Author Topic: Die hard tests  (Read 1258 times)

0 Members and 1 Guest are viewing this topic.

heatmap

Die hard tests
« on: November 16, 2017, 10:33:39 PM »
This is how people test for random qualities within a sequence of numbers.

https://en.m.wikipedia.org/wiki/Diehard_tests

The following users thanked this post: kav, Reyth

Reyth

Re: Die hard tests
« Reply #1 on: December 11, 2017, 05:04:07 AM »
Hmmm... this post is getting some views.  Who knows some of us may have experienced some rather (ahem) unusual results with RNG's...

Jesper

Re: Die hard tests
« Reply #2 on: December 11, 2017, 07:03:54 AM »
We need a very large sample to test random numbers. Not any test can 100% verify random numbers. They can give a probability of random, but never reach certain.

Those  below two number streams have the same probability.

12,12,12,12,12,12,12,12,12,12
22,34,1,6,11,0,17,28,16,5

I am almost sure there is nobody on this forum who have seen the first, and I am equally sure I am
the only one seen the second.

Reyth

Re: Die hard tests
« Reply #3 on: December 11, 2017, 07:25:26 AM »
It is useless to point out the second; its like saying the sky is blue -- its no different than ANY other sequence -- all sequences and so it holds no value.

Jesper

Re: Die hard tests
« Reply #4 on: December 11, 2017, 08:31:18 AM »
At my latitude at this time of the year, I can not see a blue sky.   When they test for randomness, they do take statistic in account, even if a large bias is theoretical possible, it is so rare they can  judge the device as weak.

dobbelsteen

Re: Die hard tests
« Reply #5 on: December 11, 2017, 10:20:21 AM »
I am not be in want of testing RNGs. I am interested , if there is a connection between our bets ( input) and the outcome of the RNG. This can be done the best with a large number bet, because the DTOP is very small.
Do some  long run tests and bet the last 5 DSs. If the loss is more than 2,7 %, do not trust the software.

The following users thanked this post: Reyth

heatmap

Re: Die hard tests
« Reply #6 on: January 16, 2018, 03:00:15 AM »
@dobblesteen from what i understand (which isnt much) about chi squared test, it actually tells you if there IS a correlation between one variable and another.