NumPy equivalent to random.getrandbits()

Question

I'm working on creating a subclass of random.Random() where the generator is actually NumPy's default Random Generator.

The reason for this is, I can leverage parallel execution using NumPy's generator via the SeedSequence object.

Overriding methods random() and seed() is fairly easy, however, I also need to override method getrandbits(), but I haven't been able to find any Numpy's equivalent function/solution. getrandbits() is used heavily by all other functions of class random.Random().

Out of curiosity, why do you need to subclass stdlib random generators? — Andras Deak -- Слава Україні
– Andras Deak -- Слава Україні, Commented Jan 11, 2022 at 15:02
you can take a look at the source for random.getrandbits and port that to numpy — jkr
– jkr, Commented Jan 11, 2022 at 15:05
@jakub indeed, my thinking is however there's an easier solution in front of my eyes and I just don't see it. Looking at the code, it boils down to replicate int.from_bytes(_urandom(numbytes), 'big'). Will work on that if I don't get the easier, expected answer. — fabiog1901
– fabiog1901, Commented Jan 11, 2022 at 15:14
@AndrasDeak The ultimate goal is to use that subclass for creating faker.Faker objects. All functions in Faker use the underlying Random generator, specifically the functions random() and getrandbits(). Class random.Random() however cannot be parallelized and the BitGenerator cannot be advanced (PCG64 can). — fabiog1901
– fabiog1901, Commented Jan 11, 2022 at 15:18
@jakub This could be the gist of the solution, unless there's a simpler, higher level function. int.from_bytes(np.random.default_rng().bytes(numbytes), 'big'). Thanks! — fabiog1901
– fabiog1901, Commented Jan 11, 2022 at 15:28

fabiog1901 · Accepted Answer · 2022-01-11 19:43:04Z

1

As @jakub mentioned in the comments, easiest way is to port the source code to NumPy.

    def getrandbits(self, k):
        print()
        """getrandbits(k) -> x.  Generates an int with k random bits."""
        if k < 0:
            raise ValueError('number of bits must be non-negative')
        numbytes = (k + 7) // 8                       # bits / 8 and rounded up
        x = int.from_bytes(self.rng.bytes(numbytes), 'big')
        return x >> (numbytes * 8 - k)                # trim excess bits

where self.rng is an instance of np.random.Generator.

answered Jan 11, 2022 at 19:43

fabiog1901

3744 silver badges13 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Sam Mason Over a year ago

you could also do something similar with a raw BitGenerator as it returns a uint64 ndarray, e.g. int.from_bytes(bitgen.random_raw(num_u64s, 'big'))

Collectives™ on Stack Overflow

NumPy equivalent to random.getrandbits()

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related