Strategy to make a 91+ LP genie

Yuniryu - Raging Tide

serbetel wrote: »

No because 1/10 has nothing to do with following numbers,
its just the start sequence of a potentially valid combination.

We dont even are able to ditch it instantly, it is in your bag anyway
and if your really think it has a low chance - then you wont loose much xp,
but yeah you could waste that week for no reason and leave the chance unused
but maybe you could explain why you would.. b:sweat

Essentially what you've said is that if you pick 1 number out of a possible 10 the chance that your number is going to come up is 50% because it's either the number you picked or not.

What kind of messed up shiit are you smokin brah?

HrunsPanda - Archosaur

Fissile - Archosaur wrote: »

@HrunsPanda

It is a custom script written for the purpose. It's not that complicated to write.

I know, but its easier to click a link and download it

I understand however that may not be an option since custom scripts usuall lack user interfaces and probably have not very well organised code so you would have to provide an explanation to go with it or it'd take me more time to figure it out than to write it

serbetel

Yuniryu - Raging Tide wrote: »

Essentially what you've said is that if you pick 1 number out of a possible 10 the chance that your number is going to come up is 50% because it's either the number you picked or not.

What kind of messed up shiit are you smokin brah?

No i said the chance to get on in this specific setup is allways 50:50
you simply have no guarantee that your 80/80 will win over my 71/80 or 1/10
you simply dont have

Adroit - Lost City

HrunsPanda - Archosaur wrote: »

Adroit, sorry if it is posted already somewhere in these 12 pages but i dont wanna go search b:surrender
Do you have a link to whatever you use for these simulations ?

PS: and is it possible to simulate with hugely larger number of genies?, because 10.000 is way too small a sample size for such a small number. You can recognize that in the standard deviation being almost as big as the average SP used. I think you should simulate a million genies.

I think i might make that excell sheet after all, even though i am not yet entirely sure how i am gonna implement the totally different variables of SP, mirages and time into one formula.

Yeah as Fiselle is correct, I wrote a short script to simulate these. If you are curious I'd be happy to share, just shoot me a pm.

It is possible to simulate more genies, but it takes a long time on this laptop.. realize that when I say I simulate 10,000 genies, that doesn't mean 10,000 genies tried and just picked the 91+ out of there.. that means I simulated until I had 10,000 91+ genies.. so for example the case where I only continued with perfect genies to level 30, then as long as it had 90% of more lucky points after that I kept going.. the average mirage cost was over 150k (meaning about 50k genies tried per 91+ genie), I ended up simulating 500 million genies using that method.. which took my laptop almost 10 hours to complete. My program right now is single-threaded, and would probably benefit from using more threads, but I'm actually on vacation right now so I start my script going and leave for the day.. then look at the results when I get back. I'm really not sure if simulating more genies would decrease the standard deviation, the SP cost per genie is all over the place.. so the standard deviation will also be very large no matter how many trials I run. I actually have two different scripts, one that runs trials and generates a log file, and another that reads the log file and gives me all the statistics I'm interested in.. and while it's running I often times check out the statistics for what it has so far. I've noticed that after ~1k 91+ genies.. the stats don't really change much, so I think the sample size I'm using is satisfactory. I'd be more than happy to give you the script and you can run it to generate your own files or add on to what I have already, but I probably won't be running more than 10k trials for my own personal use.

Decus - Dreamweaver

serbetel wrote: »

No i said the chance to get on in this specific setup is allways 50:50
you simply have no guarantee that your 80/80 will win over my 71/80 or 1/10
you simply dont have

I don't think anyone made a claim that one is absolutely possible and one isn't. I understood, and rightfully so, that one is more likely then another. It is highly inefficient to use every single combination of LPs to get to 91+

HrunsPanda - Archosaur

Adroit - Lost City wrote: »

Yeah as Fiselle is correct, I wrote a short script to simulate these. If you are curious I'd be happy to share, just shoot me a pm.

It is possible to simulate more genies, but it takes a long time on this laptop.. realize that when I say I simulate 10,000 genies, that doesn't mean 10,000 genies tried and just picked the 91+ out of there.. that means I simulated until I had 10,000 91+ genies.. so for example the case where I only continued with perfect genies to level 30, then as long as it had 90% of more lucky points after that I kept going.. the average mirage cost was over 150k (meaning about 50k genies tried per 91+ genie), I ended up simulating 500 million genies using that method.. which took my laptop almost 10 hours to complete. My program right now is single-threaded, and would probably benefit from using more threads, but I'm actually on vacation right now so I start my script going and leave for the day.. then look at the results when I get back. I'm really not sure if simulating more genies would decrease the standard deviation, the SP cost per genie is all over the place.. so the standard deviation will also be very large no matter how many trials I run. I actually have two different scripts, one that runs trials and generates a log file, and another that reads the log file and gives me all the statistics I'm interested in.. and while it's running I often times check out the statistics for what it has so far. I've noticed that after ~1k 91+ genies.. the stats don't really change much, so I think the sample size I'm using is satisfactory. I'd be more than happy to give you the script and you can run it to generate your own files or add on to what I have already, but I probably won't be running more than 10k trials for my own personal use.

Oh thanks, i see that makes a good sample size indeed, i misunderstood that number yes
I am currently traveling myself and powering my laptop with a car battery, so i dont think i feel like running 10 hour simulations

MiniST - Sanctuary

serbetel wrote: »

No i said the chance to get on in this specific setup is allways 50:50
you simply have no guarantee that your 80/80 will win over my 71/80 or 1/10
you simply dont have

this is delving into the area of philosophy, which is why i said 1+1=2 is not provable. And my variables are 1 and 2. b:pleased

Adroit - Lost City

HrunsPanda - Archosaur wrote: »

Oh thanks, i see that makes a good sample size indeed, i misunderstood that number yes
I am currently traveling myself and powering my laptop with a car battery, so i dont think i feel like running 10 hour simulations

haha yeah, don't blame you! Most of the simulations take a few hours, the less efficient strategies take quite a bit longer :P If you have any strategies you're curious about, I'd be more than happy to simulate them for you.. I'm kinda racking my brain trying to think of ideas that might be better than what I've already tried, would love to get try some different ideas :P

serbetel

Decus - Dreamweaver wrote: »

I don't think anyone made a claim that one is absolutely possible and one isn't. I understood, and rightfully so, that one is more likely then another. It is highly inefficient to use every single combination of LPs to get to 91+

That would be true if high LP would increase chance to get high LP each next decade
but infact high LP only matter in the last decade before 91 and by throwing away all other valid combinations that last decade is considerably tiny.

HrunsPanda - Archosaur

serbetel, your reasoning would make sence if there were only the option to get 1 point or 10 points.

However, being critical in what genies we accept allows us to get 9 points on every occasion except one. And since we probably get 10 points some time instead of 9 anyway, we can even accept 8 points.
Whereas when we take that first 1/10, we have to get 10 out of 10 every time after.

HrunsPanda - Archosaur

Adroit - Lost City wrote: »

haha yeah, don't blame you! Most of the simulations take a few hours, the less efficient strategies take quite a bit longer :P If you have any strategies you're curious about, I'd be more than happy to simulate them for you.. I'm kinda racking my brain trying to think of ideas that might be better than what I've already tried, would love to get try some different ideas :P

I am going to start an attempt at making a spreadsheet tonight. If i succeed at making something i deem worth sharing, you can check its conclusions with simulations

Fissile - Archosaur

Adroit - Lost City wrote: »

Yeah as Fiselle is correct, I wrote a short script to simulate these. If you are curious I'd be happy to share, just shoot me a pm.

It is possible to simulate more genies, but it takes a long time on this laptop.. realize that when I say I simulate 10,000 genies, that doesn't mean 10,000 genies tried and just picked the 91+ out of there.. that means I simulated until I had 10,000 91+ genies.. so for example the case where I only continued with perfect genies to level 30, then as long as it had 90% of more lucky points after that I kept going.. the average mirage cost was over 150k (meaning about 50k genies tried per 91+ genie), I ended up simulating 500 million genies using that method.. which took my laptop almost 10 hours to complete. My program right now is single-threaded, and would probably benefit from using more threads, but I'm actually on vacation right now so I start my script going and leave for the day.. then look at the results when I get back. I'm really not sure if simulating more genies would decrease the standard deviation, the SP cost per genie is all over the place.. so the standard deviation will also be very large no matter how many trials I run. I actually have two different scripts, one that runs trials and generates a log file, and another that reads the log file and gives me all the statistics I'm interested in.. and while it's running I often times check out the statistics for what it has so far. I've noticed that after ~1k 91+ genies.. the stats don't really change much, so I think the sample size I'm using is satisfactory. I'd be more than happy to give you the script and you can run it to generate your own files or add on to what I have already, but I probably won't be running more than 10k trials for my own personal use.

You can calculate how good your estimates of the average number of genies, average amount of spirit spent, standard deviation for number of genies, and standard deviation for amount of spirit spent by calculating the standard error for each of these. When the standard error is only a tiny fraction of the the estimator* in each case, then you can be (quantifiably) confident in your results.


* The values you are reporting are only estimates of the true mean and standard deviation of the random distribution.

Fissile - Archosaur

Adroit - Lost City wrote: »

haha yeah, don't blame you! Most of the simulations take a few hours, the less efficient strategies take quite a bit longer :P If you have any strategies you're curious about, I'd be more than happy to simulate them for you.. I'm kinda racking my brain trying to think of ideas that might be better than what I've already tried, would love to get try some different ideas :P

There is no reason it should take that long. In the time I wrote my above post i was able to simulate 10,000 successful genies for a particular strategy (over 90 million total genies). My guess is you are writing the results of every single genie out to your file, which is unnecessary and also very slow.

If you track the total number of genies simulated plus the sum of the squares of the number of genies generated per successful genie, then you can calculate the statistics at the end of your simulation with some pretty simple formulae. E.G.

For each time you get a 91+ genie (spirit is # of spirit points spent to make that 91+ genie):
SumSpirit = SumSpirit + Spirit
SumSpiritSquare = SumSpiritSquare + Spirit * Spirit

Then after you are done making all genies (i is the total number of 91+ genies... 10,000 in your case):
SpiritSampleMean = SumSpirit / i
SpiritSampleStDev = Sqr(SumSpiritSquare / (i - 1) - SpiritSampleMean * SpiritSampleMean)
SpiritMeanSE = SpiritSampleStDev / Sqr(i)


Here SpiritMeanSE is the standard error in the SpiritSampleMean estimator. b:victory
Sqr() is the square root function.

Adroit - Lost City

Fissile - Archosaur wrote: »

You can calculate how good your estimates of the average number of genies, average amount of spirit spent, standard deviation for number of genies, and standard deviation for amount of spirit spent by calculating the standard error for each of these. When the standard error is only a tiny fraction of the the estimator* in each case, then you can be (quantifiably) confident in your results.


* The values you are reporting are only estimates of the true mean and standard deviation of the random distribution.

If I were at home and bored, I'd probably read up and do this.. but being on vacation is making me very lazy. Maybe I'll look into it when I get back, just got lots of stuff to do.. and I'm already reasonably confident that my data is accurate enough to use for my purpose.

HrunsPanda - Archosaur

yes i was also making a misinterpretation of the standard deviation. Indeed, i have most always been using it only as standard error calculation and thats what i thought i was looking at (to keep an eye on your poker results and decide within a certain statistic (as high as possible ofc, but that requires 100.000s of games played) certainty your winning rate)

Looking at the numbers, i also think your data is accurate enough.

Fissile - Archosaur

Adroit - Lost City wrote: »

If I were at home and bored, I'd probably read up and do this.. but being on vacation is making me very lazy. Maybe I'll look into it when I get back, just got lots of stuff to do.. and I'm already reasonably confident that my data is accurate enough to use for my purpose.

b:thanks Actually, you will find that running 10,000 cases is overkill by a good bit.

serbetel

HrunsPanda - Archosaur wrote: »

serbetel, your reasoning would make sence if there were only the option to get 1 point or 10 points.

However, being critical in what genies we accept allows us to get 9 points on every occasion except one. And since we probably get 10 points some time instead of 9 anyway, we can even accept 8 points.
Whereas when we take that first 1/10, we have to get 10 out of 10 every time after.

It is really hard for me to belive that that many ppl fall for the wrong assumption
that pwi codes the seeder to just send out 91+ genies
wich are produced using only 8s, 9s and 10s.

You dont really belive that do you ?

HrunsPanda - Archosaur

jeez i was even giving you a context where your reasoning makes sence. Take the chance and accept that to save a little bit of your grace dude ! Must you really keep holding on the this attempt to ridicule yourself ?

Adroit - Lost City

I see your formula is pretty simple, I could probably implement that pretty quick in my script. I record each 91+ genie to my file, but any genie that doesn't make it to 91+ is not added to the file. I'm not entirely sure why my script is running so slow, it's weird that my processor usage never goes above 7%.. I would have expected at least one core to be fully utilized but that doesn't seem to be the case. I'd assume the script would run much faster on my desktop, but that is 400 miles away atm and I forgot to re-enable port forwarding for my VPN before I left.. so cannot access my home network and run the script remotely or anything..

serbetel

HrunsPanda - Archosaur wrote: »

jeez i was even giving you a context where your reasoning makes sence. Take the chance and accept that to save a little bit of your grace dude ! Must you really keep holding on the this attempt to ridicule yourself ?

How do you think the valid combinations are distributed ?

i can give you a hint - equally.

That means you can target the lower possbilities or the higher
it would make completely no difference
if there wasnt the factor of costs and time.

Adroit - Lost City

SweetieBot, take 2 points from serbetel for continuing to spam this thread with nonsense.

SweetieBot - Lothranis

Adroit - Lost City wrote: »

SweetieBot, take 2 points from serbetel for continuing to spam this thread with nonsense.

Adroit - Lost City removes 2 points from serbetel!
serbetel now has a total of -2 points and is in 108th place.
Adroit - Lost City can still remove another 3 points today.

Check this thread for the current high scores and to learn how to award points to others.

serbetel

b:chuckle you make me feel like Gandhi lol

Fissile - Archosaur

serbetel wrote: »

How do you think the valid combinations are distributed ?

i can give you a hint - equally.

Not true. The actual distribution was posted earlier in this thread by Asterelle.

serbetel wrote: »

That means you can target the lower possbilities or the higher
it would make completely no difference
if there wasnt the factor of costs and time.

And our point is neglecting costs in a cost optimization problem is nonsensical.

serbetel

Fissile - Archosaur wrote: »

Not true. The actual distribution was posted earlier in this thread by Asterelle.

You ignore the nature of statistic wich is fatal when trying to create statistical formula.

Fissile - Archosaur wrote: »

And our point is neglecting costs in a cost optimization problem is nonsensical.

Cost optimization is a sideffect but i just focus on the chances available
not on a more expensive and lower subset of chances.



There exist exactly as much chance for 91er with lower first 4 decades as for higher 4 decades - its called luck.

Fissile - Archosaur

serbetel wrote: »

You ignore the nature of statistic wich is fatal when trying to create statistical formula.

I think the university that gave me my graduate degree would tend to disagree with you on that point.

Adroit - Lost City

Times like this make me wish there was a "ban noob from thread" option.. but there isn't.. so best we can do is just ignore it.

ovenusarmanio

Serbetel could you please stick to the topic of the thread, which is what is the most cost effective method of leveling a genie to 91+? Statistical theory (sorry don't know the math terms xD) isn't what is being discussed here, it's only cost-effective genie leveling strategies. Thank you.

serbetel

Sorry but i m not using it - they do.

Not funny that insults are tolerated but so called "spamming" wich in fact is just answering questions and provocations makes a Mod take actions =\

Magnanimous_ - Heavens Tear

serbetel wrote: »

Sorry but i m not using it - they do.

Not funny that insults are tolerated but so called "spamming" wich in fact is just answering questions and provocations makes a Mod take actions =\

Because you are a complete idiot by spamming misinformation, using the wrong assumptions, trying to promote an approach that does not work, and attempting to explain probability while having a limited knowledge of it.