Random Numbers

In data science contexts random numbers are important for simulating data and for selecting random subsets of data. NumPy provides a submodule random for creating arrays of (pseudo-)random numbers.

import numpy as np

Random Number Generators

NumPy supports several different algorithms for generating random numbers. For our purposes choice of algorithm does not matter (for crypto applications it matters!). Luckily NumPy provides a default one.

We first have to create a random number generator object (or get the default one) and initialize it with a seed. The seed determines the sequence of generated random numbers. Using a fixed seed is important if we need reproducable results (when testing things, for instance).

rng = np.random.default_rng(123)    # use some integer as seed here

Getting Random Numbers

Random numbers may follow different distributions. NumPy provides many standard distributions, see Random Generator in NumPy’s documentation.

# random integers (arguments: first, last + 1, shape)
a = rng.integers(23, 42, (4, 10))

print(a)

[[23 35 34 24 40 27 27 26 29 26]
 [29 38 31 40 31 28 37 38 39 39]
 [23 32 28 27 27 38 38 27 30 37]
 [25 34 31 40 37 27 38 38 27 32]]

# uniformly distributed floats in [0, 1)
a = rng.random((4, 4))

print(a)

[[0.23155562 0.16590399 0.49778897 0.58272464]
 [0.18433799 0.01489492 0.47113323 0.72824333]
 [0.91860049 0.62553401 0.91712257 0.86469025]
 [0.21814287 0.86612743 0.73075194 0.27786529]]

# permutation of an array
a = np.array([1, 2, 3, 4 ,5])
b = rng.permutation(a)

print(b)

[4 2 5 3 1]