The effect of skills on mindamage.

[Witte]

EDIT: FOR MORE ACCURATE ANALYSIS SEE: https://www.planetcalypsoforum.com/forums/showthread.php?t=53122

After some reading in this thread, I decided to do some research. I collected the skill and mindamage of 10 different people, of which 6 all 3 weapon categories, and 4 just their main category.

The skills I collected were: ?WT, Anatomy, IRD, RDA, MMS, Serendipidy, and Wounding. I let matlab calculate what the influence of these skills is on the damage %. Here are the results:

Mindamage gain per skilllevel:

LWT: 0.6263 (1600)
ANA: 0.5028 (2000)
IRD: 0.1079 (10000)
RDA: 0.3936 (2500)
MMS: 0.2485 (4000)
SER: 0.0745 (13300)
WND: 0.2507 (4000)

In the brackets is how much skill you approximatly need to gain 1% mindamage. After adding the 7th person, the result nearly didnt change anymore, so it seems the result is pretty much complete. If you think its not correct you can PM me your skills and mindamage, so I can add them.




Matlab data:

A = [
6142 5486 2981 4408 4754 2886 0000;
1232 5486 2981 4408 4754 2886 0000;
0502 5486 2981 4408 4754 2886 0000;
6047 6272 4005 5237 5446 3991 2516;
0847 6272 4005 5237 5446 3991 2516;
1163 6524 3116 4837 5029 4121 0000;
6460 6524 3116 4837 5029 4121 0000;
0123 6524 3116 4837 5029 4121 0000;
6454 5220 3140 4209 4129 2914 0000;
0001 5220 3140 4209 4129 2914 0000;
0474 5220 3140 4209 4129 2914 0000;
8071 6998 4070 4971 5765 4451 3521;
0001 6998 4070 4971 5765 4451 3521;
1518 6998 4070 4971 5765 4451 3521;
6020 5771 3263 4509 4961 3066 0000;
3819 5771 3263 4509 4961 3066 0000;
1466 5771 3263 4509 4961 3066 0000;
3019 3480 1071 1922 3375 0000 0000;
6095 4945 3595 3454 4248 2590 0000;
5417 4559 2415 3542 4067 2022 0000;
5902 5151 2595 3950 4441 2810 0000;
7501 6681 3989 5790 6132 4062 3712]

B = [
10.00;
07.00;
06.51;
11.75;
08.44;
07.82;
11.10;
07.18;
09.94;
05.82;
06.16;
13.59;
08.55;
09.53;
10.29;
08.91;
07.40;
05.36;
09.30;
08.49;
09.42;
13.53]

C = A\B =

0.6263
0.5028
0.1079
0.3936
0.2485
0.0745
0.2507

 

[Shinobi C]

Looking good

Thx for sharing

 

[sacklitch]

Wooot! is anatomy affecting minimum dmg!?!?. i know what skill i gnna buy

 

[andy5579]

Yep, anatomy helps damage as well as healing-So if you want to be good at heling people-Start killing things

 

[CycoKick]

witte crap at u now other people will buy ana

 

[sacklitch]

that means , ignna save about 50 anatomy chips in my storage

 

[Lu Nix]

sounds resonable, have you tested the same way to find inpact of skils for HA ?
btw. my skiles (mind dmg-33,725; HA-3,9)
LWT: 5473
ANA: 4671
IRD: 2914
RDA: 3729
MMS: 4187 (markmanship ?)
SER: 2206
WND: locked

 

[Torschen]

interesting.
+rep

 

[jdegre]

Hi Witte, great effort there!, +rep...

I think that Kill Strike affects min dmg too. I must have read it somewhere, but specially the fact that it unlocks at a certain min dmg threshold, makes me believe it does affect min dmg.

I've been trying to do the same calcs but never got the time... anyway, my approach was to calculate the % weights for each skill, considering a linear mapping between the whole dmg range (25%-50%) and a wheighted average of all involved skills, capped at 10K. This way, your data are equivalente to:

- LWT : 25%
- ANA : 20%
- IRD : 4%
- RDA : 16%
- MMS: 10%
- SER : 3%
- WND : 10%

The problem I see is that if we add up everything we get 88% instead of 100%. This kinda suggests that maybe some other skill is involved too (KS again?).

And last, I've seen people reporting that WND increases min dmg much more than 1% each 4k levels. I think it was Duli who said 1% each 2k levels aprox.

Dunno, more food for thought... hope it helps.

/jdegre.

 

[[DFA]Hasse K]

Hey Witte, great post! Do u or anybody know if combat sense affect min dmg?

 

[Witte]

Hi Witte, great effort there!, +rep...

I think that Kill Strike affects min dmg too. I must have read it somewhere, but specially the fact that it unlocks at a certain min dmg threshold, makes me believe it does affect min dmg.

I've been trying to do the same calcs but never got the time... anyway, my approach was to calculate the % weights for each skill, considering a linear mapping between the whole dmg range (25%-50%) and a wheighted average of all involved skills, capped at 10K. This way, your data are equivalente to:

- LWT : 25%
- ANA : 20%
- IRD : 4%
- RDA : 16%
- MMS: 10%
- SER : 3%
- WND : 10%

The problem I see is that if we add up everything we get 88% instead of 100%. This kinda suggests that maybe some other skill is involved too (KS again?).

And last, I've seen people reporting that WND increases min dmg much more than 1% each 4k levels. I think it was Duli who said 1% each 2k levels aprox.

Dunno, more food for thought... hope it helps.

/jdegre.

Hey old thread is bumped . Havnt payed attention to this.

Your approach suggests there is indeed another skill, and from what I have heard its very likely to be KS. I the avatars I used for my tests didnt have KS. If anyone can send me their skills and mindamage including killstrike, I can add it to the list.

The point that wounding raises more damage, might be because the damage gain wasnt purely caused by wounding. When you gain it by skilling, other skills also gain. And ?WT gains fast, even at high levels. That affects the mindamage allot.

Anyway, the more data I have, the more accurate the result will be, so feel free to send more

 

[Witte]

Overview of results:

All (initial 3 avatars):
0.6262
0.3257
0.2529
0.2454
0.3904
0.2474
0.1915
+Trans:
0.6266
0.4945
0.1214
0.3682
0.2672
0.0844
0.2440
+Dolph:
0.6265
0.4988
0.1067
0.3878
0.2569
0.0787
0.2485
+Gawk:
0.6263
0.5007
0.1079
0.3865
0.2548
0.0790
0.2485
+MD:
0.6261
0.5066
0.1107
0.3880
0.2471
0.0750
0.2494
+Will:
0.6263 (1600)
0.5028 (2000)
0.1079 (10000)
0.3936 (2500)
0.2485 (4000)
0.0745 (13300)
0.2507 (4000)
+Avery:
0.6262
0.5099
0.1123
0.3899
0.2418
0.0726
0.2497
+Lu Nix
0.6260
0.5141
0.1081
0.3909
0.2380
0.0732
0.2508

Seems that by adding more people that MMS goes down in favor of Anatomy

 

[KapokWu]

C = A\B =

Your equation goes: B*C= A , but have y checked it is the right one?
Y should keep observing the same guys later, when they have advanced
further in skills, and see if y get the same C, then y could see
if the relation is linear. Also, some terms could multiply, like
KStrike*Wounding, and so on.

This is a nice piece of work if it gets -uhm- proven.

While waiting the new results y can experiment by leaving out some
rows of data and checking if y get even better fitting C.

 

[falkao]

Sry Witte,

with C= A\B you're solving in Matlab an overdetermined system of equations. What you would need here is something different namely to use linear regression. Doing this with your posted data I get the following standardized coefficients:

Variable Beta
LWT 0.79
ANA 0.22
IRD 0.04
RDA 0.14
MMS 0.07
SER 0.01
WND 0.16


Moreover, SER was not significant and therefore not needed and you can't prove with this if a skill is directly or only indirectly correlated with dmg like ana. I guess you one can do this only with chips. However, if an effect is needed in the model like ana, then there might be other skills that influence dmg not included in the model yet.

 

[Witte]

Sry Witte,

with C= A\B you're solving in Matlab an overdetermined system of equations. What you would need here is something different namely to use linear regression. Doing this with your posted data I get the following standardized coefficients:

Variable Beta
LWT 0.79
ANA 0.22
IRD 0.04
RDA 0.14
MMS 0.07
SER 0.01
WND 0.16


Moreover, SER was not significant and therefore not needed and you can't prove with this if a skill is directly or only indirectly correlated with dmg like ana. I guess you one can do this only with chips. However, if an effect is needed in the model like ana, then there might be other skills that influence dmg not included in the model yet.

Although skillgains suffer linear regression, I dont think the skills itself do. AFAIK the relation between skills and damage is linear.

I got to agree I am totaly no expert in matlab. Infact, this is the first time I used it . If you can explain me how you can calculate it using linear regression i can experiment with it abit myself. BTW what regression curve did you use?

 

[KapokWu]

Witte, y can also use: pinv(A)*B, which gives:

(0.6263 0.5028 0.1079 0.3936 0.2485 0.0745 0.2507)*1.0e-003

As y see, they are quite small ones... If y use linear regression,
y need to get it in the form C0+C1*B. Ask falkao to provide
both C0 and C1... (I have to go now, no time to
calculate them for y).

"pinv" stands for pseudoinverse, see how it is defined by "help pinv".
And congrats in taking the first steps with Matlab, a non-commercial
alternative is Gnu Octave . Anyways, keep paddling!

 

[falkao]

.. If you can explain me how you can calculate it using linear regression i can experiment with it abit myself. BTW what regression curve did you use?

You might need the statistics toolbox to calculate a "Multiple Linear Regression". You should find something under help/content/Statistics toolbox/Linear Models/Multiple Linear regression.

I used SPSS to calculate it and the model was linear i.e.

dmg = a0 + a1*LWT + a2 * ANA + ... + a7 *WND

The model works rather good and the relation seems indeed linear.

The coefficients I have given are the standardized coeffizient. So not the absolute ones. With them you can judge the relative importance of a factor but you can not calculate any predictions.

the absolute ones are:
(Constant) -0,142879822
LWT 0,000626733
ANA 0,000552338
IRD 0,000122171
RDA 0,000403968
MMS 0,000228969
SER 3,12615E-05
WND 0,000248075

 

[Witte]

Ok good work The constant is 0,14%. I think this is just within the error margin caused by rounding of the mindamage. So its hard to tell if the data is complete. I do think it is save to say that IF there is another skill involved, its influence will be very small. BTW, the coefficients you show here seem absolute to me. Its good to see they are pretty close to the results I had. (I forgot to mention the 10^-3 factor though)

 

[falkao]

Ok good work The constant is 0,14%. I think this is just within the error margin caused by rounding of the mindamage.

Yep, I guess this is due to the linear approximation. The coefficients should be similar to yours but with less estimation errors. Matlab is rather clever and knows what to do . However, with lineare regression you get also standard errors of estimated parameters and residuals.

 

[Tor]

I do not think gains can be that exact and it seems that you used direct calculations and not statistical methods to calculate the gains per skill lvl, but if I'm wrong, can you please post the variance.

 

[falkao]

I do not think gains can be that exact and it seems that you used direct calculations and not statistical methods to calculate the gains per skill lvl, but if I'm wrong, can you please post the variance.

here is the full output

Model	 	Unstandardized Coefficients		Standardized Coefficients	t	Sig.
		B	Std. Error	Beta		
1	(Constant)	-0,142879822	0,16781326		-0,851421524	0,408863086
	LWT	0,000626733	2,49725E-06	0,790264635	250,969787	5,60648E-27
	ANA	0,000552338	7,28381E-05	0,216592754	7,583088159	2,53897E-06
	IRD	0,000122171	2,70273E-05	0,037326637	4,520274896	0,000480203
	RDA	0,000403968	3,29921E-05	0,140400586	12,24438172	7,21758E-09
	MMS	0,000228969	5,85942E-05	0,069585943	3,907714266	0,001577622
	SER	3,12615E-05	5,85872E-05	0,01413011	0,533589236	0,601995155
	WND	0,000248075	9,15249E-06	0,163739891	27,10462686	1,68893E-13

 

[Gingko]

Just unlocked RDA - predicted perfectly by this formula...

Just unlocked RDA last night (https://www.planetcalypsoforum.com/forums/showthread.php?p=554739#post554739) and I was tracking progress with the formula mentioned above. It unlocked exactly at 29.50% so this was a perfect prediction!

My skills at that time:

LWT: 3572.1
ANA: 3046.5
IRD: 1349.9
MMS: 2342.8

 

[Jimmy B]

How sure are we of the results in this thread? Some chips from Trabin and Dale in the chipping thread suggest Serendipity 4% (bit of uncertainty there but I think so) and RDA 11% (no uncertainty...).

 

[Witte]

How sure are we of the results in this thread? Some chips from Trabin and Dale in the chipping thread suggest Serendipity 4% (bit of uncertainty there but I think so) and RDA 11% (no uncertainty...).

I updated the opening post with link to your thread. I think it is more relyable then these calculations.