

In [1]:

    
import numpy as np



In [2]:

    
import torch
from torch.autograd import Variable

PyTorch Autograd Introduction

Example from official documentation



In [3]:

    
x = Variable(torch.ones(2,2), requires_grad=True) # requires_grad: calculate gradients
print(x)









    



Variable containing:
 1  1
 1  1
[torch.FloatTensor of size 2x2]



In [4]:

    
print(x.data)









    



 1  1
 1  1
[torch.FloatTensor of size 2x2]



In [5]:

    
print(x.grad)









    



None



In [6]:

    
y = x + 2
print(y)









    



Variable containing:
 3  3
 3  3
[torch.FloatTensor of size 2x2]



In [7]:

    
z = y * y * 3
out = z.sum()
print(z, out)









    



Variable containing:
 27  27
 27  27
[torch.FloatTensor of size 2x2]
 Variable containing:
 108
[torch.FloatTensor of size 1]



In [8]:

    
out.backward() # backpropagation
print(x.grad)









    



Variable containing:
 18  18
 18  18
[torch.FloatTensor of size 2x2]

$z = (x + 2)^2 * 3$

$out = 3(x+2)^2$

$dout/dx = 6(x+2)' * 1 = 6(x+2)$

$= 6*3 = 18$

Tensors and Variables have the same API. Variables have the Autograd option, in that they keep track of how they were created.

If you want to use Autograd you have to use Variables.



In [9]:

    
Q = torch.eye(3)



In [10]:

    
Q









    Out[10]:





 1  0  0
 0  1  0
 0  0  1
[torch.FloatTensor of size 3x3]



In [ ]: