A Call to Arms - MoP Mechanics Testing

xstratax wrote:How many Mastery points do you need to get an accurate model of the coefficient? And how much of a difference should there be, I think I could possibly get a bunch of plot points between my Ret, PvP, and Tank items.

It only took us ~16 data points last time:
https://docs.google.com/spreadsheet/ccc ... vSlE#gid=1

Ideally we want points ranging from naked to as high a mastery value as you can get. Ideally we'd want to reach 25% mastery, if possible. The higher a mastery% you can achieve, the more accurately we'll be able to determine Cb.

If you e-mail/PM me your google account, I can give you access to that document and you can enter the data directly.

Thx for the explanation Theck.

I'll leave beta live logging over night to http://www.worldoflogs.com/reports/rt-g706k3rd2dl8wfsb/ with Colossus enchanted on a 2.6 speed wep.

New test request, #41. Should be a pretty simple one, provided you can tailor the gear to meet the requirements (0% hit, 15% exp)

Based on xtratax's data, the block DR equation hasn't changed.

Klaud's data set:

Code: Select all

General model:
     f(x) = 13+1/(1/C+0.885/x)
Coefficients (with 95% confidence bounds):
       C =       149.1  (148.7, 149.4)

Goodness of fit:
  SSE: 0.0006184
  R-square: 1
  Adjusted R-square: 1
  RMSE: 0.006421


xtratax's data set:

Code: Select all

General model:
     f(x) = 13+1/(1/C+0.885/x)
Coefficients (with 95% confidence bounds):
       C =       149.3  (149.1, 149.5)

Goodness of fit:
  SSE: 0.00028
  R-square: 1
  Adjusted R-square: 1
  RMSE: 0.004641


Combining the two:

Code: Select all

General model:
     f(x) = 13+1/(1/C+0.885/x)
Coefficients (with 95% confidence bounds):
       C =       149.2  (149.1, 149.4)

Goodness of fit:
  SSE: 0.0009435
  R-square: 1
  Adjusted R-square: 1
  RMSE: 0.005704


May update Cb from 149.1 to 149.2, but that's not a very significant change.

theckhd wrote:We'll probably need more data to nail this one down. Unfortunately, to determine it to the required accuracy (+/- 0.5% would be ideal), it's going to take 15k melee swings...

Looks like I'll be leaving my beta running overnight. Should I continue to avoid Haste, or would using Hasted swing times help rule out other mechanics? Also do you want 1 hand weapons still, or should I go 2 hand for the 3.6 base swing speed?

EDIT:

Nvm Daishan beat me to it

Hehe I stole the test

http://www.worldoflogs.com/reports/rt-g706k3rd2dl8wfsb/
colossus enchant with 2.6 speed wep.
Didn't quite get 15k hits before the weapon broke.
1140 procs / 13564 melee attacks = 8.4% proc rate
Looks pretty close to 2PPM,
If we add in Xstratax's data (153+1140)/(1361+13564)= 8.66% that's almost spot on.

Will we need any data from a slow weapon to confirm?

daishan wrote:Will we need any data from a slow weapon to confirm?

It would be nice to have a little data from a 3.6-speed weapon, yeah. Luckily we should only need around 1000 melee swings this time, since we just need to show that 8.66% is outside the confidence interval.

Kk, I'll do some after work unless someone beats me to it.

http://www.worldoflogs.com/reports/rt-i ... details/0/

3.6 speed Weapon
205/1254=16.3%
2*3.6/60=12%

If I've done that right it looks to be quite a long way out, though guess it is a small sample size.

Without any additional haste, the cooldown of Crusader Strike is reduced to 3.46 seconds when timewarp and Bloodlust are applied. I did not have any haste rating on my gear at the time. I mention this because you sounded fairly sure that it wouldn't when we were chatting a couple of weeks ago.



Unholy Aura (10% melee attack speed) does not change the cooldown.

So with a bit of haste, bloodlust will give us almost (or exactly, depending on your haste level) full ShoR uptime for the 40 seconds. Which is kinda crazy tbh.

Pyrea wrote:Without any additional haste, the cooldown of Crusader Strike is reduced to 3.46 seconds when Bloodlust is applied. I mention this because you sounded fairly sure that it wouldn't when we were chatting a couple of weeks ago.

Unholy Aura (10% melee attack speed) does not change the cooldown.

So with a bit of haste, bloodlust will give us almost (or exactly, depending on your haste level) full ShoR uptime for the 40 seconds. Which is kinda crazy tbh.

I'm not sure why I would have said that, Bloodlust should apply perfectly well to CS. That said, it doesn't give us 100% uptime on SotR. Even assuming we have a generous ~0.45 HP/s at 0% haste, 30% haste only raises that to 0.585, far below the 1.00 threshold you'd need.

<edit> At hit/exp cap, you'd get 1.2/4.5+1/6.75=~0.42 HP/sec at 0% haste. With bloodlust and x% haste, you'd generate 0.42*1.3*(1+x/100). For that to equal 1, x=100*(1/(0.42*1.3)-1) = 83.2% haste

and there's a 1-second limit to the GCD, so at best we could get to CS-J-CS-?- and pray for really good Grand Crusader procs.

daishan wrote:http://www.worldoflogs.com/reports/rt-imxyoggbbntrpfta/details/0/

3.6 speed Weapon
205/1254=16.3%
2*3.6/60=12%

If I've done that right it looks to be quite a long way out, though guess it is a small sample size.

That's actually quite a ways out. Far enough that I'm not sure what to make of it. The 15k samples with a 2.6-speed weapon more or less nailed it down to a 2 PPM enchant. This result with a 3.6-speed weapon falls about halfway between 2 PPM (12%) and 3 PPM (18%). It would fit with a 3PPM enchant using normalized weapon speed (3*3.3/60=16.5%), but that's sort of uncommon, and wouldn't agree with the 2.6-speed data either (3*2.4/60=12%, far outside the CI).

Best guess at this point is that it's a 2 PPM enchant and that our small sample size in the 3.6-speed data is causing the error. It'd be nice to get a few thousand more swings at 3.6-speed if you don't mind leaving it running overnight again.

I'll leave it logging to http://www.worldoflogs.com/reports/rt-v ... details/0/ for a few hours at least.
I double checked to make sure I hadn't got the wrong weapon speed but it's definitely heroic Experimental Specimen Slicer speed 3.6

Edit:
Servers went down so only got a couple of hours.
350/2363=14.81%
adding in logs from 20/08/12
(205+350)/(1254+2363)=15.34%

Getting closer to 2PPM but still a long way out, I'll try to do some more when I have chance.

I know its not mechanics per se, but tlitp asked me to snipe some armor values from a progression of beasts and while I've sent it as a pm, it hasn't been read yet, so in case others are also waiting for them, I'll repost them here:

Lvl 86 (Krasari Hissing Wasp, Violetdown Songspinner)
Armor: 12,810

Lvl 87 (Murkscale Striker, Dustback Mushan)
Armor: 14,530

Lvl 88 (Swooping Plainshawk, Mountain Yak, Painted Stag)
Armor: 16,250

Lvl 89 (Swamp Rodent)
Armor: 17,960

Lvl 90 (Amberscale Basilisk)
Armor: 19,680

Lvl 91 (Cracklefang(elite))
Armor: 21,400

I have only found 2 level 92 beasts - Carapin (who is dead in VoEB) and Alani, the dragon (is classified as beast) who is uncastable on - at least untill I get 9 more skyshards (which I don't realistically expect in the beta) or I encounter someone else with them.

Thanks Nooska. With that data and our known value for L93 armor (24835), we can fairly accurately determine the armor function:

Code: Select all

General model:
     f(x) = 1718*x+b-134938
Coefficients (with 95% confidence bounds):
       b =     -0.1429  (-2.556, 2.271)

Goodness of fit:
  SSE: 40.86
  R-square: 1
  Adjusted R-square: 1
  RMSE: 2.61

armor = 1718*lvl-134948.1429

It's interesting that there's still a small amount of error in this. For example:

Code: Select all

         lvl       armor     formula
          86       12810       12810
          87       14530       14528
          88       16250       16246
          89       17960       17964
          90       19680       19682
          91       21400       21400
          93       24835       24836

It's almost as if they rounded to the nearest 5 except for the L89 data point. The weird thing is that these armor values permit no exact linear match. Higher-order polynomials seem to do no better. I think there's some weird rounding going on within the function.

I can get a slightly better fit by doing a little rounding:

guess=round(343.521.*lvl-26980.541).*5

Code: Select all

         lvl       armor       guess       armor-guess
          86       12810       12810           0
          87       14530       14530           0
          88       16250       16245           5
          89       17960       17965          -5
          90       19680       19680           0
          91       21400       21400           0
          93       24835       24835           0

It's really the L88 and L89 data points that are screwing it up. For some reason, the L88 point is high and the L89 point is low. There might be a piecewise break there. Doing that I can get exact results:
L86-88: guess1=round(344.*lvl-27022).*5
L89-93: guess2=round(343.7.*lvl-26997).*5

Code: Select all

         lvl       armor       guess         a-g      guess1        a-g1      guess2        a-g2
          86       12810       12810           0       12810           0       12805           5
          87       14530       14530           0       14530           0       14525           5
          88       16250       16245           5       16250           0       16245           5
          89       17960       17965          -5       17970         -10       17960           0
          90       19680       19680           0       19690         -10       19680           0
          91       21400       21400           0       21410         -10       21400           0
          93       24835       24835           0       24850         -15       24835           0

As you can see, guess1 fits the 86-88 data points perfectly, while guess2 fits the 89-93 data points perfectly. If the L92 armor value ends up fitting this pattern, then these should be pretty close.

Just to rule out a typo in my listings I'll go and snipe L88 and L89 again in a few minutes (just have to find some)

Edit: Double checked l88 and l89, they are correct in my initial copy/paste from my notepad doc.

http://www.worldoflogs.com/reports/rt-7 ... details/0/
Been logging for most of the time I've been at work it's still going so numbers will change.
1051/(6441+456)=15.24%

adding in the previous 2 logs
(205+350+1051)/(1254+2363+6897)
1606/10514=15.27%

Still looking a very long way out, besides me putting down the wrong weapon speed (which I've triple checked) is there any other daft things I could be doing wrong?
Even checked the combat log swing times and they match with 3.6 speed, all logs have been naked using a heroic Experimental Specimen Slicer with zero haste.

Code: Select all

[10:06:23.437] Zaishan hits Training Dummy 1 (O: 5800)
[10:06:27.031] Zaishan hits Training Dummy 1 (O: 6502)
[10:06:30.632] Zaishan hits Training Dummy 1 (O: 7631)
[10:06:34.254] Zaishan hits Training Dummy 1 (O: 6068)
[10:06:37.844] Zaishan hits Training Dummy 1 (O: 6712)
[10:06:41.452] Zaishan hits Training Dummy 1 (O: 7039)
[10:06:44.980] Zaishan hits Training Dummy 1 (O: 6213)
[10:06:48.571] Zaishan hits Training Dummy 1 (O: 7493)
[10:06:52.295] Zaishan hits Training Dummy 1 (O: 6111)
[10:06:55.823] Zaishan hits Training Dummy 1 (O: 6021)
[10:06:59.412] Zaishan hits Training Dummy 1 (O: 6740)
[10:07:03.029] Zaishan hits Training Dummy 1 (O: 7856)
[10:07:06.685] Zaishan hits Training Dummy 1 (O: 5728)

daishan wrote:http://www.worldoflogs.com/reports/rt-7vptbj3aob5f5gla/details/0/
Been logging for most of the time I've been at work it's still going so numbers will change.
1051/(6441+456)=15.24%

adding in the previous 2 logs
(205+350+1051)/(1254+2363+6897)
1606/10514=15.27%

Still looking a very long way out, besides me putting down the wrong weapon speed (which I've triple checked) is there any other daft things I could be doing wrong?

Not that I can think of. There must be something we're missing.

With a 3.6-speed 2h, we're getting 15.27% +/- 0.69%.
With a 2.6-speed 1h, we're getting 8.66% +/- 0.45%.

It's almost as if the enchant is 2 PPM for a 1H (8.67%), but 2.5 PPM for a 2H (15%).

<edit> What I think we need is a third set of conditions. Grab a Wooden Mallet, which is the fastest 2H-weapon that's easily accessible, and see what the proc rate is with that.

<edit2> strike that, you won't be able to enchant Colossus on it. Hrm...

<edit3> I have an idea. I'm going to put Colossus on a Ti'tahk I have lying around on my warlock and log that for an hour or two.

<edit4> In the meantime, see if you can find a 1.6- or 1.7-speed one-handed weapon to try out. Most of the int/spi PvP maces are fast and high-enough ilvl for the enchant, if you have the currency to buy them.

Found a 1.7 speed dagger on the AH just need to get the pre made to some dummies.

Edit: logging to http://www.worldoflogs.com/reports/rt-k ... details/2/

Edit: oops kind of failed on the logging, luckily Theck's log shows my toon as well http://www.worldoflogs.com/reports/rt-1 ... details/3/

I got a 1.6 with colossus on it. How many hrs of whacking u need?

Probably only 1-2k swings, which is 25-50 minutes with a 1.6 weapon (faster if you have some haste).

Logging with Titahk here: http://www.worldoflogs.com/reports/rt-18luggodiisxb74y/

Also, Klaud, if you have some time, I could really use information from test #41 (which should only take 10 minutes) and #38 (the L90 talents part).

What kind of info do u need?

Test #38: Seal of Insight interactions

• Level: 90
• Gear: naked (white weapon if needed)
• Target: any (low level ideal)
• Attack sequence: With Seal of Insight active, log 10-15 casts of each L90 talent. Repeat the process with Seal of Insight off (use any other seal).
• Goal: 10-15 casts of each L90 talent with SoI active, the same amount with SoI inactive.
• Report: AP, SP, upload log.

Test #41: Expertise->Spell Hit conversion

• Level: 90
• Gear: any, but with exactly 0% hit and 15% expertise
• Target: boss-level dummy (if you can find 2+ boss-level dummies together, all the better)
• Attack Sequence: Consecrate until the world ends
• Goal: 20 Consecration casts, or the first missed tick.
• Report: upload log

Mostly the "10-15 casts of each L90 talent with SoI active, the same amount with SoI inactive." on #38 though, as I think the logs I did where enough for the WoG and EF.
Correct me if I'm wrong just don't want Klaud wasting his time.