A Call to Arms - MoP Mechanics Testing

[mythor]

when using warrior 85 beta dodge data:

Code: Select all

d2_try1_fit =
     General model:
     d2_try1_fit(x) = 1./(1./C+k./x)
     Coefficients (with 95% confidence bounds):
       C =       88.64  (86.4, 90.87)
       k =      0.9541  (0.9524, 0.9558)

d2_try1_gof =
           sse: 7.1744e-05
       rsquare: 1.0000
           dfe: 9
    adjrsquare: 1.0000
          rmse: 0.0028


That gave me the feeling k was still 0.956. So with k fixed:

Code: Select all

d2_try2_fit =
     General model:
     d2_try2_fit(x) = 1./(1./C+0.956./x)
     Coefficients (with 95% confidence bounds):
       C =       91.09  (90.31, 91.86)

d2_try2_gof =
           sse: 1.2137e-04
       rsquare: 1.0000
           dfe: 10
    adjrsquare: 1.0000
          rmse: 0.0035


Parry with k = 0.956

Code: Select all

p1_try3_fit =

     General model:
     p1_try3_fit(x) = 1./(1./C+0.956./(x./243.7))
     Coefficients (with 95% confidence bounds):
       C =       238.7  (238.5, 239)

p1_try3_gof =

           sse: 1.3492e-05
       rsquare: 1.0000
           dfe: 14
    adjrsquare: 1.0000
          rmse: 9.8169e-04


[theckhd]

Playing around with matlab files provided by Theck I find the following for 85 prot war to make the results fit:

Code: Select all

k = 0.956
Cp = 238.7
Cd = 91


k still seems to be the cataclysm value. Cp is not that far off but Cd seems to be pretty different. I think I need more data. Is there more warrior data available somewhere?

Just took some of my own data on beta with a L85 pre-made warrior:
https://docs.google.com/spreadsheet/ccc ... vSlE#gid=4

Confirmed that your dodge formula is different, and still using the cataclysm k coefficient:

Code: Select all

General model:
     f(x) = 1/(1/C+0.956/x)
Coefficients (with 95% confidence bounds):
       C =       91.57  (91.06, 92.07)

Goodness of fit:
  SSE: 0.0001821
  R-square: 1
  Adjusted R-square: 1
  RMSE: 0.003606


Parry data shows the same k discrepancy, given the known value of str->parry conversion factor. It's not clear if the parry cap is subtly different than the paladin one, or if we're just limited by noise and a small data set.

Code: Select all

General model:
     f(x,y) = 1/(1/C+k/(x/243.9+y))
Coefficients (with 95% confidence bounds):
       C =       238.4  (236.8, 239.9)
       k =      0.9556  (0.9551, 0.9562)

Goodness of fit:
  SSE: 0.0001639
  R-square: 1
  Adjusted R-square: 1
  RMSE: 0.003551


[benebarba]

Parry data is giving me:

Code: Select all

General model:
     f(x,y) = 1/(1/C+0.885/(x/a+y))
Coefficients (with 95% confidence bounds):
       C =       237.2  (234, 240.4)
       a =       243.8  (243.3, 244.3)

Goodness of fit:
  SSE: 0.0546
  R-square: 1
  Adjusted R-square: 1
  RMSE: 0.03649


That doesn't really narrow down C much better than our earlier data sets though. If we knew a exactly that would help - didn't GC give us that number?

Not sure when they'll be shutting down the beta (assuming tuesday), but I can try to get some more data, if you think it'll help. Not sure what kind of spread I can get, or if you have a particular area that's poorly fit I can try to construct sets with numbers in that range.

[Klaudandus]

sept 11, if I recall correctly

[Nooska]

sept 11, if I recall correctly

That is indeed the date they've announced as their expected shutdown (subject to change)

[theckhd]

Parry data is giving me:

Code: Select all

General model:
     f(x,y) = 1/(1/C+0.885/(x/a+y))
Coefficients (with 95% confidence bounds):
       C =       237.2  (234, 240.4)
       a =       243.8  (243.3, 244.3)

Goodness of fit:
  SSE: 0.0546
  R-square: 1
  Adjusted R-square: 1
  RMSE: 0.03649


That doesn't really narrow down C much better than our earlier data sets though. If we knew a exactly that would help - didn't GC give us that number?

Not sure when they'll be shutting down the beta (assuming tuesday), but I can try to get some more data, if you think it'll help. Not sure what kind of spread I can get, or if you have a particular area that's poorly fit I can try to construct sets with numbers in that range.

The more data at the extremes, the better. Here's the surface fit plot I have right now:

The places for improvement would be:

1) high parry rating, low str (i.e. 9%+ parry from rating, <2k str, maybe holy gear reforged/gemmed for parry) - basically to fill out the predrparry axis

2) high parry rating, increasing STR (i.e. take the above gear set and start swapping in prot/ret pieces) - fills out the last side of the square

3) push for maximum parry - get as much high-parry, high-str gear as possible to try and push your overall parry% as high as possible. This could include using the Deathwing stacking STR trinket and parry on-use trinkets. I may try and do a little of this today, if I can find a parry on-use trinket.

4) fill in more data points in the center - not as useful, but every little bit helps

[theckhd]

My google-fu is failing me this morning... Can anyone find the blue post that gave us the exact STR->Parry conversion at L85? Or am I misremembering, and there was never such a post?

[benebarba]

My google-fu is failing me this morning... Can anyone find the blue post that gave us the exact STR->Parry conversion at L85? Or am I misremembering, and there was never such a post?

I'm only finding the 4.2 change to 27% (up from 25) of strength to parry rating.

[daishan]

Level 85 Draenei Paladin premade

Holy Gear (all gear reforged to parry, parry gemmed):
Strength: 165
Mastery: 1139 (+6.35%)
Stam: 7380
Dodge: 0 rating, 5.01%
Parry: 1839 rating, 11.26%
Block: 27.62%

Half Holy half Prot gear (all gear reforged to parry, parry gemmed):
Strength: 1738
Mastery: 1112 (+6.20%)
Stam: 7380
Dodge: 547 rating, 7.26%
Parry: 2643 rating, 20.87%
Block: 27.62%

Adding in some Ret gear (all gear reforged to parry, parry gemmed):
Strength: 2760
Mastery: 921 (+5.14%)
Stam: 7380
Dodge: 623 rating, 7.56%
Parry: 2549 rating, 24.54%
Block: 26.50%

More Ret gear (all gear reforged to parry, parry gemmed):
Strength: 3461
Mastery: 1260 (+7.03%)
Stam: 7380
Dodge: 920 rating, 8.71%
Parry: 2587 rating, 27.34%
Block: 28.25%

Ret/Prot gear (all gear reforged to parry, parry gemmed):
Strength: 4234
Mastery: 1146 (+6.39%)
Stam: 7380
Dodge: 920 rating, 8.71%
Parry: 2594 rating, 30.23%
Block: 27.66%


<edit> I was going to copy my Pally over again so I could have it at level 85 with all the DS hm gear but the character copy looks to be broken atm.
Want we to use the level 86 copy I have already for the high strength gear sets or will doing it with a Level 86 just make it over complicated?

[theckhd]

<edit> I was going to copy my Pally over again so I could have it at level 85 with all the DS hm gear but the character copy looks to be broken atm.
Want we to use the level 86 copy I have already for the high strength gear sets or will doing it with a Level 86 just make it over complicated?

Can't use 86 data - the STR->Parry conversion will be different, and I don't know the ratio.

Could you give me the naked info for that Dranei paladin though?

[theckhd]

While you're at it with that Holy set (which I assume has a lot of haste), could you spec EF and take a quick log of a couple EF casts with high haste? I don't think the tick interval is affected by spell haste, but I don't know that I've seen a confirmation of that either way.

[daishan]

Level 85 Draenei Paladin premade

Naked
Strength: 165
Mastery: 0 (+0.00%)
Stam: 179
Dodge: 0 rating, 5.01%
Parry: 0 rating, 3.68%
Block: 21.52%

<edit>
EF Logs http://www.worldoflogs.com/reports/rt-s ... details/3/
1st and 2nd EF casts with zero haste on gear 3rd and 4th EF casts with 2326 (+18.16%) haste

According to the in game tooltip that much haste reduces the tick interval to 2.54 sec.

[benebarba]

moar dataz!

human 85 protection pally, Holy gear (no strength)
Strength: 164
Parry: 301 rating, 4.95%

Parry: 538 rating, 5.94%

Parry: 733 rating, 6.75%

Parry: 953 rating, 7.66%

Parry: 1014 rating, 7.91%

Parry: 1455 rating, 9.71%

Adding strength+parry pieces:
Str: 737, Parry: 1609 rating, 12.81%
Str: 1451, Parry: 1923 rating, 17.03%
Str: 1964, Parry: 2056 rating, 19.62%


I'll add more as I get it

[theckhd]

With the new data points, along with the exact str->parry conversion ratio at L85 (243.6), this is what the fit looks like:

Code: Select all

General model:
     f(x,y) = 1/(1/C+0.885/(x/243.6+y))
Coefficients (with 95% confidence bounds):
       C =       236.1  (235.4, 236.8)

Goodness of fit:
  SSE: 0.07522
  R-square: 1
  Adjusted R-square: 1
  RMSE: 0.03483


So our best estimate of the parry cap Cp is 236.1 +/- 0.7

[mythor]

After plugging in some more data from a level 85 warrior from beta:

Code: Select all

General model:
    dodge_fit(x) = 1 / (1 / Cd + 0.956 / x)
Coefficients (with 95% confidence bounds):
    Cd =        91.42  (91.01, 91.83)

dodge_fit_GoF =
           sse: 3.1974e-04
       rsquare: 1.0000
           dfe: 24
    adjrsquare: 1.0000
          rmse: 0.0037


Code: Select all

General model:
    parry_fit(x) = 1 / (1 / Cp + 0.956 / (x / 243.6))
Coefficients (with 95% confidence bounds):
    Cp =        237.1  (237.1, 237.2)

parry_fit_GoF =
           sse: 1.5810e-06
       rsquare: 1.0000
           dfe: 14
    adjrsquare: 1.0000
          rmse: 3.3604e-04


Cp_paladin and Cp_warrior are so close I want them to be the same. After 5.0.4 hits EU I'll use a fresh dataset to confirm these findings.