron due to Caustic Slime.

Thelen · 01/25/11, 5:28 PM

When should you renew rupture? Just before it runs out or when it has run out? If there is any difference in dps.

Zippers · 01/25/11, 5:34 PM

Originally Posted by Thelen

When should you renew rupture? Just before it runs out or when it has run out? If there is any difference in dps.

From what I understand you can refresh an equal CP rupture with <2sec left on the duration. Otherwise you should (have to?) wait for it to run out.