202_PyTorch中文教程：變量-Variable

202_PyTorch中文教程：變量-Variable

torch中的變量（Variable）用於構建計算圖（computational graph），可是相對Tensorflow和Theano的靜態圖不一樣的是，Torch的graph是動態的。 所以，torch中不須要佔位符（placeholder），能夠直接傳遞變量（variable）到計算圖中。

• 更多參考: https://morvanzhou.github.io/tutorials/
• 油管頻道: https://www.youtube.com/user/MorvanZhou

依賴軟件包

• torch

import torch
from torch.autograd import Variable

tensor = torch.FloatTensor([[1,2],[3,4]])            # build a tensor
variable = Variable(tensor, requires_grad=True)      # build a variable, usually for compute gradients

print(tensor)       # [torch.FloatTensor of size 2x2]
print(variable)     # [torch.FloatTensor of size 2x2]

tensor([[1., 2.],
        [3., 4.]])
tensor([[1., 2.],
        [3., 4.]], requires_grad=True)

到此，tensor 和 variable 看起來同樣。

可是, tvariable是graph的一部分, 而不是auto-gradient的一部分。

t_out = torch.mean(tensor*tensor)       # x^2
v_out = torch.mean(variable*variable)   # x^2
print(t_out)
print(v_out)

7.5
Variable containing:
 7.5000
[torch.FloatTensor of size 1]

v_out.backward()    # backpropagation from v_out

$$ v_{out} = {{1} \over {4}} sum(variable^2) $$

the gradients w.r.t the variable,

$$ {d(v_{out}) \over d(variable)} = {{1} \over {4}} 2 variable = {variable \over 2}$$

let's check the result pytorch calculated for us below:

variable.grad

Variable containing:
 0.5000  1.0000
 1.5000  2.0000
[torch.FloatTensor of size 2x2]

variable # this is data in variable format

Variable containing:
 1  2
 3  4
[torch.FloatTensor of size 2x2]

variable.data # this is data in tensor format

1  2
 3  4
[torch.FloatTensor of size 2x2]

variable.data.numpy() # numpy format

array([[ 1.,  2.],
       [ 3.,  4.]], dtype=float32)

Note that we did .backward() on v_out but variable has been assigned new values on it's grad.

As this line

v_out = torch.mean(variable*variable)

will make a new variable v_out and connect it with variable in computation graph.

type(v_out)

torch.autograd.variable.Variable

type(v_out.data)

torch.FloatTensor