Here is a demo:
x = torch.ones(10).geometric_(.03)
print(x)
geo = Geometric(.03)
print(geo.sample([10]))
Result:
>>> x = torch.ones(10).geometric_(.03)
>>> print(x)
tensor([1., 1., 1., 1., 1., 1., 1., 1., 1., 1.])
>>> geo = Geometric(.03)
>>> print(geo.sample([10]))
tensor([ 2., 51., 30., 9., 10., 58., 14., 26., 108., 75.])
Not a huge problem, but it’s counter-intuitive and inconsistent.