Any statistics that anybody gives from a raid setting is not useful for determining the "true" proc rate of LnL outside of the ICD for the purposes of future theorycrafting. What we need, pure and simple, is tests from dummies, similar to what Hungtar and I have done for many, many procs.
I think the 15% number that's being tossed around is probably what we'll see in the real world, where many ticks of BA are going to be within the 22 second ICD, but with a dataset of now ~200 procs giving a proc rate of 21.8% (197/902), I doubt that we'll get much lower than a 20% final rate, which would sound like the sort of nice round number that Blizzard would go for if they were going to stealth buff it.
As far as delaying the BA cast, there's probably some modeling that could be done as to how many ticks would be inside the ICD as to whether there's a benefit in delaying or not.
In any case, we need more dummy data for outside the ICD, so take 10 minutes, get 15 or so procs, and either PM or post your results.
[Edited: Removed dupicate Bronnum data from the final tally]
There was not a change to the proc rate. It's well known that spells/items with internal cooldowns seem to 'save' procs that should have happened, but didn't according to the ICD. Which is why your DPS trinkets will usually proc within an attack or two after coming off the ICD.
There was no actual change to the proc rate, stop claiming otherwise.
As far as delaying the BA cast, there's probably some modeling that could be done as to how many ticks would be inside the ICD as to whether there's a benefit in delaying or not.
Following up on my earlier mathcraft, I've built simulations for all hybrid strategies, such as delaying BA for ICD if only 1 tick would be suppressed, otherwise using BA on CD. Again assuming a 15% proc rate for now, there is no delaying strategy that appears beneficial when considering solely BA. The situation in my earlier post is actually the best-case for delay. Hybrid strategies are monotonically inferior to the "always delay" approach. For BA delay to be a net gain, an average of 1 LNL proc every 15-20 minutes of rotation time would have to counteract the loss of 80-100 seconds of BA ticks (and the +6% damage buff) over that same timeframe, which is patently impossible. No proc rate within the error bars of the available data results in a different conclusion.
There is, however, a situation where "BA delay" is advantageous. That is in combination with Frost Trap, based on its currently-reported behavior. First and foremost, I believe this behavior to be a bug. Additionally, the number of fights where it is viable is relatively small. However, with those concerns in mind....
Frost Trap has a 100% chance of triggering LNL if it affects a target. The LNL triggered by Frost Trap does not currently invoke the 22 second ICD, which is what I believe to be the bug. However, Frost Trap is itself affected by the ICD -- a trap triggered <= 22 seconds after a BA-triggered LNL will not proc LNL. Under the theoretically situation in which enough snareable adds exist to allow perfect use of Frost Trap, the optimal strategy is to Frost Trap, consume the LNL charges (maintaining the 2s interweaving as we normally do), then immediately BA. If that BA does not result in an LNL proc, Frost Trap on its CD. If BA does proc LNL, Frost Trap on the LNL ICD. My modeling of this is rather rougher than the pure-BA models, because issues like the GCD and LNL charge use timing are germane. However, any such introduced errors are certainly smaller than the 2.5-fold increase in LNL procs it implies. Although Trap-BA-(delay)-Trap cycles lose an average of 60 seconds of BA time per minute, the ideal circumstance rockets up to over 16 LNL procs in the average 5 minute fight. Again, this system requires:
1) A supply of Frost Trap-capable adds.
2) Nontrivial timing.
3) Belief that Frost Trap-LNL is intentionally failing to activate the ICD (or that they don't care enough to fix it, if it's bugged).
In Ulduar, Razorscale, Ignis, Auriaya, Thorim P1 (arena), Freya, and Mimiron are the fights most likely to benefit from this approach for some or all of the combat's duration. In CotC, it is probably not viable until Faction Champions (as Blizzcast 10 reveals they have no "boss" immunities).
Originally Posted by Spamurai
There was not a change to the proc rate. It's well known that spells/items with internal cooldowns seem to 'save' procs that should have happened, but didn't according to the ICD. Which is why your DPS trinkets will usually proc within an attack or two after coming off the ICD.
There was no actual change to the proc rate, stop claiming otherwise.
Increasing amounts of data strongly suggest otherwise. Regardless, the conclusions from my modeling hold at 6% as well as at 15%, for both BA-only situations (do not delay) and BA+Frost Trap situations (delay Frost Trap/LNL/BA cycles for ICD on BA-triggered procs).
There was not a change to the proc rate. It's well known that spells/items with internal cooldowns seem to 'save' procs that should have happened, but didn't according to the ICD. Which is why your DPS trinkets will usually proc within an attack or two after coming off the ICD.
There was no actual change to the proc rate, stop claiming otherwise.
That is nearly an irrelevant fact and an invalid argument. At a 6% proc rate it takes on average 16.67 BA ticks to proc LnL. The tests above are designed to fire BA only outside of the LnL cooldown, and there is enough evidence to establish an average of 5-6 BA ticks per LnL. 902 BA ticks were tested in these conditions and at 6% they would generate 54 LnLs. They generated 197. Something certainly changed.
Even in the case of the two raid experiences posted, a full 2/3 of the procs would have to be triggered by traps. On my run I had 531 BA ticks. At 6% that would generate 31 LnL procs, assuming that any BA tick was able to generate a proc. Under the assertion above, 49 of the 80 actual LnLs would have to have come from traps, at most 10 were triggered by traps.
There is already enough statistical evidence to prove the claim that the proc rate is now far higher than 6%, or that the mechanics have changed in an equivalent way. There is just not enough evidence to prove a specific figure to the tenth of a percentile.
Last edited by Difool : 08/12/09 at 1:49 PM.
Reason: Updated (corrected) numbers per subsequent posts)
Our testing should prove one way or another in the end. By only counting the tick's prior to the proc and coming up with a total % rate we can establish the non cumulative theoretical rate. We can then sample a large enough pool of total ticks over a period of time including ticks while on ICD to get a percentage proc. Under a large enough sample size either model will give approximately the same proc over time under and infinite condition. Unless Blizzard states one way or another, it will be almost imposable to determine one way or another. We need a value that lets us model as close to game play as possible.
324 BA ticks were tested in these conditions and at 6% they would generate 19 LnLs. They generated 47. Something big changed.
I'm wondering where you're getting the 324 from. If we're merely adding Bronnum's and Hungtar's data together, it's 47 procs (9 from Bronnum and 38 from Hungtar) out of 231 (59 from Bronnum and 172 from Hungtar). Bronnum I believe was wrong in his/her tally of Hungtar's data; It's 172 ticks, not 265. This is creating some confusion here. So far, the only clean datasets, consisting of solely procs outside of the 22sec internal CD, that we have to go with are the following:
Stubacca: 150 procs out of 671 tics
Bronnum: 9 procs out of 59 tics
Hungtar: 38 procs out of 172 tics
Total is 197 procs out of 902 tics, or 21.8%
Let's use this as the starting point.
[8/13/09] With Bronnum and Lahiri's new data: 276 out of 1272 tics, 21.7%
From those 605 procs, not all were eligible for LnL procs because you did not wait for the 22 second cooldown of LnL to finish:
0:02'56.297 Kibu gains Lock and Load. #1423
0:02'58.703 Kibu Lock and Load was removed from Kibu. #1439
0:03'07.985 Heroic Training Dummy is afflicted by Black Arrow. #1508
0:03'11.031 Heroic Training Dummy suffers 1036 Shadow damage from Kibu Black Arrow. #1533
0:03'14.063 Heroic Training Dummy suffers 1006 Shadow damage from Kibu Black Arrow.
Edit: by any chance, do you still have the WoWCombatLog.txt?
Bronnum I believe was wrong in his/her tally of Hungtar's data; It's 172 ticks, not 265. This is creating some confusion here. So far, the only clean datasets, consisting of solely procs outside of the 22sec internal CD, that we have to go with are the following:
Stubacca: 150 procs out of 671 tics
Bronnum: 9 procs out of 59 tics
Hungtar: 38 procs out of 172 tics
Total is 197 procs out of 902 tics, or 21.8%
Let's use this as the starting point.
You're absolutely right, thanks for the correction
For the sake of simplicity and eaze of dummy testing, I believe it would be a possiblity to cast BA on every CD and simply filter out the ticks affected by 22 sec ICD on LnL manually in the log. Personally, I would find this approach more consisten compared to stop watch timing as I used in my last sample ( /hint only 9 procs :P ).
I sure do agree the 6% proc chance is history for sure, but to me the ~20% chance still seems too high; but as Tarus claims, our testing will tell in time. Regarding trap dancing and modelling, it already seems too impossible to handle correctly in most situations. Though, if this will be 'exploited' (assuming it's a bug as it was when wotlk hit) in arenas, I'm quite sure Blizz will fix it, at least they should.
You're absolutely right, thanks for the correction
For the sake of simplicity and eaze of dummy testing, I believe it would be a possiblity to cast BA on every CD and simply filter out the ticks affected by 22 sec ICD on LnL manually in the log. Personally, I would find this approach more consisten compared to stop watch timing as I used in my last sample ( /hint only 9 procs :P ).
...
Or just record the combat log of your testing, upload it somewhere and send me a PM with an URL to your log and let the parser do the counting. Please no raid logs and include your character name. And only logs of English or German language clients. Of course, i can't guarantee the parser works correctly. But if it doesn't, i'll find the bugs eventually. Or someone else will.
Counts are:
LnL procs out of 22sec cd = 29
BA tics with possible LnL proc = 139
20.86% significant higher than the earlier 15% mentioned.
BA tics distribution: Mean = 4.79 -/- standard dev. = 4.84
Kurtosis of 6.58 showing how affect my test was to 5 outliers (tics: [22, 12, 11, 12, 11])
Removing those give Kurtosis value of 3.33 - wts RNG!
Let's hope Shandara will update speadsheet to re-model the proc rate of LnL soon'ish
Finding the new LnL proc rate is useful for a simulation type test like simulationcraft. However for the spreadsheet I'm having a hard time coming up with an appropriate formula that can model around the cooldown and BA uptime properly. This is why I was trying to get an actual rough proc rate based on LnL procs versus BA ticks (without compensating for the cooldown) because the easiest thing to do for now would be to figure out that and use it for the spreadsheet calculations. The rotation test kind of does a simulation but unless you're using randomization for it, the procs will always come at consistent intervals which won't really reflect real world behavior very well.
Based on my looking through our combat logs, the rate we're going to see LnL proc rates based on a normal shot rotation is more in the 10-15% range, which is still much higher than the old rate but not going to match up with the higher pure proc rate. For now I've upped the rate on FemaleDwarf.Com to 12% while I'm waiting for a better formula or more accurate number. Given that a lower than maximum resourcefulness and LnL talent investment will not scale in a consistent manner with an estimated proc rate like this, I would prefer to get a good formula but accounting for all the variables involved is pretty messy.
A possible way to circumvent the 'inside-22-sec-cd-issue' would be to simply count all number of tics within or not within LnL cooldown. This is not a good model in any way, but it will indeed be closer to the observed proc rate during actual encounters.
For my last test (post #3011) I had 228 tics of BA, making the proc rate 12.79%. Just to achknowledge your choice for FemaleDwarf.com
I recorded the combat log of auto-shooting at a dummy with BA every 38 seconds. To get the timing right i used the following macro:
That way, LnL is always off cooldown when the timer expires (that /qt .. stuff is a feature of Quartz). After a bit off afk auto-shooting with black arrows, i threw the combat log in a little parser i wrote (UtilityBase :: Paste code) and got:
(wws parse of the comat log used: Wow Web Stats. the parser should work ok as long as no other SV hunter appears in the combat log).
This is very interesting data. The first thing that sticks out to me is that you had quite a bit of procs at 1 or 2 ticks. Has anyone considered the possibility that each tick has a different chance to proc?
Using your data the breakdown looks like this:
No. of Ticks
Procs at this tick
Percentage of occurrence
1
9
0.236842105
2
7
0.184210526
3
6
0.157894737
4
2
0.052631579
5
1
0.026315789
6
3
0.078947368
7
0
0
8
1
0.026315789
9
2
0.052631579
10
3
0.078947368
14
1
0.026315789
21
1
0.026315789
I don't think it's too far off to consider the possibility that the first tick has a 25% chance to proc LnL and subsequent chances have a 5% less chance every time (first 25%, second 20%, third 15%, fourth 10%, fifth 5%). This would still mean that on average each tick has a 15% chance. If this is the case it would also mean that the chance you get a proc on a full duration BA (assuming the cooldown doesn't interfere) is 1-(.75*.8*.85*.9*.95)=56.4% which is only slightly different than 1-.85^5=55.6%.
Again, it is way too soon to draw any conclusions yet - I'm just throwing it out there as a possibility that might be worth keeping an eye on. Hungtar's data is great, but too small of a sample to really draw any conclusions yet. The reason it would matter at all is because if the internal cooldown prevents the first couple ticks of BA from being applicable it will have an adverse effect on "average" LnL proc chance (you'd end up with the last three ticks at 15%, 10%, 5% or 10% average, instead of a flat 15%).
Originally Posted by Florrie
What flavour of hipster racism am i missing today?
Again, it is way too soon to draw any conclusions yet - I'm just throwing it out there as a possibility that might be worth keeping an eye on. Hungtar's data is great, but too small of a sample to really draw any conclusions yet. The reason it would matter at all is because if the internal cooldown prevents the first couple ticks of BA from being applicable it will have an adverse effect on "average" LnL proc chance (you'd end up with the last three ticks at 15%, 10%, 5% or 10% average, instead of a flat 15%).
Actually, if this were the case you'd expect a steeper dropoff from the number of procs at 1 tic, 2 tics, etc. We just need more data to fit this thing to a curve with a constant proc rate: p = r*(1-r)^(n-1), where p is the percent of procs that will happen at a specific tic, r is the proc rate of black arrow, and n is the tic number. I'll post my proc rate data for the 150 I did last week, comma separated so somebody can just import it.
if any of those helps, and yes those were all timed with the CD.
Ran my parser against the combat log and got 477 ticks and 94 procs (19.71%, not that the WWS reports the total number of BA ticks, of those 477 were eligible for LnL procs). Not posting the whole output but the proc count for each tick count is:
Edit: Assuming my parser is right, between procs #24 and #25 and procs #31 and #32 BA was applied while LnL was on cooldown (~17 and ~19 seconds after the last LnL proc respectively). Don't know how much of an impact this has on the overall data.
Seeing as it should be benefitial to delay Black Arrow (though there's no actual math behind this yet, is there?), would Resourcefulness still be a good talent? Or might those points be better spent in, say, Imp. Aspect of the Hawk or Imp. Stings?
I've been curious about the same thing (dropping 3/3 Trap Mastery for 3/3 Imp Hawk) and I spreadsheeted it to see it comes up to a ~60-70 dps. I'm thinking it scales much better for people with slow weapons, such as [Magnetized Projectile Emitter]. I'll give it a shot and post the WWS here as soon as I get a chance.
You need those 3 points in order to reach the next tier of talents. Trap mastery is the only dps option to put them in, the other choices don't provide any dps boost.
You CAN take points out of Resourcefulness though, would those 3 points used elsewhere be a net dps gain? Obviously the points in Trap Mastery are a dps gain but the conversation here is, I think, aimed at whether waiting the extra 6 seconds that Resourcefulness talent gives nets higher dps due to less clipping of the internal cd LnL now has given its seemingly increased proc rate.
You CAN take points out of Resourcefulness though, would those 3 points used elsewhere be a net dps gain? Obviously the points in Trap Mastery are a dps gain but the conversation here is, I think, aimed at whether waiting the extra 6 seconds that Resourcefulness talent gives nets higher dps due to less clipping of the internal cd LnL now has given its seemingly increased proc rate.
As was stated above, it's a bad idea (see this post). But i think i have an explanation that will make this more obvious.
While you may get more LnL cooldown clipping with 3/3 Resourcefulness, you will also get more Black Arrow ticks. You want LnL on cooldown as often as possible, as that means more LnL procs. To increase the chance of BA being on cooldown, you increase the number of BA ticks. Because the other variables that determine LnL procs you cannot influence (besides not taking 3/3 of the LnL talent which might change the proc rate, which probably is not what you want).
08/11/09, 5:52 PM
Similar Threads
