How to use “retain_variables” in Variable backward function.
I tried the following code:
import torch
from torch.autograd import Variable
x = Variable(torch.ones(2, 2), requires_grad = True)
y = x + 2
y.backward(torch.ones(2, 2), retain_variables=True )
print "first gradient of x is:"
print x.grad
z = y * y
gradient = torch.ones(2, 2)
z.backward(gradient)
print "second gradient of x is:"
print x.grad
import torch
from torch.autograd import Variable
x = Variable(torch.ones(2, 2), requires_grad = True)
y = x + 2
y.backward(torch.ones(2, 2), retain_variables=False)
print "first gradient of x is:"
print x.grad
z = y * y
gradient = torch.ones(2, 2)
z.backward(gradient)
print "second gradient of x is:"
print x.grad
Both print the same results:
first gradient of x is:
Variable containing:
1 1
1 1
[torch.FloatTensor of size 2x2]
second gradient of x is:
Variable containing:
7 7
7 7
[torch.FloatTensor of size 2x2]