seed¶

delu.random.seed(base_seed: int | None, /, *, one_cuda_seed: bool = False) → int[source]¶

Set diverse global random seeds in random, numpy and torch.

Note

For all libraries, different deterministically computed (based on the base_seed argument) seeds are set to ensure that different libraries and (by default) devices generate diverse random numbers.

Usage

>>> import random
>>> import numpy as np
>>>
>>> def f():
...     return (
...         random.randint(0, 10 ** 9),
...         np.random.rand(10).tolist(),
...         torch.randn(20).tolist(),
...     )
...
>>> # Numbers sampled under the same random seed are equal.
>>> delu.random.seed(0)
>>> a = f()
>>> delu.random.seed(0)
>>> b = f()
>>> a == b
True

Pass None to set a truly random seed generated by the OS:

>>> # Save the generated `seed` for future reproducibility:
>>> seed = delu.random.seed(None)
>>> a = f()
>>> # Reproduce the results:
>>> delu.random.seed(seed)
>>> b = f()
>>> a == b
True

Parameters:

base_seed – an integer from [0, 2**64) used to compute diverse seeds for all libraries. If None, then an unpredictable seed generated by OS is used and returned.
one_cuda_seed – if True, then the same seed will be set for all CUDA devices, otherwise, different seeds will be set for all CUDA devices.

Returns:

the provided base_seed or the generated one if base_seed=None.