[Randtest-develop] randtest 1.0
Status: Alpha
Brought to you by:
drmike2
|
From: Mike R. <er...@es...> - 2002-01-02 05:02:31
|
Howdy folks, Happy New Year y'all :-) I've pushed the longest runs test has far and as hard as I can. I'm now going to give up. It seems that there's no shortcuts that work, and brute force computing the chi^2 tables can take a *very* long time, even on 1GHz computers. However, for block sizes less than 500 bits, the calculations are exact and can be found in less than 1 hour. For 10000 bit blocks, computing m=1 or m=2 is *really* long, and even for m=6 or m=8 it's going to take a while using brute force. The "bigfloat" package was expaned to include polynomial manipulation so I could add the natural logarithm via Tchebyshev expansions. This was just cut and paste from other stuff, but this is a good excuse to get it out to the public domain. If anyone wants to take charge of the randtest project, I'll be very happy to turn over the passwords. The major things that need to be done are to mesh the hex input and FIPS tests so that hex input can feed the NIST tests (which now only work with binary data) and FIPS tests can work with binary data. More tests can also be added. The general idea is that there's no cabal of high priests that dictate what blocksizes or probabilities get used. Everything should be computed from basic principles so the users know the right answer is really coming out (but it all still needs lots of testing to prove that :-) It's kind of cool that problems from the 1800's still can't be computed in "reasonable" time even with 1 GHz processors. But I'm heading for physics problems from the 1910's, so I leave randtest in the state it's in. I hope it will be a small core of what comes next. Patience, persistence, truth, Dr. mike |