In place randomization

1

In torch we can do torch.zeros(3, 4):randn() for in place randomization of values. This doesn’t seem to work in pytorch. Is this (minor but convenient) feature available in pytorch? (There is in place torch.zeros(3, 4).random_() but that’s just random floats, it seems)

2

you can do:

a = torch.zeros(3, 4) # or whatever shape
torch.randn(3, 4, out=a)

But yea we didn’t implement some methods. We just didn’t get around to it.

3

:randn() just sampled values from a standard normal distribution, so torch.Tensor(3, 4).normal_() is equivalent.

4

FYI, this isn’t implemented yet:

>>> torch.cuda.FloatTensor(3,2).random_()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'torch.cuda.FloatTensor' object has no attribute 'random_'

But, @smth’s suggestion works greatly.