Fill up ESI data...

[Doer]

Let's tackle this already

From Kyl's and my data​

Please: I need slope information from people with 6.5k+ skills. All you have to do is look in auction at chips for that skill and tell me 1) the tt of the chip and 2) your start and end level. If the chip is going to give you less than ~15 levels (~10 levels over 10k skills) or over ~30 levels it's probably not useful information. Repeat this three times with each skill over 6.5k for slightly different size chips (the range of a few levels gain) and it will be more accurate.

If we can get some confidence about the slope at intervals up to or above 10k, we can chose a suitable fitting equation and revive the utilities that predict tt for chipping above 5k skills. Then i can finish my program module to calculate the cost/value and optimal way to chip up or down...


That's a really slippery progression slope.

 

[Doer]

Good idea, Doer, let's finish this.

Yeah, i'm talking to myself. It's late.

I spent some time at the auction of CP, soliciting help from the various ubers and noobers that hang out there, and thanks to the help of General Turdgidson, Amber, Skippie, Stryker, Ravaj, and sacklitch, i have this to show you:

So, some comments. First, there is clearly a divergence from the nicely fitting polynomial that worked very well for the first 6k. It is hard to tell where it happens due to the oscillation but i'm guessing around 7k. Coincidentally (or not), that equation predicted a slope of exactly 1 PED/level at 8k skills (which is the new "10k"), but if it's not longer in effect at that level i guess that's just a cute coincidence after all. Anyway, after the initial polynomial/pseudo-exponential portion, there is actually a bit of downward curvature to the increase in slope. The magnitude of the oscillation just keeps increasing, making it hard to be sure where the mean value lies from the few data points.

Gah.

So there is a whole lot of oscillation going on here, and it appears to be increasing in period (although to verify that i'd need more data points at the high levels). I tried subtracting a quadratic from the points above 7k and then fitting some kind of function to the flattened result, but it's a mess, obviously.

Anyway, it seems that the best approach is what Carebear did for the Entropia Tools calculator: split it into three segments. I can sorta eyeball one ~straight line from 7k to 10k (with a sinusoidal function thrown in) and another from 10k up.

I'll mess around with it some more when i am less tired -- it's starting to look like a sinusoidal hysteresis diagram.

Morning now, and i can finally remember what this looks like to me: strange attractors. I realized that when Jimmy asked if skills all have the same "weight" on chip. I have assumed they did and i believe that all the calculations people have done before have assumed the same, but those high level points there would certainly make more sense if there were two (three) distinct weightings pulling the result curve back and forth... time will tell!

Here's what i need from you

  Skill start 	TT      Skill End      
Stryker	12120	11.46	12131
MMS     12120	14.33	12134
	12120	18.76	12138

If you have a skill above ~6.5k, your participation would be much appreciated. Just find three similar tt chips on auction that give you from 15-20 (ideally) levels of skill and report like that one above. Over 10k skills a chip that only gives you 10 or so would probably be ok, too.

Thanks

 

[Jimmy B]

lol@Doer talking to himself

Good luck with it anyway, it would be useful to know. I'd help if I could but my highest skill just over 6300 so a little below what you need...

 

[Bogger]

       Skillstart     TT  Skill End      
Bogger     7109     9.95   7134
Handgun    7109    11.71   7139
           7109    14.68   7147

-
Dont understand your math but let me know if you need bigger gaps between tt of ESI. (9,95 was smallest i could find atm)

 

[Trabin]

  Skill start 	TT      Skill End      
Trabin	7361	12.24	7390
Rifle	7361	10.31	7385
	7361	17.24	7400

  Skill start 	TT      Skill End      
Trabin	7387	1.95	7391
BLPWT	7387	8.04	7404
	7387	10.94	7410

  Skill start 	TT      Skill End      
Trabin	7026	15.38	7057
Anatomy	7026	10.41	7046
	7026	12.55	7051

 

[Doer]

-------Skill start TT Skill End
Bogger---7109 --9.95 7134
----------7109 11.71 7139
----------7109 14.68 7147

Dont understand your math but let me know if you need bigger gaps between tt of ESI. (9,95 was smallest i could find atm)

Thanks, Bogger! Basically i want multiple points just for comparison to make sure the first one isn't an extreme case or something. It also helps me see the slope of the slope at that point in case i have to go to the second derivative to get anything useful. So: the closer together the better, and the lower the better (as long as it's above ~10 levels to minimize rounding errors).

The overall idea is that, since it's very hard to get data for the full tt value of a skill above 6k (it only comes when people chip out and share their results, or even more rarely, when someone buys a huge chip), we'll be able to get more data for the tt vs. level above 6k by looking at the slope at various points. That will show the trend of the curve already calculated for up to 6k, justifying either extrapolation or allowing a new fit to be made for that range by integration of a fitting equation.

Better yet would be data on the actual tt in skills that one gets from a huge chip (1-6k+) or in extracting a huge amount of skill from 6k+-1, but in the absence of data like that we'll have to go with the slopes.

If anyone sees a huge chip on auction that would take a noob in that skill from 1 to over 6k in a skill, please mention it here so i can check it out!

Edit: ty Trabin, too.

 

[Trabin]

Ooops - just seen you wanted close together and over 6 skills - let me re-do the middle one (was trying for furthest apart.....

  Skill start 	TT      Skill End      
Trabin	7387	10.10	7408
BLPWT	7387	8.04	7404
	7387	10.85	7410
	7387	10.94	7410

Not a lot of chips to choose from for this skill atm - but this should be a little better data.

 

[Jimmy B]

Apologies if this has been brought up before...

Do we know for a fact that the same amount of skill in different skills fill up the same amount of ESI?

 

[Doer]

Ooops - just seen you wanted close together and over 6 skills - let me re-do the middle one (was trying for furthest apart.....

  Skill start 	TT      Skill End      
Trabin	7387	10.10	7408
	7387	8.04	7404
	7387	10.85	7410
	7387	10.94	7410

Not a lot of chips to choose from for this skill atm - but this should be a little better data.

Thanks, Trabin. To be honest as i look at this it might in the end require some larger gaps so i can just get an average slope, but i'm not sure that will work properly and if this fails i'll move on to that.

The problem with small chips, or rather chips that give less than 10 levels of gain, is that the error can be quite large. In the case of that chip that gave you 4 levels, with a +/- 1 level of uncertainty, the slope ranges from 0.39-0.65, which is clearly not a good thing. The one that gives you 17 is +/- ~5% error, which is much more acceptable. This changes as the overall slope goes up, so with higher skills a slightly smaller level gain is acceptable, but i'm going to recommend 15 as the ideal minimal gain for the purposes of checking this for below 10k skills. I'll update the above posts to make that clear to future testers.

Areas with large gaps:
6550-6900
7400+

Jimmy: It has been discussed and some say it doesn't. The overall thought was that they do but i certainly haven't tested it and have been going on the assumption that they do. I believe CareBear's tool assumed the same and trust that any large discrepencies would have been reported. If anyone has evidence that all skills aren't the same weight, please let me know. Looking at the plot for high levels, it could certainly be interpreted as two different weighted skills rather than an oscillation. That would be an exciting discovery in terms of ultimately predicting tt of skill levels, but poor in terms of me repeating my data collection process.

So, please let me know what the skill is when you report, if that's ok. I'd rather have the numbers and not a skill to go with it if for some reason you'd rather not say.

*sigh*

I'm going to have to go back and further refine the original data from Kyl now that i've got a better idea what's important. It might change the lower curve a bit, but not likely visually -- just better for fitting. Kyl, if you recall which skills were which on that last data you posted, please let me know.

 

[Doer]

Here's an interesting teaser based on the data so far. I fit a logarithmic curve to the slope data from 6k up and then integrated it and used it to plot tt vs. skill level. It predicts almost two full ESIs (2500 PED) for 10k skills, something i think to be the case (it was one before the VU tyvm MA ) and almost one full ESI for 8k skills, which i think is the new 10k as i mentioned before. I expect that the resulting curve will move toward those values as we add more data points and it becomes a better approximation.

Note that i'm ignoring the oscillating (slowdown/speedup) part of the advancement due to the fact that there's no good way to approximate that with the data so far.

Neomaven said he'll help out with more data so i hope to put to rest or promote the different weighted skills hypothesis, as well as refine the fit of high skills.

In any case we now have what i think is a passably good approximation for skills vs. tt up to and beyond 16k skills.

 

[jdegre]

I've made available on-line the data points I've collected so far:

http://spreadsheets.google.com/pub?key=pqOMP11FL84qZIKbo4Iasbg&gid=0

It's a Google Spreadsheet so it has its limitations compared with Excel, but I really don't like to post Excel files online.

The good thing is that I can made it collaborative, if someone wants to contribute to it.

- Sheet1 is the collected data points, most of them are temptative (in red) , and are refined iteratively from the chips I find in auction.
- Sheet2 is the graphic made from these data. It must be manually rebuilt everytime new data is added in Sheet1

That's it. Just for whatever it's worth...
/jdegre.

PS.: Contributions much appreciated would be if someone with rifle or lwt skills around 7k-8k can browse some chips in auction (there are currently some of them around 200-300 ped) and share their skill gain. There is no point in getting data around 10k yet, because the graphic must be built progressively from bottom to top with no gaps in between.

 

[Doer]

Ok let's summarize what's going on then, for my sake as well as the others' who want to help out.

• Goal: find an equation that will accurately predict the tt of a skill at a certain level
• Challenges:
  • it's a complex relationship, with one sinusoidal (wave-like) component
  • The only way to get data is by observing how many skill levels implants of a certain tt value will give
  • There is rounding uncertainty when looking at that information
  • Maximum implant size is 1250 PEDs, and it's hard to find any filled implants of larger sizes to test
• Approaches
  • jdegre's approach is to build up, from 0.01 tt, direct tt-to-level connection by "stacking" chips. For example once you know that 1-4,000 skills is exactly 50 PEDs (it's not really), you can then find a skill at 4k and find what another 50 PED chip will add to it.
  • my approach is to find the slope of the curve by seeing how many levels per PED chip can be gained for small chips at that level. It doesn't require a foundation below it (assuming good continuity in the original function) to be a useful data point but is more subject to uncertainty.

So in conclusion, jdegre's approach is going to give better results in the long run but may be difficult to apply at very high levels due to a shortage of data, my approach will allow a "reasonable" extrapolation to be made from wherever his data can reach, unless a perfect fitting equation can be found to describe the data, in which case i will be able to check the result.

What is needed: data from auction for 7k+ skill levels that shows the beginning skill, the end skill with the chip, and the tt of the chip. For my purposes those chips that give 10-20 levels should be reported, and for jdegre's the biggest chips you can find should be reported.

Does that sound right, jdegre?

 

[jdegre]

sorry doer, it seems that my post created some confusion, because I didn't realize that my and your approach were different. I thought that you were following the same principle of building the curve "incrementally" (sp?).

both approaches are in fact complementary, so definitely your system is more than welcome too.
let's see if we can get nice useful data for an interval of skills as big as possible (although the whole curve most probably will be close to impossible to determine).

cheers,
/jdegre.

 

[Doer]

sorry doer, it seems that my post created some confusion, because I didn't realize that my and your approach were different. I thought that you were following the same principle of building the curve "incrementally" (sp?).

both approaches are in fact complementary, so definitely your system is more than welcome too.
let's see if we can get nice useful data for an interval of skills as big as possible (although the whole curve most probably will be close to impossible to determine).

cheers,
/jdegre.

No problem, i just wanted to clarify that was what you were doing for my sake. I think when you hit about 10k, and the going gets tough, we can use my data above 10k to carry on the curve and have a satisfactory solution. In the meantime i'll continue with it just in case it proves useful below 10k or i can get a good fit that applies over the whole range from it.

I assume you used all the data provided by Kyl and Actam in your google spreadsheet? If so i will adopt that refined set to use as the basis for my tests.

If you could add me for access i will add any new data to it. It would be helpful if you could explain your exact method, though.

 

[jdegre]

I assume you used all the data provided by Kyl and Actam in your google spreadsheet? If so i will adopt that refined set to use as the basis for my tests.

If you could add me for access i will add any new data to it. It would be helpful if you could explain your exact method, though. I have the login Davidfalkayn there.

Yes I used all the data points from actam and kyl, and a bunch of skill chip that I browsed on auction.

I have added you as colaborator for the doc, so you should be able to edit it as well.

The method I've used to fill in the gaps is based on aproximations based on the slope form the closest points in the curve. It is not exact, of course, and it must be refined as long as new data is entered.

/jdegre.

 

[Doer]

The maximum error is about +-1.5 PED except for one small peak just before 6k that reaches about 3 PED difference -- but i think that data point is bad.

This fitting makes the bad data points pretty glaring when subtracting. I think with better data the deviation can be brought down to +-1 PED. The equation is segmented into regions from 1-550, 551-4000, and 4000-6000. Like so:

And then i add a sine function. I think the range 550 to at least 5k should be all the same function if the data is perfected, but perhaps not.

You can see that including a sinusoidal component can reduce the overall error if the exponential portion is sufficiently good.

 

[Rony]

Someone must inform Connor that his programmers waste servers CPU time by adding useless sin calls.
In fact I dont know if you found what is the source of this variations but I remember that during my studies I learned that everything in EU gets a runtime adjustement. Also I remember that some time ago you would get different skill levels on different servers from the same chip.

 

[Doer]

Someone must inform Connor that his programmers waste servers CPU time by adding useless sin calls.

lol

I have no idea what would cause the variation per server. Is that still the case?

Incidentally, the sine period and its multiplication factor that i arrived at by trial and error are both such clearly human-chosen values that i have no doubt they are correct.

There is not enough data above 6k to get a good fit, but i think the results are still sufficiently good for almost anyone's purposes in determining ESI size up to ~800 PEDs within a reasonable margin of error.

I tried using the integrated fit to the slope data from 6k-16k as the fit to the few points of data available above 6k from jdegre and the fit is too poor to be useful yet.

Conclusion: more data is needed both at the low end (6.5kish) to extend jdegre's range, and throughout the 6k-20k region for slope information.

Feel free to test out the function and report your findings. It has rudimentary ability to calculate final skill level if you give it a chip size, but due to the free server's limitations on recursion sometimes it errors out.

 

[EyeContact]

I found this function in this thread:
ESI cost = 0.00197044 * skill * exp(0.00054208 * skill)

I know it's not perfect, but I wanna use it anyway.

Problem is, I'm having a hard time to change the formula to get the skill on the other side of the equasion. Formula I came up with (which doesn't work) is SKILL = SQRT(LN(ESI/0.00197044)/0.00054208)

Help

 

[Doer]

I found this function in this thread:
ESI cost = 0.00197044 * skill * exp(0.00054208 * skill)

I know it's not perfect, but I wanna use it anyway.

Problem is, I'm having a hard time to change the formula to get the skill on the other side of the equasion. Formula I came up with (which doesn't work) is SKILL = SQRT(LN(ESI/0.00197044)/0.00054208)

Help

Unfortunately math isn't very intuitive for me. I fiddled with that equation you picked and don't think an easy solution is possible for skill. However, i never was good at that kind of thing.

What is it you are trying to do? I will add a skill+tt feature to the php script if you are trying to figure out what level a certain chip will get you to. It will be more useful that way, and get you much better results than taking a single equation and solving for skill.

I was hoping to get a full fit up to 10k done before making a chipping calculator utility, so we need some more data in the 6.5k+ range. There are some big chips on auction right now, so if people could look at them and report their begin and end skill level and the tt of the chip it would expedite this.

Right now the data from jdegre's direct approach working upward and my slope approach from high levels is not complete enough to match up:

 

[jdegre]

I found this function in this thread:
ESI cost = 0.00197044 * skill * exp(0.00054208 * skill)

I know it's not perfect, but I wanna use it anyway.

Problem is, I'm having a hard time to change the formula to get the skill on the other side of the equasion. Formula I came up with (which doesn't work) is SKILL = SQRT(LN(ESI/0.00197044)/0.00054208)

Help

you cannot solve "y = x*exp(x)" in an algebraic way. in mathematica and other math software, there is a function called ProductLog(x) which represents this "inverse" function.

http://documents.wolfram.com/mathematica/functions/ProductLog

/jdegre.

PS: In case you're interested, the function you're after is:

skill = ProductLog(0.00054208 * ESIcost / 0.00197044) / 0.00054208

 

[Doer]

I'm glad to know i wasn't losing my mind thinking there wasn't an algebraic form of that.

I added a function to the PHP program to find a final skill level given an initial level and a chip size in PEDs.

You simply enter a level and (optionally) a chip tt in the URL:

http://esi.coolinc.info/?level=1000&chip=12

It can also calculate number of levels from 1 if given a tt value (the inverse function) instead of a level:

http://esi.coolinc.info/?tt=10

The maximum skill level it will allow at the moment is 7.5k to avoid misuse at ranges it is not yet designed for.

edit: Alright that server didn't like my tail recursion so i have it using a dumber but less server-inflammatory iterative function. If anyone encounters a problem let me know. Please test this out and see if it gives you good results for chipping.

This calculator is only as good as the data i have, so if you are getting odd values then it means we need more data. *hint hint*

 

[wizzszz]

I'm glad to know i wasn't losing my mind thinking there wasn't an algebraic form of that.

I added a function to the PHP program to find a final skill level given an initial level and a chip size in PEDs.

You simply enter a level and (optionally) a chip tt in the URL:

http://esi.coolinc.info/?level=1000&chip=12

It can also calculate number of levels from 1 if given a tt value (the inverse function) instead of a level:

http://esi.coolinc.info/?tt=10

The maximum skill level it will allow at the moment is 7.5k to avoid misuse at ranges it is not yet designed for.

edit: Alright that server didn't like my tail recursion so i have it using a dumber but less server-inflammatory iterative function. If anyone encounters a problem let me know. Please test this out and see if it gives you good results for chipping.

Try this : http://esi.coolinc.info/?level=229&chip=0.37

Predicts to end up at level 342.6, but should be 381 according to MA

Pretty large deviation at such low levels...

Have you taken into account that different skills might have a different formula? And that merging those data might have caused the sine in the graph, especially at higher levels?

/Edit:

Checked another: http://esi.coolinc.info/?level=140&chip=0.5

Predicted: 370.0, MA: 376

 

[MG Mighty]

Try this : http://esi.coolinc.info/?level=229&chip=0.37

Predicts to end up at level 342.6, but should be 381 according to MA

Pretty large deviation at such low levels...

Have you taken into account that different skills might have a different formula? And that merging those data might have caused the sine in the graph, especially at higher levels?

/Edit:

Checked another: http://esi.coolinc.info/?level=140&chip=0.5

Predicted: 370.0, MA: 376

Thanks I didn't get how to use that until you've demonstrate:

So here is an example of data not fitting

The tt value of your skill (at level 1) is 0.00 PEDs, and a chip of 0.37 PEDs tt will raise you to level 235.3.

that is false
A 0,37 chip will take a level 1 avatar to level 255

000.27 => 00199
000.28 => 00204
000.29 => 00210
000.30 => 00216
000.31 => 00221
000.32 => 00227
000.33 => 00233
000.34 => 00238
000.35 => 00244
000.36 => 00249
000.37 => 00255
000.38 => 00259
000.39 => 00264
000.40 => 00268
000.41 => 00273
000.42 => 00277
000.43 => 00282
000.44 => 00286
000.45 => 00291
000.46 => 00295
000.47 => 00299
000.48 => 00303
000.49 => 00308
000.50 => 00312
000.51 => 00316
000.52 => 00320
000.53 => 00324
000.54 => 00328
000.55 => 00332
000.56 => 00335
000.57 => 00339
000.58 => 00343
000.59 => 00347
000.60 => 00350
000.61 => 00354
000.62 => 00357
000.63 => 00361
000.64 => 00364
000.65 => 00367
000.66 => 00371
000.67 => 00374
000.68 => 00377
000.69 => 00380
000.70 => 00383
000.71 => 00387
000.72 => 00390
000.73 => 00393
000.74 => 00396
000.75 => 00398
000.76 => 00401
000.77 => 00404

About your 229 starting skill adding a 0,37 will give accorddling to my data:
000.32 => 00227 << 229 << 000.33 => 00233
=>
000.69 => 00380 << 229+0,37ESI << 000.70 => 00383

And 381 is bettween 380 and 383

 

[Doer]

Try this : http://esi.coolinc.info/?level=229&chip=0.37

Predicts to end up at level 342.6, but should be 381 according to MA

Pretty large deviation at such low levels...

Have you taken into account that different skills might have a different formula? And that merging those data might have caused the sine in the graph, especially at higher levels?

/Edit:

Checked another: http://esi.coolinc.info/?level=140&chip=0.5

Predicted: 370.0, MA: 376

Thanks for testing it out. IMO that's not too bad, really. I didn't focus too much on the fit from 1-550 because it comprises just a few PECs worth of tt value, which will be in the noise for chipping anything significant--i don't expect anyone to need it for anything useful. Deviation noted, though, thanks.

It might be a better idea to just do a poly fit to the first segment. I'll try that next iteration and repeat your test. If you find any other particularly large deviations please let me know for future test cases.

Regarding different skills having different weights: Jimmy brought that up a few days ago and it had me concerned, but every skill chip i look at to add data to jdegre's data fits perfectly with the sinusoidally-modified exponential curve. There may be some skills that have a different weight, but neither jdegre or i have found one that doesn't fit the curve.

MG, you come out of the woodwork to flash a bit of your wondrous data, and then go away again until someone else makes something available, whereupon you just come to try to show them up and then disappear again. Why not share what you have so it does someone some good? jdegre had a nice list list built up of tt to level data. If you have some points above 6k we could refine the fitting a lot and have a more useful tool. In any case, it concerns me not at all that my formula deviates by a dozen levels at < 500 skill because that's not going to be particularly relevant for people using a ESI tt tool.

 

[jdegre]

Try this : http://esi.coolinc.info/?level=229&chip=0.37

Predicts to end up at level 342.6, but should be 381 according to MA

Pretty large deviation at such low levels...

I think we have two problems here:

- doer's script does not calculate well the inverse function (for the reasons he posted about server-side script limitations, i guess). in the previous example, 229 levels is calculated as 0.33 tt; adding 0.37 tt makes it to 0.70 tt. the script makes that equal to 342. however, the same script, puts 342, as being 0.60 tt when used like this: http://esi.coolinc.info/?level=342
i'd recommend not to use (yet) the "tt -> skill" feature, to avoid mixing sources of errors. however, please do use the "skill -> tt" feature to check if the formula works ok.

- afaik, there is a single polynomial to interpolate all data from 1 to 7.5k levels, and this creates necessarily errors which could be lowered by splitting the interpolation is smaller segments (0-2k, 2k-4k, 4k-6k, ...)

/jdegre.

ps.: pls ppl, nobody is at 6k-7k skill levels and have a couple of minutes to browse big chips on auct and report here???

 

[Witte]

LWT

6672->7177 185.95 PED
6672->7036 127.71 PED
6672->7034 126.63 PED
6672->6982 100.17 PED
6672->6939 79.62 PED
6672->6858 48.61 PED

Rifle

5751->6819 273.77 PED
5751->6532 191.15 PED
5751->6505 180.40 PED
5751->6415 147.46 PED
5751->6256 110.80 PED

Anatomy

5911->6224 77.44 PED
5911->6115 55.34 PED

Hope that helps

 

[jdegre]

LWT

6672->7177 185.95 PED
6672->7036 127.71 PED
6672->7034 126.63 PED
6672->6982 100.17 PED
6672->6939 79.62 PED
6672->6858 48.61 PED

Rifle

5751->6819 273.77 PED
5751->6532 191.15 PED
5751->6505 180.40 PED
5751->6415 147.46 PED
5751->6256 110.80 PED

Anatomy

5911->6224 77.44 PED
5911->6115 55.34 PED

Hope that helps

woot, thx a lot !... that's exactly the type of data we need, because that's exactly the range where we have bigger gaps.

+rep's (and to everybody that supplies more data, hint, hint.. )
(witte, must spread before blah, blah...)

i've included the new data in the spreadsheet and updated the graphic, which looks quite ok. you can find it here:

http://spreadsheets.google.com/pub?key=pqOMP11FL84qZIKbo4Iasbg&gid=0


/jdegre.

 

[Doer]

I think we have two problems here:

- doer's script does not calculate well the inverse function (for the reasons he posted about server-side script limitations, i guess). in the previous example, 229 levels is calculated as 0.33 tt; adding 0.37 tt makes it to 0.70 tt. the script makes that equal to 342. however, the same script, puts 342, as being 0.60 tt when used like this: http://esi.coolinc.info/?level=342
i'd recommend not to use (yet) the "tt -> skill" feature, to avoid mixing sources of errors. however, please do use the "skill -> tt" feature to check if the formula works ok.

- afaik, there is a single polynomial to interpolate all data from 1 to 7.5k levels, and this creates necessarily errors which could be lowered by splitting the interpolation is smaller segments (0-2k, 2k-4k, 4k-6k, ...)

/jdegre.

ps.: pls ppl, nobody is at 6k-7k skill levels and have a couple of minutes to browse big chips on auct and report here???

The script uses a piecewise (0-550,550-4k, 4k-6k, 6k-7.5k) exponential fit following witte's approach, and is currently discontinuous when mapping from tt->level, so there are even more severe problems than you think. In the cases where a value is input that lies right on the boundary, it will stop iterating at a very sub-optimal value.

The inverse mapping is mainly to help out people who might need an easy approximation (EyeContact was asking about it), but as jdegre said, only the forward function level->tt should be used for testing.

I have a bunch of new data that should make a better fit possible in the mid to higher range thanks to MGMighty. I'll update this post when i have a new fit and script up.

Here's the deviation from the fit using all of sheet2 for 6k+ data, jdegre. The interpolations are obvious because they tend to cut across the sine, so we can tell when the resultant point is "good enough" when the interpolation converges on the sine curve.

I haven't updated the script but do have somewhat better fitting curves, as you can see. Also apparent should be the need to have a sine component at high skill levels to avoid too much error due to the exponentially increasing magnitude of the sine wave.

I still can't get convergence of the slope from the fit to 6k+ and the slope data i collected at 6k-16k levels, so our points above 7k need some refinement i think.

I'm quite pleased with the overall fit. Besides the points that are obvious interpolation errors we have a maximum error of less than 1.5% up to 7.8k (except the relatively unimportant 1-550 range where a single PEC can make the difference of several levels). In addition, my equation gives less error than a naiive poly fit approach would.

 

[Trabin]

Didn't have a massive amount to choose from, but I hope at least some of this helps....

Handgun:

5493->6541 239.32 PED

Rifle:

7395->7934 273.77 PED
7395->7768 191.15 PED
7395->7740 180.40 PED
7395->7657 147.46 PED
7395->7582 110.80 PED
7395->7558 96.74 PED

Combat Sense:

5870->6291 98.88 PED

Weapons Handling:

5725->6565 207.39 PED

Anatomy:

7042->7248 77.44 PED
7042->7179 55.34 PED

LWT:

5912->6592 185.95 PED
5912->6438 127.71 PED
5912->6435 126.63 PED
5912->6339 100.17 PED
5912->6252 82.65 PED
5912->6236 79.62 PED