1 import torch
 2 from torch.autograd import Variable
 3 
 4 # Variable in torch is to build a computational graph,
 5 # but this graph is dynamic compared with a static graph in Tensorflow or Theano.
 6 # So torch does not have placeholder, torch can just pass variable to the computational graph.
 7 
 8 tensor = torch.FloatTensor([[1,2],[3,4]])            # build a tensor
 9 variable = Variable(tensor, requires_grad=True)      # build a variable, usually for compute gradients
10 
11 print(tensor)       # [torch.FloatTensor of size 2x2]
12 print(variable)     # [torch.FloatTensor of size 2x2]
13 
14 # till now the tensor and variable seem the same.
15 # However, the variable is a part of the graph, it's a part of the auto-gradient.
16 
17 t_out = torch.mean(tensor*tensor)       # x^2
18 v_out = torch.mean(variable*variable)   # x^2
19 print(t_out)
20 print(v_out)    # 7.5
21 
22 v_out.backward()    # backpropagation from v_out
23 # v_out = 1/4 * sum(variable*variable)
24 # the gradients w.r.t the variable, d(v_out)/d(variable) = 1/4*2*variable = variable/2
25 print(variable.grad)
26 '''
27  0.5000  1.0000
28  1.5000  2.0000
29 '''
30 
31 print(variable)     # this is data in variable format
32 """
33 Variable containing:
34  1  2
35  3  4
36 [torch.FloatTensor of size 2x2]
37 """
38 
39 print(variable.data)    # this is data in tensor format
40 """
41  1  2
42  3  4
43 [torch.FloatTensor of size 2x2]
44 """
45 
46 print(variable.data.numpy())    # numpy format
47 """
48 [[ 1.  2.]
49  [ 3.  4.]]
50 """