The difference between np.random.seed(int) and np.random.seed(array_like)?

Question

In Python's numpy library, the np.random.seed method can accept two different types of parameter: int and array_like[int].

What's the difference between them? Such as: np.random.seed(2) and np.random.seed([2013, 1, 4]).

Answer 1

The state of the underlying Mersenne Twister PRNG is very large, 624 32-bit integers, to be exact. If given an integer seed, the initialization routine will run a smaller PRNG to expand that single 32-bit integer out to the full 624-element state. This does mean that you can't access the vast majority of the possible states.

Similarly, if given a sequence of integers as the seed, then a different smaller PRNG will use that to expand out to 624 elements, but one that can use the whole array that you pass it. This lets you access the whole space of initial states, if such a thing matters to you. This algorithm is shared between the standard library's random module and numpy.random.