/prog/ - Probability change

Name: Anonymous 2013-09-26 6:37

Suppose i have a set [A,B,C]
Probability of A =10% B=40% C=50%
If probability of C lowers to 25%, how the remainder value is distributed(proportionaly)?
I know it suppose to get like 15(+50%),60(+50%),25 but what if the set grows?
exmaple 5,77,3,15(-5) becomes 6.65,102,4,10
I'm looking for a formula for arbitrary sets of probability(not just 3).

Name: Anonymous 2013-09-26 6:50

You have a set X.
Do whatever you like to the set.
Once you're done scale the set such that the sum of all elements = 1 == 100%.
scaled X = (1 / sum(X))

Also do your own homework or you'll never learn

Name: >>2 2013-09-26 6:51

scaled X = (1 / sum(X)) * X
rather

Name: Anonymous 2013-09-26 6:58

>>2
>scale the set such that the sum of all elements =1
the problem, 1/sum(x) gives
(1 / sum(X)) * X
5,77,3,15 -> 5,77,3,10 -> (100/95)~=*1.05-> 5.25,80.85,3.15,10
(5 * (100 / 95)) + (77 * (100 / 95)) + (3 * (100 / 95)) + 10 = 99.4736842105
Which is 0.5% less than 100%

Name: Anonymous 2013-09-26 7:04

>>4
You didn't scale 10.

Name: Anonymous 2013-09-26 7:14

>>5
What? 10 is the new probability and cannot be altered(change from 15 to 10).

Name: Anonymous 2013-09-26 7:32

>>6
Then scale everything except the new probability, i.e. multiply it by (1 - p) / (sum(X) - p).

Also, you probably should do a proper Bayesian inference instead of pseudo-scientific probability adjustments with no actual basis in statistics.

Name: Anonymous 2013-09-26 7:47

>>7
I suck at statistics. Bayesian inference is probably much more complex and slower than this simple formula and harder to understand.
I use something like val1,val2,val3 where val1+val2 give probability to val3(manually adjusted).

Name: Anonymous 2013-09-26 8:36

>>8
I use something like val1,val2,val3 where val1+val2 give probability to val3(manually adjusted).
The problem is, without understanding how actually "val1+val2 give probability to val3", like, what happens in reality, your way of just adding them, then doing meaningless math tricks to get probability back to normal, is practically guaranteed to make no sense and yield weird results. When you say that it's easier to understand, I'm not sure you understand what "understand" is supposed to mean...

What are you trying to do?

Name: Anonymous 2013-09-26 8:48

>>9
letter1 'A' with letter2 'B' give 0.1% probability that letter3 will be 'Z'
computer generates random letters using 0%[prob1][prob2][prob3]100% intervals
letter3 will be in the one interval on which the probability lands.
By shrinking the interval(and proportionally adjusting all others) the probability is reduced(as set by tester or file input).
the current 2 letter generate third, then second and third generate fourth,etc. at random word length(2-16) the word is printed to screen.

Name: Anonymous 2013-09-26 9:07

What's the probability of getting dubs on a slow board with predictable post numbers?

Name: Anonymous 2013-09-26 9:13

>>11
1/6
pure random chance

Name: Anonymous 2013-09-26 9:15

>>11
depends on post rate, close to 100%.
If you select threads off the front page its 100%. Regardless its zero-effort dubs, unlike imageboards where you adaptive post rate adjustments(take whole board average post rate and predict the next number based on current) to create the possibility of dubs.

Name: Anonymous 2013-09-26 9:20

>>13
Yes, but does that take into account the recently discovered race condition that we found in Shii_{_{_t}}chan? Does it factor in the fact that lower numbers are more likely than greater ones? Have you considered the extra processing time due to CAPTCHA?

Name: Anonymous 2013-09-26 9:25

What drives a man to study statistics?
Fear of determinism?
The desire for anarchy and hatred of the illuminati?
The belief that humans are in essence statistics machines and that they can only be emulated using our inherently rational and probabilistic artificial intelligence techniques?
Or simply so they can sound smart on the internet?

Name: Anonymous 2013-09-26 9:27

>>14
>Yes, but does that take into account the recently discovered race condition that we found in Shiitchan?
What condition?
>Does it factor in the fact that lower numbers are more likely than greater ones?
What?
Captcha:
1.you prefill captcha before it or use OCR
2.when the timer sets off(with latency to server adjustment) it autoclicks() submit.

Name: Anonymous 2013-09-26 9:29

>>15
I don't study statistics i write code.

Name: Anonymous 2013-09-26 9:31

>>16
Are you saying that you know what Bayesian inference is but don't know what race conditions are?
What is this I don't even.
Are you a mathematician pretending to be a programmer or something?

Name: Anonymous 2013-09-26 9:37

>>18
1.I know what is a "race condition" . I don't know what kind of race condition is on shiichan(i assume something related to post numbers, as this is about dubs)
2.I'm not a programmer or mathematician, i just happen to have a hobby which involves writing code.

Name: Anonymous 2013-09-26 10:59

>>16
/g/fags can't quote

Name: Anonymous 2013-09-26 11:06

>>20
>le pedophile elitism
>le pedophile meme pipes

Name: Anonymous 2013-09-26 11:42

>>10
Ah, OK then.

I don't quite get why you need to adjust intervals as such. You get your matrix of probabilities from counting letter combinations in the wild, I assume, so why not alter those counts directly?

Also, are you sure that manually adjusting some of the about 20000 weights will have any noticeable effect? What do you mean that this can be done "by file input", do you actually have no idea that you're supposed to construct the probability matrix from file input directly?

Name: Anonymous 2013-09-26 11:48

>>22
file input e.g. wikipedia text will contain many repeating sequences like "-ing" which will skew the result towards more common combinations.
There are 26*26*26 probabilities, of course changing them is slow.
There will be a command like S:AFG -50.6% which will reduce probability of AFG by 50.6%.
Its like interactive dialogue, computer shows you text you reduce prob of bad combos.

Name: Anonymous 2013-09-26 11:57

The purpose is to create phonetically correct(with a liberal definition of correctness) words, not to create words from rigid datasets. File input is for adjusting the probabilities like "batch mode", not "learning good letter combinations from limited linguistic corpus".
The words may not exist at all, but possible to pronounce. In other words i'm trying to make a program which will create random phonetically-correct words.

Probability change

1 Name: Anonymous 2013-09-26 6:37

2 Name: Anonymous 2013-09-26 6:50

3 Name: >>2 2013-09-26 6:51

4 Name: Anonymous 2013-09-26 6:58

5 Name: Anonymous 2013-09-26 7:04

6 Name: Anonymous 2013-09-26 7:14

7 Name: Anonymous 2013-09-26 7:32

8 Name: Anonymous 2013-09-26 7:47

9 Name: Anonymous 2013-09-26 8:36

10 Name: Anonymous 2013-09-26 8:48

11 Name: Anonymous 2013-09-26 9:07

12 Name: Anonymous 2013-09-26 9:13

13 Name: Anonymous 2013-09-26 9:15

14 Name: Anonymous 2013-09-26 9:20

15 Name: Anonymous 2013-09-26 9:25

16 Name: Anonymous 2013-09-26 9:27

17 Name: Anonymous 2013-09-26 9:29

18 Name: Anonymous 2013-09-26 9:31

19 Name: Anonymous 2013-09-26 9:37

20 Name: Anonymous 2013-09-26 10:59

21 Name: Anonymous 2013-09-26 11:06

22 Name: Anonymous 2013-09-26 11:42

23 Name: Anonymous 2013-09-26 11:48

24 Name: Anonymous 2013-09-26 11:57