Chromatic Orb probabilities Spreadsheet (Update: N = 1570)
" This is basically true. If you're trying to calculate the odds of getting a specific color combination in total, then you do need to take into account the fact that it's impossible to roll the exact same color combination that you had before. If you didn't have to worry about that you could just calculate P (as above) and since you're rolling from 0 - 1 the true probability would be P. But since we've eliminated a possibility, we're not actually rolling from 0 - 1. Let's let P_old be the probability that we roll the exact same set of colors that we had previously. The true probability should be P / (1 - P_old). So if P_old is very unlikely then the true probability approaches P, but if P_old is very likely (maybe because we only have 1 socket on the item), then the true probability will be quite distorted. | |
|
Bumping to get added visibility, and to let people know that I've added the code to GitHub so that anyone can see and download the code for themselves.
It's all written in MATLAB so in order to run it you'd sadly need access to that program. I can't vouch for whether or not the code would work on Octave. At some point I might rewrite it in Perl or Python, but that would really just be an exercise for myself more than anything else. | |
|
Added some data from recolouring my shav. I expected to provide more data but the first 4off-colours were the ones I wanted and not some crappy combinations like I usually get :D
| |
" Thanks! And grats on that sick roll. Gotta love those off-colors. | |
|
I just used around 1000 chroms trying to get 5 off colors (still not successful), shame I didn't know about this thread earlier. I will probably attempt some more once I get more chroms. Will try to keep a log of my tries. Pure Dex chest.
Spoiler
I AM MAD ZAP!ZAP!ZAP! ME SOOO WIZZARD! Last edited by gathor on Feb 12, 2014, 3:39:19 AM
| |
" Good luck with it! Yeah, that's a real project getting 5 off-colors on that. On the plus side, I think the devs confirmed that chromes roll off the reduced stat requirements rather than the originals so that makes it slightly easier. I will happily take any data if you start rolling it again. I'm not playing as much lately so I've only been slowly stocking up on chromes again to start adding my own data. | |
|
added a lot of data for you.
|
|
" Thanks! This is really good data. At first I was pretty worried because the estimate dipped from 14 to 12 with the addition of the new data. But I think this is because 1) the process is random enough that variation of a couple points with only 1200 data points is actually not so rare, and 2) the early data is biased towards high values of X because people tend to be rolling for rare outcomes. If they don't get it for a while, then the bias is very small, but if they get it quickly the bias becomes larger. | |
|
Just to throw in, I don't think the probability is a constant.
I'm pretty sure it draws from an altering seed. http://www.pathofexile.com/forum/view-thread/378472/page/1 This was a 191 pure str chest within 63 chromes.  Casually casual. Last edited by TheAnuhart on Feb 15, 2014, 8:17:45 PM
|
|
" Hmm, I'm not really sure what you're getting at here. What do you mean by altering seed? Sure the pseudo-random process has to be seeded somewhere. But let's say that there's some table of socket colors, GGG seeds it somewhere in that table and then when you roll a chrome you just get the next set of socket colors. Sure, it might be implemented that way, but if we want to model it then all we care about is how that table was created in the first place, because that will reflect the underlying probabilities. At least that's how I'm reading your comment right now. I think there's good evidence right now that the model is capable of explaining the data quite well despite the fact that it only has one parameter. The exact estimate of X is going to shift around quite a bit just because the process itself is so random, but the model has no trouble mimicking the patterns that we see in the data. |
