Hi !
To clear things up
If you want to do the equivalent of numpy.random.choice:
a = np.array([1, 2, 3, 4])
p = np.array([0.1, 0.1, 0.1, 0.7])
n = 2
replace = True
b = np.random.choice(a, p=p, size=n, replace=replace)
In pytorch you can use torch.multinomial :
a = torch.tensor([1, 2, 3, 4])
p = torch.tensor([0.1, 0.1, 0.1, 0.7])
n = 2
replace = True
idx = p.multinomial(num_samples=n, replacement=replace)
b = a[idx]
Careful, np.random.choice defaults to replace=True
But torch.multinomial defaults to replacement=False