PE|EU WikiTools

[Immortal]

<I really really don't know where to post this now I r conufzzd, and too {lazy;tired;drunk} t ocare... but it's cough topical and I posted it in the wrong thread once already >_> >

In another thread where I was answering how to predict agility gains over time... I used 6100-6000 (i.e. 100*agi) to give approximations as to time required hunting... turned out to be a bad example due to continuing fluctuation in the algo at that point...

d(C(x))/dx=sin(g(x))*exp(f(x))

Skill Value	SKill Volume	d(C(x))/dx
5600		214.84		18.89
5700		229.51		14.67
5800		241.2		11.69
5900		257.48		16.28
6000		279.33		21.85

(from here onwards normal exponent d(C(x))/dx=exp(f(x)))

Skill Value	SKill Volume	d(C(x))/dx
6100		298.19		18.86
6200		318.89		20.7
6300		341.39		22.5
6400		365.68		24.29
6500		391.74		26.06
6600		419.57		27.83
6700		449.13		29.56

Just needed to confirm it was (intended to be) piecewise like that...I wondered also if it was consistent across attributes

 

[Doer]

Sorry, i have no idea what you're saying. Could you explain using more words, please.

 

[Immortal]

Okay using chipping optimiser (which I assume is based on master-curve (which I can't find any other appropriate thread to post in)) I took a sample skill volume from 6100 to 6000, which ended up giving me an odd result... So I investigated further

Standard master-curve (or whatever it's referred to nowadays) V(x)

dV/dx

Definable dV/dx regions

Discontinuities

So my questions with relation to dV/dx
1) Are those discontinuities intentional?
2) Are the discontinuities consistent with data gathered
3) Is there any reasoning why the function dV/dx changes at those discontinuities
4) Is attribute volume and dV/dx known to be similarly affected? If not... can we assume the function for attributes can be arrived at by deconvoluting the function from dV/dx and removing the sin component?

 

[Doer]

Okay using chipping optimiser (which I assume is based on master-curve (which I can't find any other appropriate thread to post in)) I took a sample skill volume from 6100 to 6000, which ended up giving me an odd result... So I investigated further

So my questions with relation to dV/dx
1) Are those discontinuities intentional?
2) Are the discontinuities consistent with data gathered
3) Is there any reasoning why the function dV/dx changes at those discontinuities
4) Is attribute volume and dV/dx known to be similarly affected? If not... can we assume the function for attributes can be arrived at by deconvoluting the function from dV/dx and removing the sin component?

The chipping optimizer uses jdegre's fit to the skill volume curve. I imagine he made it piecewise, as did i. You should inquire about that in the chipping optimizer thread.

You will find some transitions if you use the wikitools calculator (my fit), although i did smooth between functions by moving through a weighted average of the two at the transition. It's one of the reasons the tt->level inverse is a bit "jumpy".

These come about because there was no one function that perfectly fit all the data. I used exponentials and sines and i think polynomials for some segments. You can read through the last bit of the thread where we worked through it to see the impressions and so forth: https://www.planetcalypsoforum.com/forums/skills/58525-fill-up-esi-data.html

 

[Immortal]

Had a problem with getting correct values on progress calc for health. I double checked with Chipping Optimiser which gives correct values.

Pretty please can you add some sword dmg/hit

 

[Doer]

Had a problem with getting correct values on progress calc for health. I double checked with Chipping Optimiser which gives correct values.

Pretty please can you add some sword dmg/hit

What specifically was the discrepancy for the health? Was it over or under, and by how far?

I haven't had more than a few minutes at a time to think EU for the last few weeks and it's all spent on EF and convincing myself that the multitarget chipping calculator algorithm is correct. Knowing that there's interest for the progress tool will be enough to shift my attention that way next time i can code.

Thanks for the feedback.

 

[Immortal]

What specifically was the discrepancy for the health? Was it over or under, and by how far?

I haven't had more than a few minutes at a time to think EU for the last few weeks and it's all spent on EF and convincing myself that the multitarget chipping calculator algorithm is correct. Knowing that there's interest for the progress tool will be enough to shift my attention that way next time i can code.

Thanks for the feedback.

Yeah I'm intending to start using it to compare predictions short term vs longterm, and see how those work when it comes to the actual amount of work required. I'm regularly logging intervals of 3k weapon expense so the data sets should be coherent.

I did a check on all individual skills to see if there were any unexpected skill health contributors affecting it: there aren’t. Starting health with all values=1 is 89.2

Health consistently over by 0.9-1.0

A couple of future features that would be nice:

• Skill level prediction
  How long to get individual skills up to a certain level (say i set a goal of shortblade skill to 4000, i can then compare it to data gathered via the chipping optimiser and my own expenses datas to how much cheaper it is to naturally skill)
• Extraneous professions prediction
  What the Profession levels might look like will be when reaching said goal (similar to chipping optimiser output where affected professions are listed)
  e.g. while I might like to skill health, some of those gains will contribute to evader as well
  
• automatic "% contribution TT calculation"
  I could use this to quickly see how those skills that contributed to the profession gain were distributed (by TT value)

 

[Doer]

Yeah I'm intending to start using it to compare predictions short term vs longterm, and see how those work when it comes to the actual amount of work required. I'm regularly logging intervals of 3k weapon expense so the data sets should be coherent.

I did a check on all individual skills to see if there were any unexpected skill health contributors affecting it: there aren’t. Starting health with all values=1 is 89.2

Health consistently over by 0.9-1.0

A couple of future features that would be nice:

• Skill level prediction
  How long to get individual skills up to a certain level (say i set a goal of shortblade skill to 4000, i can then compare it to data gathered via the chipping optimiser and my own expenses datas to how much cheaper it is to naturally skill)
• Extraneous professions prediction
  What the Profession levels might look like will be when reaching said goal (similar to chipping optimiser output where affected professions are listed)
  e.g. while I might like to skill health, some of those gains will contribute to evader as well
• automatic "% contribution TT calculation"
  I could use this to quickly see how those skills that contributed to the profession gain were distributed (by TT value)

Ok, i added brawler, whipper and swordsman. It should be possible to print out a list of all the skills along with the volume gain and i will try to add it later tonight. The other things will be more involved.

 

[wizzszz]

I did a check on all individual skills to see if there were any unexpected skill health contributors affecting it: there aren’t. Starting health with all values=1 is 89.2

Health consistently over by 0.9-1.0

Oh noes - don't tell me you have health predicted wrong by Jdegres Chipping Optimizer...

Base health is 88.00, influence of the attributes/skills is listed here
(should be the same values the chipping optimizer uses, not sure if wikitools were updated with the new contribution values for attributes)

Do you, by chance, have any of the unlocks listed under "contribution to health still unknown"?

 

[Doer]

Oh noes - don't tell me you have health predicted wrong by Jdegres Chipping Optimizer...

Base health is 88.00, influence of the attributes/skills is listed

FYI i use 88.8 as the base health. Can't remember why except that it fit best last time i compared a few different avatars. That could account for almost one hp overprediction if it's off.

I added the skill volumes to the progress calculator. It also lists % contributions for comparison. If you find any discrepancies, please let me know. Tailoring is probably off...

 

[Immortal]

Oh noes - don't tell me you have health predicted wrong by Jdegres Chipping Optimizer...

Do you, by chance, have any of the unlocks listed under "contribution to health still unknown"?

Jdegre's Chipping Optimiser correctly estimates my health. However if i set all values =1(instead of 0), ofc i get an extra 0.2 there.

And yes I am paying attention to any unlocks i get for their health contribution but don't have any of the missing ones yet

 

[wizzszz]

FYI i use 88.8 as the base health. Can't remember why except that it fit best last time i compared a few different avatars. That could account for almost one hp overprediction if it's off.

You seem to have missed the thread i've linked (and don't read the summary in first post only) - the values there are very likely correct, tested with very different avatars (i.e. Neomaven and a complete newb) and resulted in minimal errors only, so i have little doubt the values there are correct.

Jdegre's Chipping Optimiser correctly estimates my health. However if i set all values =1(instead of 0), ofc i get an extra 0.2 there.

Don't really get where those 0.2 come from - shouldn't you have 9 stamina?
Those 9 points alone are 0.8325 HP - however, for 1 stamina it would fit.

1 psyche and 1 intelligence give 0.0125 HP each, 1 strength is 0.05 and 1 agility is 0.025 - sums up to exactly 0.1 HP.

If you count a stamina of 1 this is another 0.0925 which results in a base health of

88.1925 for stamina=1 avatars
88.9325 for stamina=9 avatars

Doer very likely still uses the old formula which calculates 1 HP for every 40 points in every attribute.

Glad that the chipping optimizer is correct - otherwise it would mean the formula is still off.

 

[Immortal]

Don't really get where those 0.2 come from - shouldn't you have 9 stamina?
Those 9 points alone are 0.8325 HP - however, for 1 stamina it would fit.

I don't have 9 stamina

 

[wizzszz]

I don't have 9 stamina

n00b...

 

[Immortal]

n00b...

I know I only just hit 60 str :O

Meanwhile, doer's been doing a fantastic job with the progrees calculator

Sword Dmg

Skill	Volume	Levels	Fraction	Contribution
Anatomy		10.93	25.43	0.04	20%
Wounding	5.78	61.14	0.04	10%
Melee Combat	4.93	200.26	0.07	5%
MDA		5.44	291.46	0.29	14%
Longblades	24.53	257.91	0.46	25%
Serendipity	3.25	22.71	0.01	4%
Inflict Melee	1.7	223.9	0.08	5%
				
				
	56.56	1082.81		83%

Sword Hit

Skill	Volume	Levels	Fraction	Contribution
Dexterity	1.02	17.22	0	2%
Melee Combat	4.93	200.26	0.11	9%
Perception	1.15	28.87	0	2%
Coolness	2.44	33.6	0.01	5%
Bravado		2.65	232.26	0.1	7%
Longblades	24.53	257.91	0.59	39%
Courage		2.34	22.12	0.01	4%
Serendipity	3.25	22.71	0	1%
Martial Arts	1.92	244.32	0.09	6%
Combat Reflexes	2.78	25.47	0.01	5%
Combat Sense	2.86	99.05	0.02	4%
Heavy Melee	2.58	136	0.06	7%
				
				
	52.44	1319.79		91%

first 9k I got 0.0084% skill/ped on sword (including skill contributors on hit+dmg, but not excluding factors such as evade,CS,CR gained via evade/dodge/other weapons used). Going to split runs (have three atm) and use to predict, then compare to actual gains.

Is the value fraction intended to just raw skill rather than TT? Both values could be interesting side by side

 

[Doer]

You seem to have missed the thread i've linked (and don't read the summary in first post only) - the values there are very likely correct, tested with very different avatars (i.e. Neomaven and a complete newb) and resulted in minimal errors only, so i have little doubt the values there are correct.

You seem to have forgotten that i was an active part of that thread and the final cracking of the attribute contribution.

I hadn't updated the esi.coolinc.info site with the new contributions since finding them on my local computer. It should be correct now.

Immortal, the fraction is the contribution of the skill to pro. standing divided by the total increase in pro. standing (so it's level-based).

 

[Immortal]

Immortal, the fraction is the contribution of the skill to pro. standing divided by the total increase in pro. standing (so it's level-based).

Yup understood the raw skill/total skills threw me

I think an interesting development option would be use weapon type
And then calculates gains based on the split (i.e. all sword contributions – we assume dmg+hit contribute equally?)
Something like

		run 9k (1,2,3)	actual	predicted
Dexterity	1.02	2%	0.01	0.01
Melee Combat	4.93	14%	0.06	0.08
Perception	1.15	2%	0.02	0.01
Coolness	2.44	5%	0.03	0.03
Bravado		2.65	7%	0.03	0.04
Longblades	24.53	64%	0.32	0.37
Courage		2.34	4%	0.03	0.02
Serendipity	3.25	5%	0.04	0.03
Martial Arts	1.92	6%	0.03	0.03
Combat Reflexes	2.78	5%	0.04	0.03
Combat Sense	2.86	4%	0.04	0.02
Heavy Melee 	2.58	7%	0.03	0.04
Anatomy		10.93	20%	0.14	0.11
Wounding	5.78	10%	0.08	0.06
Melee Damage 	5.44	14%	0.07	0.08
Inflict Melee 	1.7	5%	0.02	0.03
		76.3	174%

Meanwhile, here's a pretty graph using cross-correlated data of both the chipping optimiser, progress calculator and a standard loot return assumption

 

[wizzszz]

You seem to have forgotten that i was an active part of that thread and the final cracking of the attribute contribution.

Umm, yes... sry - it's been a while... just remember Jimmy B. being faster...

I hadn't updated the esi.coolinc.info site with the new contributions since finding them on my local computer. It should be correct now.

At least we've tracked down the issue - already thought we need to review the formula!

DefCon set back to 5 - situation normal - pheew!

 

[Immortal]

Very interesting results when I looked at the smaller intervals

I graphed all sword-related skills and their expected gains to the actual TT gains recorded for the 3k intervals
The later runs correlate far more with expected skill gains.

No solid explanation for this yet. the TT value of my total skills between for run1,run2,run3 is 40,30,30, yet all spent 3k, and sword related skills volume accounts for 50%, 95% and 95% of skills.

odd...

 

[Immortal]

Why am I posting this in her? don't know I just found my result interesting and this program makes it very easy to analyse
Both runs are 9k.

		Run1	Run2
Dexterity	1.02	1.74
Melee Combat	4.93	7.74
Perception	1.15	1.32
Coolness	2.44	2.74
Bravado		2.65	3.76
Longblades	24.53	35.84
Courage		2.34	2.68
Serendipity	3.25	3.89
Martial Arts	1.92	3.05
Combat Reflexes	2.78	3.51
Combat Sense	2.86	2.63
Heavy Melee 	2.58	4.75
Anatomy		10.93	12.14
Wounding	5.78	6.49
Melee Damage 	5.44	8.34
Infl. Melee Dmg	1.7	2.58
	Total	76.3	103.2
	STR	1.95	2.12

each 3kped run yielded 30, 30, 30 ped TT, until yesterdays run Which garnered 50ped TT (and 1kskill instead of 6-700) in a standard 3.1k sword run. I found this interesting. If it continues it drastically reduces the amount of time required to be spent. A point of note was that in Run2, 94% of TT gain is in sword related skills, compared to 85% in the previous 9k.

 

[Doer]

Why am I posting this in her? don't know I just found my result interesting and this program makes it very easy to analyse
Both runs are 9k.

		Run1	Run2
Dexterity	1.02	1.74
Melee Combat	4.93	7.74
Perception	1.15	1.32
Coolness	2.44	2.74
Bravado		2.65	3.76
Longblades	24.53	35.84
Courage		2.34	2.68
Serendipity	3.25	3.89
Martial Arts	1.92	3.05
Combat Reflexes	2.78	3.51
Combat Sense	2.86	2.63
Heavy Melee 	2.58	4.75
Anatomy		10.93	12.14
Wounding	5.78	6.49
Melee Damage 	5.44	8.34
Infl. Melee Dmg	1.7	2.58
	Total	76.3	103.2
	STR	1.95	2.12

each 3kped run yielded 30, 30, 30 ped TT, until yesterdays run Which garnered 50ped TT (and 1kskill instead of 6-700) in a standard 3.1k sword run. I found this interesting. If it continues it drastically reduces the amount of time required to be spent. A point of note was that in Run2, 94% of TT gain is in sword related skills, compared to 85% in the previous 9k.

I've also seen that gain by profession is more uniform than individual professions (reflecting phases of "dynamics", perhaps, or just chance), but i've not seen such a large departure like that before.

 

[Immortal]

I am writing this tutorial here and linking to it to give this thread more exposure.

How to analyse skillgains related to Health​

Here is my suggestion on how to work out the quickest/cheapest/easiest way to skill health using tools available via this forum.

Take a skill scan using jdegre's software. http://jdegre.net/pe/scanner/
If you have problems using it or questions, write them in jdegre's thread:https://www.planetcalypsoforum.com/forums/skills/95277-skill-scanner.html?highlight
Save the file (.csv)
Go do one activity for the day, logging all expenses.
Take another skill scan.
Save the file (.csv)

Compare both files using the progress calculator
http://esi.coolinc.info/progress.cgi
You will need to cut and paste the data from the files. I find Notepad easiest to use.

Enter a goal level (realistic)

The most meaningful figures are underlined. The last figure is the amount of "real" skill you gained.

Now repeat the process for different activities (e.g. mining, hunting with pistol, hunting with melee) to determine what would be the best way for you to achieve your goals.

 

[Immortal]

unable to access the site

 

[Doer]

unable to access the site

Yeah i noticed that. I thought it was just a temporary outage but now i'm not so sure. I'll see about putting it on the entropedia server when i get back from freezing my ears off bike touring this weekend.

Edit: Survived; sorta. I see the site is back up so have fun.

 

[Immortal]

<can only express self with smiley>
<tested some other profs>
sword dmg


evader+ health

values returned (hit)

46	6.1
46.1	9.1
46.2	9.7
46.3	10.2
46.4	10.8
46.5	11.3
46.6	11.2
46.7	11.7
46.8	12.2
46.9	13.5
47	14.1
47.1	14.6
47.2	15.2
47.3	15.7
47.4	15.3
47.5	16.8
47.6	17.4
47.7	19
47.8	19.6
47.9	20.2
48	20.8
48.1	21.3
48.2	21.9
48.3	23.8
48.4	24.4
48.5	25
48.6	25.6
48.7	26.2
48.8	26.8
48.9	27.5
49	28.1
49.1	28.7
49.2	29.3
49.3	29.9
49.4	32.2

values returned (dmg)

59	141.6
59.1	142.7
59.2	143.7
59.3	144.8
59.4	145.9
59.5	147
59.6	148.1
59.7	149.2
59.8	155.8
59.9	157
60	158.1
60.1	159.2
60.2	160.3
60.3	161.5
60.4	162.6
60.5	163.7
60.6	170.7
60.7	171.9
60.8	173.1
60.9	174.3
61	175.4

59.7	149.2
59.71	149.3
59.72	149.4
59.73	149.5
59.74	149.6
59.75	149.7
59.76	149.8
59.77	149.9
59.78	150.1
59.79	150.2
59.8	155.8

 

[Doer]

My Immortalese is still pretty rudimentary. Are you "saying" that sometimes the predicted repetitions changes in an unexpected fashion? Are your skill gains each interval the same?

I don't remember all the function calls the little predictor routine has to make but i think it uses the inverse(level->volume) function which is not smooth and not precise (ie it's a hack approximation). If i ever implement the multi-target chipping calculator i have outlined somewhere i will first need to make it behave like a true inverse function so that will hopefully resolve the apparent incongruity here that is causing you so much consternation.

In the meantime i can add this. Disclaimer: The progress calculator was meant more as an approximate guide than an engineering tool.

 

[Immortal]

Yes, and in some cases the approximation seems to be cumulative (first graph) where it jumps suddenly...and continues it's regular gradient

It seems to have problems with some numbers in particular

The error curve though seems to be consistent - and it's having trouble with some numbers in particular.

Specifically on larger N, it seems to be building on a 'bad thing' note how the gradient is consistent, and the error is periodic.

note also how the evader sub-peak @ 37.3 can be correlated to the health sub-peak at 173

I thought you might have used an iterative function, since you are calling for threshold?

like in following example, where precision is 1.0, but the number of reps doesn't go backwards

//i=skill nominal figure, eg. agi=1, ...., surface comp=117
//gain is difference in initial rep
Find N_reps()
{
while prof<level_asked
for i<N_skills
skill=skill+gain
i++
ENDFOR
j++
ENDWHILE
return j
}

To gain precision, can use a follow up function

Find N_reps_high_precision()
{
prof=j-1
for prof<level_asked
for i<N_skills
skill=skill+gain*0.1
i++
ENDFOR
j=j+0.1
ENDWHILE
return j
}

So instead of relying on the hack inverse, actually do a vol_convert then interpret using real curve?

Maybe too much calc req'd for iterations though? Don't know :/

As you say it's an approximation tool, and it's good for that, however in the DMG approximation graph, the error appears to be cumulative - so I think this may result in bad approximations. (and i certainly hope it doesn't take me 14kped of sword to get to level 45 hit, which happens to be half a level away - 44->44.5 took ~ 4kped i can't imagine such a jump is required to 45)

 

[Immortal]

I did a mock test on the first target of 45 sword hit.
I asked the function to tell me how many iterations: it returns 3.5
It also returned the volum change. (volc)
i used chip-out feature on jdegre's to gather the volume data.

I then added 3.5*10/9*volc to each skill
using this data i subbed the values into excel, to get the expected volumes after 3.5 iterations.

i then used jdegre's to estimate what skill level (yes, tedious) corresponded to which volume. After these were determined, the Skillchip optimiser tells me I should be at 45.32 (where as this one tells me i should be at 45). This would correspond to 50/82*3.5 iterations, or between 4-5kped like i would expect.

I would say a difference of 60% is significant enough to review the approximation method.

 

[Doer]

I can't think of any reason why it would be cumulative in one profession and not the others. They are all treated the same...

Except--remember that it can't take into account attribute changes because there is not a corresponding level->volume mapping available for those yet. They play such a small role in progress after they hit level 50 or 60 that it shouldn't matter much, but i guess that could account for a cumulative error. It may also just be a case of cumulative rounding errors in the skill calc.

In trying to find an easy way to handle large repetition situations i think i compromised by choosing a trick that has problems with very nearby targets. Perhaps that's the origin of the huge difference between expected and predicted value.

If you keep this up i'm going to have to go back and open that rusty bucket of bolts and poke around. Damn you.

 

[Immortal]

I did a manual 100-iterations and compared wikitool output to assess it's accuracy. I got the profession levels from those 100 iterations and then asked wikitools to give it's approximation
Profession (level) - wikitool output
Sword hit (58.64) - 124.4
Sword dmg (58.00) - 125.9
Evader (42.33) - 129.4
Paramedic (39.16) - 124.9
Health - (182.8) - 125.8

So It consistently overpredicts the amount of time required by 25% - a significant factor if assessing whether to chipin or skill manually.