Seed setup for random number generator to ensure reproducibility

1

According to the reproducibility session of Pytorch Doc, we can set seed for random number generator in Pytorch, python, and numpy-related libraries respectively.

import torch
torch.manual_seed(0)

import random
random.seed(0)

import numpy as np
np.random.seed(0)

For a generic ML workflow, involving numpy-based libraries, like scikit-learn; and pytorch, etc. The best practice to ensure reproducibility is to set the seed using all the above three ways. Is that right?

2

Yes, that’s right, but the linked docs also mention more needed steps to ensure reproducibility between runs, so you would also need to check these points.