pythonrandomnumpycython

Correct way to generate random numbers in Cython?


What is the most efficient and portable way to generate a random random in [0,1] in Cython? One approach is to use INT_MAX and rand() from the C library:

from libc.stdlib cimport rand
cdef extern from "limits.h":
    int INT_MAX
cdef float randnum = rand() / float(INT_MAX)

Is it OK to use INT_MAX in this way? I noticed that it's quite different from the constant you get from Python's max int:

import sys
print INT_MAX
print sys.maxint 

yields:

2147483647  (C max int)
9223372036854775807  (python max int)

Which is the right "normalization" number for rand()? EDIT additionally, how can the random seed be set (e.g. seeded based on current time) if one uses the C approach of calling rand() from libc?


Solution

  • The C standard says rand returns an int in the range 0 to RAND_MAX inclusive, so dividing it by RAND_MAX (from stdlib.h) is the proper way to normalise it. In practice, RAND_MAX will almost always be equal to MAX_INT, but don't rely on that.

    Because rand has been part of ISO C since C89, it's guaranteed to be available everywhere, but no guarantees are made regarding the quality of its random numbers. If portability is your main concern, though, it's your best option, unless you're willing to use Python's random module.

    Python's sys.maxint is a different concept entirely; it's just the largest positive number Python can represent in its own int type; larger ones will have to be longs. Python's ints and longs aren't particularly related to C's.