Module prand

source code

Mersenne Twister based random number generator

This module is intended to provide a reasonable fast random number generator object that allows for (1) multiple generators running in parallel and (2) saving/restoring of generator state to allow easy regeneration of number sequences and (3) use a welll documented algorithm (Mersenne Twister) that can easily be implemented or exists in Matlab.

Starting with Matlab 7.1 the built in RAND() function uses the same Mersenne Twister engine that python's (2.4 on) random.Random() object uses. This means you can go back and forth pretty easily. The only catach is that python's state vector is composed of signed long's, while matlab's is unsigned. This means that to take the python state vector into matlab you need to add 2^32 to all negative values in the state vector before passing to the RAND function:

>> s = [state vector from puthon..] >> s(s<0)=s(s<0)+(2^32); rand('twister', s);

Will put the generator at the same state the python generator was at when the state was saved. RAND(100) in matlab will now generate the same random sequence the python call:

>> PypeRandom(seed=statevector).rand(100)

would. At least to about +- 1.0e-11 as far as I can tell..

Author -- Matt Krause (matthew.krause@yale.edu)