[Randtest-develop] Yet another version put up

[Mike R. <er...@es...>]

OK, I fixed a lot of bugs in the integration routine.  It had some
pathelogical problems for moderate values of chi^2 (I was getting p-values
larger than 1!).  

I also added the ability to loop on the binary rank tests.  So it does
both big and little endian analysis for binary matricies of size 8 and up
(barfs for sizes less than that).  I need to write up the documentation
for the math, it's based on the NIST doc's, but I used more levels.

The comand line args I added were -x which means use the input file as
binary and -p<value> which means start the binrank rank test with
dimension value and go up to nWordSize (inclusive).  It seems to work ok
since the file cvs-1.11.1p1.pdf gave these kinds of results:

dimension = 8
Little endian chi^2 = 245604616.624208,  p-value = 0.000000
Big endian chi^2 = 245604616.624208,  p-value = 0.000000

dimension = 9
Little endian chi^2 = 3.102747,  p-value = 0.315851
Big endian chi^2 = 5.510977,  p-value = 0.643256

dimension = 10
Little endian chi^2 = 21.423159,  p-value = 0.999326
Big endian chi^2 = 24.049784,  p-value = 0.999788

dimension = 11
Little endian chi^2 = 5.328044,  p-value = 0.622830
Big endian chi^2 = 12.855175,  p-value = 0.975226

dimension = 12
Little endian chi^2 = 146304292.933449,  p-value = 0.000000
Big endian chi^2 = 70124717.373940,  p-value = 0.000000

dimension = 13
Little endian chi^2 = 7.113900,  p-value = 0.787694
Big endian chi^2 = 1.592002,  p-value = 0.097785

dimension = 14
Little endian chi^2 = 2950633.552405,  p-value = 0.000000
Big endian chi^2 = 2950620.678294,  p-value = 0.000000

i.e. even byte boundaries and ascii boundaries gave unrealistic
(non-random) results.  Whis is what you'd expect from ascii data!

More testing is needed, so beat on it when you get a chance.

Patience, persistence, truth,
Dr. mike