[Raid] Executioner vs. Mongoose, preliminary numbers - Page 7

• Aldriana · 11/16/07, 10:19 PM

Well, it's sort of a "the more the better" type situation. Half an hour is good, an hour is better, and 2 hours is even better.

The real answer is "the total amount of data collected needs to be large enough to distinguish between the various proc rate options" - which basically means we need enough data at two different speeds that the various possible proc models no longer overlap - hence, the amount of data we need with Talon depends on the amount of data we get with weapons of other speeds.

That said: another hour or so with Talon would probably help quite a bit.

Warhound · 11/16/07, 10:23 PM

I have it on blade of inmafy 2.6 weapon speed.

Would you like me to upload date or is the data from the previous poster sufficient.

Im sorry, although im a frequent reader im no expert at all this. Basically you want me to /combatlog and record white hits for an hour or so if im not mistaken ?

That shouldnt be a problem, if you want me to download a buff cancelling mod as well please let me know

Ill be glad to help

• Aldriana · 11/16/07, 10:49 PM

Yeah, autoattack with no offhand equipped and no haste procs equipped (passive haste is okay, just make sure to say how much you had on) against a servant for... basically as long as you're willing to, and then post the combatlog and I'll run my analysis program over it.

Zxahn · 11/16/07, 11:42 PM

3 hours, 3000 swings with jin'rohk later

combat log is here

protspec so no flurry, only haste was from the sword itself bringing it down 3.6 speed exactly

I tried using the cancel macro thingy but it wouldnt work, had a go at canceling it manually but found you cant cancel the buff by right clicking it.

• Aldriana · 11/17/07, 12:22 AM

Great data, thanks.

149 periods of uptime, including at least 183 procs in 3000 swings works out to a proc rate somewhere between 5.24 and 6.96%, aka somewhere between .85 and 1.13 PPM.

So, this still overlaps both ranges (that is, both the fixed-percent range and the PPM range) of the Talon data; so we need either significantly more Talon data, or, perhaps easier, a decent-sized data set with a really fast (2.0 or under, and 1.5 would be better) weapon.

Zxahn · 11/17/07, 10:28 AM

20 mins of spamstring came up with 52% uptime with my jin'rohk, not too bad imo

Zips · 11/17/07, 12:48 PM

While this post may have a limited use for theorycrafting with Executioner, Surlybeve had Executioner on his MH Warglaive and Mongoose on his offhand. From what I saw it was equal or slightly inferior to Mongoose.

Surlybeve - WWS

Illundai · 11/17/07, 1:13 PM

I think Rogues are the one class it's going to be a hard choice, seeying they get ap, crit, haste and dodge from Mongoose. While Warriors and Shamans only get crit, dodge and haste. I'd guess Executioner is better for both Shamans and Warriors, although it's hard for me to let go of my double mongoose procs

.

Avyrana · 11/18/07, 12:14 AM

At first, a general question:

as a dw warrior, you only make use of BT, HS an WW.
BT doenst require a weapon, and HS just adds a fix amont of damage to your MH.

Only WW uses your melee weapons for an extra attack (--> improves the proc chance). since 2.3, both weapons are used.
in my offgear, i use dragonstrike an talon of azshara, both with 2.70 speed.

regarding this, there is no difference in enchanting executioner/mongoose respectively mongoose/executioner, since as a warrior you doesnt use your MH more frequently.

this is only incorrect, if the enemy's hp is lower than 20%, cuz if you execute then, you use the MH more often than the OH.

to sum up: for warriors the question whether enchanting your MH or OH with a procing enchant, isnt that important compared to rouges.
therefore i cant subscribe executioner/potency to be the best combo for warriors.

----------

afawk, executioner cant stack but refreshes the effect. consequently its "useless" to enchant it on both weapons.

the only question remaining for warriors is:
higher dps boost with:

a) executioner + mongoose (no matter which one's enchanted on the MH and OH)

b) executioner + potency
c) mongoose + potency
(since the 20% thing with execute, potency should be on the OH then)
d) double mongoose (since 2.3 changes)

i estimate option a) to be the best one, but thats only a personal pow.it was mentioned before, that mongoose improves with executioner up, as potency does. however, does the combination of ex+mo surpass doulbe mo?

correct me, if im wrong, but for warriors its enough to do autohits to servants in blasted lands with each of these enchant-combos and point out, with which one there was dealt the highest amount of damage for a longer period of time.

Evy · 11/18/07, 2:11 AM

Originally Posted by Avyrana

At first, a general question:

as a dw warrior, you only make use of BT, HS an WW.
BT doenst require a weapon, and HS just adds a fix amont of damage to your MH.

Only WW uses your melee weapons for an extra attack (--> improves the proc chance). since 2.3, both weapons are used.

I might be wrong, but since both PPM and %-based proc are still plausible in theory, doesn't it mean that it won't necessarily improve your proc chance when you whirlwind? Or was it something I missed and we've seen it proc off WW?

Side note on testing: Whenever my guildie with executioner logs on, I'll be putting it on my [Shard of Azzinoth] to try to fill in the testing gaps for <2.0 weapons (I'd rather not spend the gold on re-enchanting one of my crappier 1.8 or OH 1.4-1.5 daggers). I'll try to run tests tomorrow or whenever I get it. Should I just be logging /combatlog? Or should I use mods like procwatch as well?

Edit: Nvm, read back and answered my own question.

Nateo · 11/18/07, 2:54 AM

Here's a ~40 minute test using a Brutalizer (1.6 speed) on Blasted Lands mobs. Would've been longer, but I don't have any way to keep myself from being kicked by the server when I leave the house for 7 hours

Executioner Proc Rate Combat Log 1

Working on another (hopefully) 1 hour sample now, so that should give you guys some decent sample sizes for a faster weapon with Executioner.

Out of curiosity, what programs/scripts do you guys use to parse these logs? Someone mentioned that even without having a way to remove the Executioner buff during the logging, they would still be able to get reasonable results from those combat logs. I guess I just want to make sure procs while the buff is still up is somehow being taken into account -- whether through a simple double count when the duration of the buff is >15 seconds or some other way I'm not yet aware of.

Anyways, hope this helps -- let me know if you need anything else!

Zure · 11/18/07, 3:03 AM

Originally Posted by Avyrana

At first, a general question:

as a dw warrior, you only make use of BT, HS an WW.
BT doenst require a weapon, and HS just adds a fix amont of damage to your MH.

Only WW uses your melee weapons for an extra attack (--> improves the proc chance). since 2.3, both weapons are used.
in my offgear, i use dragonstrike an talon of azshara, both with 2.70 speed.

regarding this, there is no difference in enchanting executioner/mongoose respectively mongoose/executioner, since as a warrior you doesnt use your MH more frequently.

If you are not hit capped, Heroic Strike will cause your main hand to hit more often by converting a white hit (24% base miss) to a yellow hit (5% base miss).

• Aldriana · 11/18/07, 4:02 AM

Originally Posted by Evy

I might be wrong, but since both PPM and %-based proc are still plausible in theory, doesn't it mean that it won't necessarily improve your proc chance when you whirlwind? Or was it something I missed and we've seen it proc off WW?

PPM doesn't mean it doesn't proc off instant attacks - it means that the proc rate depends on your current weapon speed.

Originally Posted by Nateo

Here's a ~40 minute test using a Brutalizer (1.6 speed) on Blasted Lands mobs. Would've been longer, but I don't have any way to keep myself from being kicked by the server when I leave the house for 7 hours

Executioner Proc Rate Combat Log 1

Working on another (hopefully) 1 hour sample now, so that should give you guys some decent sample sizes for a faster weapon with Executioner.

Ran the numbers on this, and the results are, um, interesting. 1449 total swings, 42 uptime periods, at least 58 procs. Hence, proc rate lies between 3.00 and 5.01%, or 1.12 and 1.88 PPM

Now, why is this interesting? Well, consider the two options:

1) Executioner is a fixed % enchant. Lets look at the data we have so far on this:

With Jin'rokh, we found the range to be most likely between 5.3 and 7.0%
With Talon, we found the range to be most likely between 3.4 and 6.6%
With Brutalizer, we found the range to be most likely between 3.0 and 5.0%

Note that there is no possible percentage that lies within all 3 of those ranges. Now, normally this would be evidence that it's not, in fact, a fixed percent, and thus most likely is a PPM. Except that...

2) Executioner is a PPM enchant. Looking at the data we have so far:

With Jin'rokh, we found PPM to most likely be between .86 and 1.12.
With Talon, we found PPM to most likely be between .76 and 1.47
With Brutalizer, we found PPM to most likely be between 1.12 and 1.88

Note that these don't really overlap very well either; and the one place they do touch (1.12 PPM) is sort of a wierd proc rate for an item to have.

So, what does this mean? Well, it means either

A) One of our data sets is somewhat statistically abnormal (basically, we got a flukish data set with a proc rate outside the normal range), or
B) The proc behavior is something other than PPM or fixed% - and if anyone has any idea what that might be, I'd love to hear it.

Case A can be easily diagnosed by just taking more data sets and seeing if, at least, *most* of the data sets agree on the proc rate. If we can't get things to settle down with another couple data sets, then we'll have to come up with a theory for option B.

Out of curiosity, what programs/scripts do you guys use to parse these logs? Someone mentioned that even without having a way to remove the Executioner buff during the logging, they would still be able to get reasonable results from those combat logs. I guess I just want to make sure procs while the buff is still up is somehow being taken into account -- whether through a simple double count when the duration of the buff is >15 seconds or some other way I'm not yet aware of.

I use a python program to rip through logs and pull out proc information. I handle refreshing procs by assuming anything over 16 seconds involves 2 procs, anything over 31 seconds involves 3, and so on. For this, I'm just using a simple version that counts attacks and procs and does the basic statistics involved to count things out; I do have more sophisticated variants to deal with haste procs, dual wielding, and the like (though the answers tend to be less accurate the more complexity you introduce to the system). Anyway; code follows if anyone feels like messing with it. It's uncommented, but, as it's python, it's still pretty readable.

speed = 2.7
input = open("exec2.txt", "rb")

def countProcs(values):
  duration = 15

  procs = 0

  for value in values:
    temp = value
    while temp > 1:
      procs = procs + 1
      temp = temp - duration

  return procs    

def getTime(timeString):
  splitString = timeString[5:17].split(":")
  return 3600*int(splitString[0]) + 60*int(splitString[1]) + float(splitString[2])

attacks = 0
procStart = 0
procDur = []
procSpacing = []

while True:
  line = input.readline()
  if line == "":
    break

  location = line.find("You gain Executioner")
  if location > 0:
    procSpacing.append(getTime(line)-procStart)
    procStart = getTime(line)

  location = line.find("Executioner fades from you.")
  if location > 0:
    procDur.append(getTime(line) - procStart)

  location = line.find("You hit Servant")
  if location > 0:
    attacks = attacks + 1

  location = line.find("You crit Servant")
  if location > 0:
    attacks = attacks + 1

procs = countProcs(procDur)

procRate = float(procs) / attacks
procVar = (procRate * (1 - procRate) / attacks)**.5

minProc = procRate - 1.96*procVar
maxProc = procRate + 1.96*procVar

print "Haste Procs: " + str(procs)
print "Uptime Periods: " + str(len(procDur))
print "Total Swings: " + str(attacks)
print "Proc Rate: " + str(procRate)
print "Proc Std. Dev: " + str(procVar)
print "Proc rate lies between " + str(minProc) + " and " + str(maxProc) + "\n"

print "PPM Values"
print "-----------------------------"
print "Mean: " + str(60*procRate/speed)
print "Min: " + str(60*minProc/speed)
print "Max: " + str(60*maxProc/speed)

Nateo · 11/18/07, 4:10 AM

2nd set of logs should be done in about 10-15 minutes (going for the 2 hour mark on this one). And just to be sure -- you are changing that Speed = 2.7 to 1.6 for Brutalizer correct? (I'm sure you are, just want to make sure all our bases are covered)

Hopefully the new data will help out a bit!

EDIT: 2-hour log in Blasted Lands (same setup as first 40-minute log)
Executioner Proc Rate Combat Log 2

• Aldriana · 11/18/07, 4:21 AM

Yeah, I reran the numbers on Talon as the last thing I did before copying that over. For the Brutalizer data set I had speed set to 1.6, and for the Jin'rohk data it was 3.7/(1+45.*13/20500).

• Aldriana · 11/18/07, 5:43 AM

With the new data set:

125 uptime periods, 165 procs, 4614 hits. Proc rate thus lies between 3.04% and 4.11%, or 1.14 to 1.54 PPM.

Since these are the same setup as the previous data set, we can combine them to obtain greater precision. This yields a total of:

167 uptime periods, 223 procs, 6063 hits. Proc rate thus lies between 3.2% and 4.2%, or 1.2 to 1.55 PPM.

So, um... any ideas what's going on? Since the ranges for Brutalizer are now totally disjoint from the range for Jin'rohk. The only thing I can think of is that it's a fixed percent proc, with a different percentage for 1H and 2H weapons? Seems unlikely, but I don't really have any other ideas.

Warhound · 11/18/07, 6:06 AM

I have a very crazy theory in mind, Please refrain from ripping me if im wrong =P

Did the testers spec out of flurry before recording data ? From my understanding, which might be incorrrect, flurry should be screwing some of the calculations

Nateo · 11/18/07, 6:10 AM

Originally Posted by Warhound

I have a very crazy theory in mind, Please refrain from ripping me if im wrong =P

Did the testers spec out of flurry before recording data ? From my understanding, which might be incorrrect, flurry should be screwing some of the calculations

My spec is the same as my Armory -- no flurry, 8/5/48 Protection, using Brutalizer with shield OH, no haste items or effects.

EDIT: An idea/question came to me and I didn't want to use a new reply to post it so hopefully someone sees this and can answer.

I was always under the impression that if a proc is PPM-based, it would have a certain % chance to proc based off of the base weapon speed of whatever weapon you have equipped -- before any haste effects were counted. If that's correct, then trying to derive a PPM off the Jin'rokh data where we already counted the haste on the weapon toward the base weapon speed would be slightly faulty wouldn't it?

Just throwing some ideas out there to try to resolve this. Although I get the feeling I'll just end up enchanting my Cataclysm's Edge and spending a few hours out with that

songster · 11/18/07, 6:47 AM

Originally Posted by Aldriana

With the new data set:

125 uptime periods, 165 procs, 4614 hits. Proc rate thus lies between 3.04% and 4.11%, or 1.14 to 1.54 PPM.

Since these are the same setup as the previous data set, we can combine them to obtain greater precision. This yields a total of:

167 uptime periods, 223 procs, 6063 hits. Proc rate thus lies between 3.2% and 4.2%, or 1.2 to 1.55 PPM.

So, um... any ideas what's going on? Since the ranges for Brutalizer are now totally disjoint from the range for Jin'rohk. The only thing I can think of is that it's a fixed percent proc, with a different percentage for 1H and 2H weapons? Seems unlikely, but I don't really have any other ideas.

Also, the odd one out seems to be the Brutaliser rather than the Talon. Could it be that they've bumped up proc rates for sword/board relative to dual wield? To answer that we really need a much tighter Talon dataset, though. Worth looking into, though.

Karmon · 11/18/07, 6:52 AM

Originally Posted by Aldriana

So, um... any ideas what's going on?

Originally Posted by Aldriana

I handle refreshing procs by assuming anything over 16 seconds involves 2 procs, anything over 31 seconds involves 3, and so on.

Hm may be I am wrong, but for anything without internal cooldown that should be ..at least 2 procs..
May be thats screwing your number. Just a guess though

Rishina · 11/18/07, 6:55 AM

If they have bumped up the sword and board proc rate. Nateo should test with brutalizer and another weapon and then with brutalizer and a shield, see if theres any difference.

kerrigor · 11/18/07, 6:57 AM

A slight sidedrift from this fun DPS talk, but is there any data on using Executioner as a threat boost enchant for tanking? I've so far been unable to find any solid data, and have no idea where to start testing myself. Help on either matter would be appreciated.

Nateo · 11/18/07, 6:58 AM

Originally Posted by Rishina

If they have bumped up the sword and board proc rate. Nateo should test with brutalizer and another weapon and then with brutalizer and a shield, see if theres any difference.

I'll try to do some samples with Brutalizer/weapon by tomorrow. Only problem is I'm not quite hit capped with DW yet, so not sure how good/useful that data will be. I suppose -20 levels should be fine though...

EDIT:

Originally Posted by kerrigor

A slight sidedrift from this fun DPS talk, but is there any data on using Executioner as a threat boost enchant for tanking? I've so far been unable to find any solid data, and have no idea where to start testing myself. Help on either matter would be appreciated.

I actually enchanted my Brutalizer with Executioner for the single purpose of using it as my threat weapon. I don't have a final word on it yet, because of 1) Not quite sure on the actual proc rate / mechanics (see above) and 2) Latency/GCD issues made it hard to quantify during raids this week.

hannigaholic · 11/18/07, 7:01 AM

Originally Posted by Aldriana

So, what does this mean? Well, it means either

A) One of our data sets is somewhat statistically abnormal (basically, we got a flukish data set with a proc rate outside the normal range), or
B) The proc behavior is something other than PPM or fixed% - and if anyone has any idea what that might be, I'd love to hear it.

To be honest I think the most likely explanation is that we simply don't have enough data for the long-term proc rates to have evened out yet. I'm sure somebody, with a better grasp of statistics than me, will be able to work out how many swings we'd need to be even 90% certain of the results but I seem to remember that, when reading about the testing to determine boss miss rates, it was in the order of tens of thousands of swings, minimum.

I would imagine that with more data we would see either the procs per minute or the simple proc rate (and my money would be on the procs per minute, given the behaviour of other high end enchants) come down to a much smaller range.

Mideci · 11/18/07, 7:19 AM

Honestly, if 3 data sets are large enough and disjoint, there >>is<< enough data to tell you something is 'wrong'. You don't need long-term data...

Clearly, the disjoint data sets mean there is at least a second-order mechanic at work of some kind. What that is, I don't currently have a great theory.