Using NumPy to generate a sequence of random integers with variable max/min

user_3808639 0
Viewed 988

numpy.random.uniform supports array arguments, but numpy.random.randint does not:

import numpy as np 
two_random_uniforms = np.random.uniform([0, 1], [3, 4])
this_raises_exception = np.random.randint([0, 1], [3, 4])

The behavior I want from the second line of code is equivalent to:

result = []
for lown, highn in zip([0, 1], [3, 4]):
    result.append(np.random.randint(lown, highn))

Is there a way to accomplish the random integer generation with variable max/min without writing a Python loop? The above workaround will be unacceptably slow for the large arrays that my application requires. I can write the loop in Cython, but I'd rather use NumPy if possible.

random numpy python performance
1 Answers

One simple thing you could do is just generate floats and round them.

def my_randint(low, high):
    return np.floor(np.random.uniform(low, high))

It's possible this procedure might tweak the resulting distribution a bit, to make some integers ever so slightly more likely than others, but unless you are working with cryptography or some other sensitive application, it probably won't matter.

user_565410
Related