Rogue DPS Simulation Spreadsheet - Page 34

Grigori · 11/11/09, 2:33 PM

Originally Posted by Ikutaba

It is quite literally that script wrapped in an addon. The only thing the addon "does" is activate the first script automatically upon logging in. There is no difference, you perceived one.

edit: In fact, the addon should be faster for people with slower PCs, because it checks your DP stack differently than the script does. The script checks every time something "happens" to the mob, the addon specifically checks for your DP stacks. Which is why the addon swaps to your DP dagger when you swap targets instead of waiting for you to attack.

Actually, the various Outfitter and macro solutions in the "changes" thread all use the same DP timing logic bural's macros use. Some of the solutions check more often than others, but they all (attempt to) implement the same strategy. The addon is strategically different in that it waits for a MH swing to swap. Unfortunately, it waits for a MH swing unconditionally, which is not necessarily strategically sound.

In theory, an addon will decide between waiting for the MH swing to swap, waiting for the OH swing to swap, and swapping immediately (based on conditions including where the MH/OH swing timers are, whether the swap is DP->IP or IP->DP, etc.). For example, if you are using three daggers with identical speed, and your Mutilate refreshes DP 0.9 sec from your swing, you swap immediately clipping the swing so your next swing is with IP, but if you are 0.1 sec from your swing with DP, you wait for the swing before swapping.

In practice, the best you can hope for is probably Mavana modifying his 'sheet to model the wait/no-wait decisions and you code a few pre-calculated inflection point values into the addon. Given that Blizzard is now "leaning toward" doing something about weapon swapping, modifying the 'sheet is probably not worth the effort at this point.

These are my observations based on empirical results from building an addon+macro weapon swap package for an Assassination Rogue: On the IP->DP swap, triggering the swap off MH swing with a slightly longer DP lead time beats triggering the swap ASAP once inside your DP lead time; on the DP->IP swap, triggering the swap ASAP after the DP refresh beats triggering the swap off MH swing after the DP refresh.

Mavanas · 11/11/09, 6:48 PM

I am worried a little about timing of swapping and swings.
- First of all, how do you determine when the next swing occurs? Do you assume the delay is the same as the previous one or you track the haste at every point of time?
- Second, I know that certain processes occur on the client side and some on the server side. How does this work into the exactness of swap timing? Is there a delay between when the addon issues a command to swap and when the swap actually occurs? Can it happen that by forcing the timing of swings and swaps you actually hurt yourself because the swap always occurs 100 miliseconds after the swing (just an arbitrary lag amount)?

Antiarc · 11/11/09, 6:52 PM

Short version, I keep a record of the last 10 swings, and the mainhand swing speed during those last 10 swings, and look for an appropriately-sized gap between the current swing and a past swing. If a gap of the appropriate size is found, it's considered a mainhand swing. This means that 2x weapons of the same speed won't have great mainhand swing speed detection, but even worst-case, the swap always happens immediately following either an offhand or mainhand a swing.

The actual events used are the SWING_DAMAGE and SWING_MISSED sub-events of the COMBAT_LOG_EVENT_UNFILTERED event. It's about the best resolution we can get, but anecdotally, it "feels" pretty good just from developing and testing it.

Demi9OD · 11/11/09, 7:21 PM

I'm noticing a marginally longer swap time to IP since I upgraded from 0.4.2 to 0.4.5, at least compared to Grigori's script. Deadly is proccing on top of a 5 stack occasionally now, and it never used to before.

Antiarc · 11/11/09, 7:31 PM

Swaps aren't immediate, as the mod looks for a mainhand swing before swapping. It's entirely possible to proc Deadly with an offhand swing while waiting for a mainhand swing.

Athariel · 11/12/09, 6:32 AM

Have you considered the "wasted" (if any) damage (i.e. damage that did not happen due to having the deadly poison weapon equipped) due to waiting for a switch? Have you wighted up the damage lost due to the possibility of a poison proc of the non-deadly poison weapon versus the damage of your mainhand?

i.e. (Mainhand swing + x Offhand swings + possible poison proc from mainhand - damage lost due to swing reset) > ( -damage lost due to swing reset + x offhand swings + x offhand poison procs)

Of course, "damage lost due to swing reset" will be in most cases higher in the scenario where you swap instantly than in the scenario where you wait for a mainhand strike, but does the possibility of extra poison procs weight up? I addition, over a long fight, how does the seconds lost due to waiting weight up versus a full round of instant / wound poison?

I'm just throwing my thoughts up in the air. There are many 'if's here, but overall, what is better?

Grigori · 11/12/09, 1:59 PM

Originally Posted by Mavanas

I am worried a little about timing of swapping and swings.
- First of all, how do you determine when the next swing occurs? Do you assume the delay is the same as the previous one or you track the haste at every point of time?
- Second, I know that certain processes occur on the client side and some on the server side. How does this work into the exactness of swap timing? Is there a delay between when the addon issues a command to swap and when the swap actually occurs? Can it happen that by forcing the timing of swings and swaps you actually hurt yourself because the swap always occurs 100 miliseconds after the swing (just an arbitrary lag amount)?

You are still losing swing time when triggering off swings: The server swings and sends the swing notice to the client. The client receives the swing notice and sends the swap command to the server. The server receives the swap command and swaps. Network latency applies (twice), so the swing time gained by triggering off the swing event over letting the swap occur at random points into the swing is markedly less than what is possible in theory.

In terms of auto attack damage for the hand you are triggering from, if the swing trigger were competing against randomly distributed triggers, it's obvious that the swing trigger will win given most normal combinations of network latency and haste. Of course, there are other factors at play here. How much are you clipping your other hand's swing? How much poison damage are you losing by delaying the swap? Is your swing trigger actually competing against randomly distributed triggers?

The math is complicated enough that even a good function to estimate the inflection points will probably not be available by methods other than goode olde simulation. Of course, you can still test competing settings empirically. In my experience, triggering off MH swing + poison timer wins over triggering off only poison timer for the IP->DP swap. DP->IP ASAP swap wins over waiting for MH swing for Assassination fast-fast/fast. DP->IP on MH swing wins over ASAP swap for Assassination slow-fast/fast. Combat did slightly better with DP->IP ASAP swap over waiting for MH swap, but it was not a clear trend.

Vonlego · 11/12/09, 2:55 PM

The method I used to determine which time to swap to deadly was pretty simple and you might want to look into it yourself Demi.

First off I have read, in this thread, that mutilate wishes to swap to deadly around 3 seconds remaining and combat should swap closer to four seconds. Simply set the 'time' to (ideal swap time) + (average time on weapon swing).

With a heavy haste mutilate set a main hand swing timer is usually just above one second. Duration on swing can vary between 'practically zero' and the full duration. Simple math, (0+MHSpeed)/2, of course finding the average swing timer on my dagger is roughly half a second. Add your desired swap time to this number and calibrate your addon with the total.

In practice, I've had my stack fall off once during the full ToC up to Anub; however, that may have been user error on my part. One problem I do see with using an addon over doing it manually is for cases when you wish to refresh the stack before leaving the target, such as Icehowl's crash. That being said, in this particular case the randomness of where he jumps to usually results in a lost stack anyways.

Axodry · 11/12/09, 3:47 PM

Hi Mavanas,

I know that I can override the the actual talents from your standard macro for 51/18/2, but I was just curious as to why you chose to take 2 points in Quick Recovery instead of Murder? I was under the impression that if your expertise capped, that talent is no longer useful (perhaps this is standard incase your not exp capped?). As you know, Murder is currently getting buffed on the PTR which may or may not come to fruition, but if it does, I think that will be a talent that can't be overlooked anymore.

That all being said, according the the spreadsheet, even though i'm not yet in BIS gear, wpn swapping ruptureless 51/18/2 is coming out ahead of the traditional 51/13/7 (even without regemming from 20agi to 20ap/10 haste). One thing i've noticed is that on a fight like Koralon where there is not alot of movement, 51/13/7 is coming out ahead by around 1k dps in reality (I typically do around 7k and only hit 6k y'day). Any insight as to why this might be? On a fight like Anub, I noticed a significant increase with the ruptureless cycle going the other way (1k or so). I've tried using both the macro with the blizz UI as well as the poisonswapper addon. It feels like to me it's much harder to be aware of your surroundings if your having to watch for when your DP is about to drop-off when having to do it manually.
Thanks,

Mavanas · 11/12/09, 3:52 PM

I have a pretty good analytical method that I used before adding weapon swapping to simsheet to determine if weapon swapping was worthwhile. Based on that method, proper timing of swap to MH swing is definitely worth it. It adds about 60 dps (from 540 to 600). The only thing that the analytical method does not take into account is the interaction of GCD imposed by the weapon swapping and your regular rotation, but the effect of timing swaps to swings on that is very minimal.

Interaction with offhand swing we can also ignore, on average it does not matter if you time the swap randomly or you time it to your mainhand swing, the effect on the offhand white damage and poison damage should be zero on average since offhand swings are independent from mainhand swings. In other words, in both cases, I still expect that the swap will occur in the middle of the offhand swing.

Vonlego · 11/12/09, 4:16 PM

Mavanas I'm having some peculiar results coming back from my simulator. Long story short is I'm due for a Death's Verdict here pretty soon and was wanting to check out the results, and which trinket to swap. I currently use DMC:G and Blood of the Old God running mutilate with a mix of 245/258 with a few choice 226 (belt/cloak from uld10). My gear is far from perfect or top of the line, but it is very solid. I run weapon swap, lightning reflexes, gem AP/Haste, and use the four piece.

This is the thing confusing me -- I swapped out BoTOG for DV258. The first time I ran the simulator it came out as a loss of dps, and the second time it came off as a minor increase (~20 dps). White hit simply can not be that valuable even with a poison build. Any hints on how to fix this?

Update: I ran the 245 version to see if I perhaps messed up the cells pertaining to the 258 somehow. The same thing happened --

Trial starts out as a ~150 dps upgrade, p-value staying very small. 20% in or so the p-value begins to jump rapidly and the dps slowly but steadily drops.

Update 2: Dropping the DMC:G instead of the BoTOG results in a 250 dps for an extended period of time, then dropping to +80 dps over the last section. P-value stayed steady at 0 for the majority of the trial. I find it hard to believe that the upgrade is only 80 dps, but perhaps I'm just misinformed.

Killme888 · 11/12/09, 4:35 PM

One thing to add is that if you swap with 3 weapons of the same speed, your swings after the first swap will sync up and the benefits of swapping after a swing will probably be even greater. I run with 3x Bladebreaker and looked at one of my recent parse to see if the swings desynced at all and the only times that the swings aren't in sync are the few seconds before the first swap, and about 5x over a 3 minute fight where the largest variance was 11 ms (others were 4-6 ms) which is negligible.

Seditions · 11/12/09, 5:52 PM

Originally Posted by Vonlego

This is the thing confusing me -- I swapped out BoTOG for DV258. The first time I ran the simulator it came out as a loss of dps, and the second time it came off as a minor increase (~20 dps). White hit simply can not be that valuable even with a poison build. Any hints on how to fix this?

Update: I ran the 245 version to see if I perhaps messed up the cells pertaining to the 258 somehow. The same thing happened --

Trial starts out as a ~150 dps upgrade, p-value staying very small. 20% in or so the p-value begins to jump rapidly and the dps slowly but steadily drops.

Update 2: Dropping the DMC:G instead of the BoTOG results in a 250 dps for an extended period of time, then dropping to +80 dps over the last section. P-value stayed steady at 0 for the majority of the trial. I find it hard to believe that the upgrade is only 80 dps, but perhaps I'm just misinformed.

You are most likely crit capping - having both DMC:G and Death's Verdict procs up.

Antiarc · 11/12/09, 6:12 PM

Originally Posted by Vonlego

The method I used to determine which time to swap to deadly was pretty simple and you might want to look into it yourself Demi.

First off I have read, in this thread, that mutilate wishes to swap to deadly around 3 seconds remaining and combat should swap closer to four seconds. Simply set the 'time' to (ideal swap time) + (average time on weapon swing).

The addon does indeed do this, though a little smarter. It sets the time-to-swap at (ideal swap time) + (time remaining until next swing). It's not perfect, but the intent is that if you have a 1.8 speed mainhand, are 1.0 sec through your swing, have your ideal swap time set at 3.0, then the mod will initiate a swap once the timer is under 3.8, in theory actually executing the swap at 3.0 seconds.

The observations about the instant swap to a damage poison being ideal are interesting. I'll have to do some testing and see what I come up with, as well. Admittedly, though, I am more of an implementation guy than a theorycrafter.

Mavanas · 11/12/09, 7:21 PM

Axodry, I have quick recovery selected by default because as soon as you select murder talents, it calculates dps as if the target is murderable and you have murder talent, i.e. you are doing 4% more damage on average. If you want to see what your dps would be with murder talent but on a non-murderable target, out those two points into fleet footed or something that does not have any effect on dps in the simulation. Also I can't really comment on your observation regarding 51/18/2 vs 51/13/7 because I am not sure how you collect your "real" data. How do you make sure the buffs and their uptimes are fixed? Different buffs or lack thereof effect the two specs differently. What is the effect of your regular RNG? If you run the simulation 300 times and then look at backup, in those 300 times your dps may differ by 1k between minimum and maximum, that's RNG of a fight under pretty favorable conditions where you simply stand still and dps. If you add movement, varying buff uptimes, buffers dying, sunders falling off, varying reaction time, then the RNG grows even bigger. That is why real raid data is a fairly unreliable check for theorycrafting.

Vonlengo, check crit capping indeed. Also, while the simulation is running, any intermediate results are not very important. The variance of initial dps results is really small, so the dps at 20% progress is meaningless. The only number that matters is the final dps when the simulation has finished running.