|
Science Forum Index » Space - Consult Forum » Test for randomness
Page 1 of 1
|
| Author |
Message |
| Guest |
 Posted: Sat Jan 20, 2007 4:16 am |
|
|
|
|
Dear All,
I want to whether Kolmogrov Simirnov test can be used to test the
randomness for the continuous variables
(Run test can be used to test for dicotonomus variables and continous
variables can be coverted using the median )
regards,
vinaitheerthan |
|
|
| Back to top |
|
| David Winsemius |
| Posted: Sun Jan 21, 2007 1:56 pm |
|
|
|
Guest
|
vinaitheerthan@aims.amrita.edu wrote in news:1169281001.066461.313400
@a75g2000cwd.googlegroups.com:
Quote: Dear All,
I want to whether Kolmogrov Simirnov test can be used to test the
randomness for the continuous variables
(Run test can be used to test for dicotonomus variables and continous
variables can be coverted using the median )
KS test measures the difference between 2 distributions. It has one sample
and two sample configurations. In the one sample set-up, a specified
distribution (e.g a Normal or a uniform) is compared to a sample
distribution. It would let you ask the question, "Does this sample look
like it was drawn randomly from a Normal distribution" but it would not be
a test of "randomness" per se. There was at least one extended thread on
one or more of the sci.stat.* groups in the last year on the merits of the
KS test versus other options, but again this was not really a discussion of
testing for "randomness".
--
David Winsemius |
|
|
| Back to top |
|
| David A. Heiser |
| Posted: Tue Jan 23, 2007 4:08 pm |
|
|
|
Guest
|
<vinaitheerthan@aims.amrita.edu> wrote in message
news:1169281001.066461.313400@a75g2000cwd.googlegroups.com...
Quote: Dear All,
I want to whether Kolmogrov Simirnov test can be used to test the
randomness for the continuous variables
(Run test can be used to test for dicotonomus variables and continous
variables can be coverted using the median )
regards,
vinaitheerthan
++++++++++++++++++++++++++++++++++
The KS Test is not a test for randomness.
Knuth (volume 2) used it to test for certain properties of a random number
sequence.
There are very many properties of a random number sequence that are
measured/tested for. There is a large number of tests reported as being used
in the literature, but only a very few are readily available. The
appropriate test sequences to be used are dependent on the use of the
resulting sequence. If the application is non-cryptographic, then a freely
available test sequence (the Diehard II tests) is applicable. It requires a
data set of 30 million RN's represented as a unsigned 32 bit sequence.
David Heiser |
|
|
| Back to top |
|
| |
