import导入

import torch#基本的torch函数 import torch.autograd as autograd#自动求导 import torch.nn as nn#神经网络类都在这个里面 import torch.nn.functional as F#几乎所有的激励函数 import torch.optim as optim#优化

创建Tensors

#create 1D vector V_data = [1., 2., 3.] V = torch.Tensor(V_data)#我用的是pyCharm编辑器，输入torch给的提示没有Tensor函数，其实是有的 print(V)
1 2 3 [torch.FloatTensor of size 3]
#create 2D vector M_data = [[1., 2., 3.], [4., 5., 6.]] M = torch.Tensor(M_data) print(M)
1  2  3 4  5  6 [torch.FloatTensor of size 2x3]
#create 3D vector T_data = [[[1.,2.], [3.,4.]],          [[5.,6.], [7.,8.]]] T = torch.Tensor(T_data) print(T)
(0 ,.,.) =  1  2  3  4 (1 ,.,.) =  5  6  7  8 [torch.FloatTensor of size 2x2x2]

获取Tensor部分值

#我就觉得这里比TensorFlow好用多了QAQ print(V[0]) print(M[0]) print(T[0])
1.0 1 2 3 [torch.FloatTensor of size 3] 1  2 3  4 [torch.FloatTensor of size 2x2]

产生随机数据

x = torch.randn((3,4,5)) print(x)
(0 ,.,.) =  1.4533  0.0593  0.2027 -1.0107 -0.3175 -0.1847  0.3021  0.0848 -1.2445 -0.5568 -0.2796 -0.5961 -0.3000 -0.2782  1.4920  1.4030  1.0875 -0.5814 -1.2006  0.2690 (1 ,.,.) = -0.7093 -0.4939  0.7491  0.9133  0.4221  1.3949  2.5685 -0.4359 -0.7788  1.0251  1.6232 -1.2432  0.3403 -1.0551  1.3790 -1.5632 -0.9772  0.3963 -0.1890  0.0032 (2 ,.,.) = -0.3360 -0.5571 -0.6641 -1.5845 -0.8766 -0.1809 -1.0035  1.7093  0.9176  1.6438  1.6955  0.6816  0.5978 -0.1379 -0.3877  1.0876  1.2371 -0.7378 -0.7647  0.0544 [torch.FloatTensor of size 3x4x5]

Tensor运算

x = torch.Tensor([1., 2., 3.]) y = torch.Tensor([4., 5., 6.]) z = x + y print(z)
5 7 9 [torch.FloatTensor of size 3]
[res] torch.cat( [res,] x_1, x_2, [dimension] )
x_1 = torch.randn(2, 5) y_1 = torch.randn(3, 5) z_1 =torch.cat([x_1, y_1])#没有最后一个参数，默认是0，则最终维度的第0维度为x_1与y_1第0维度的和，最终维度的其他维度不变.以下同理 print(z_1) x_2 = torch.randn(2, 3) y_2 = torch.randn(2, 5) z_2 = torch.cat([x_2, y_2], 1) print(z_2)
0.6372  0.7380  0.9324  0.0626 -0.3678 1.1819  2.1591  0.2445  0.0064  0.7760 -0.7765 -0.6797  0.1814  0.3948  1.7398 -0.2957 -0.6972  3.7052 -0.1943  0.4159 0.7385 -0.2365  1.4243 -0.0044 -0.7645 [torch.FloatTensor of size 5x5] -0.0256 -0.6597 -0.1897  0.4361  0.1680  0.6513 -0.0433 -1.5741 -1.4514  0.0949 -0.7783  0.8568 -0.8722  0.0364 -0.0998  0.9265 [torch.FloatTensor of size 2x8]

Tensor维度变型reshaping

x = torch.randn(2, 3, 4) print(x)
(0 ,.,.) =  0.6294 -0.3965  1.3737  1.6951 -0.5477 -1.5385 -0.0288  0.8104 -0.4208 -0.4469  0.0184  0.9507 (1 ,.,.) = -0.2843 -0.0695 -0.1747  2.3774  1.1067  0.1980 -2.0712 -0.0670 -1.4900  0.0716 -0.7605  0.4611 [torch.FloatTensor of size 2x3x4]``` view转换维数
print(x.view(2,12))#将234 -> 2*12
Columns 0 to 9

0.6294 -0.3965 1.3737 1.6951 -0.5477 -1.5385 -0.0288 0.8104 -0.4208 -0.4469

-0.2843 -0.0695 -0.1747 2.3774 1.1067 0.1980 -2.0712 -0.0670 -1.4900 0.0716
Columns 10 to 11

0.0184 0.9507

-0.7605 0.4611

[torch.FloatTensor of size 2x12]```
print(x.view(2,-1))#-1的话，意味着最后的相乘为维数，这里为2*之后的成绩
#和上面的一样 Columns 0 to 9 0.6294 -0.3965  1.3737  1.6951 -0.5477 -1.5385 -0.0288  0.8104 -0.4208 -0.4469 -0.2843 -0.0695 -0.1747  2.3774  1.1067  0.1980 -2.0712 -0.0670 -1.4900  0.0716 Columns 10 to 11 0.0184  0.9507 -0.7605  0.4611 [torch.FloatTensor of size 2x12]

Computation Graphs and Automatic Differentiation

x = autograd.Variable(torch.Tensor([1., 2., 3]), requires_grad=True) print(x) print(x.data)#.data显示具体数据
#找不同 Variable containing: 1 2 3 [torch.FloatTensor of size 3] 1 2 3 [torch.FloatTensor of size 3]
y = autograd.Variable( torch.Tensor([4., 5., 6]), requires_grad=True ) z = x + y print(z.data)
5 7 9 [torch.FloatTensor of size 3]
.creator是生成器
print(z.creator)
#因为是z = x + y 所以，z运算是add <torch.autograd._functions.basic_ops.Add object at 0x10b04f128>
显示z中所有元素的和
s = z.sum() print(s)
Variable containing: 21 [torch.FloatTensor of size 1]
s.backward()#反向传播 print(x.grad)#对x求导
Variable containing: 1 1 1 [torch.FloatTensor of size 3] #答案解释 #x = [1,2,3] #y = [4,5,6] #z = x + y = [x0+y0, x1+y1, x2+y2] #s = z.sum() = x0+y0+x1+y1+x2+y2 #x.grad 在s运算中对x求导   也就是当中的x0,x1,x2求导  为1，1，1

Deep Learning Building Blocks: Affine maps, non-linearities and objectives

Affine maps

也可以说是线性映射，即为f(x) = Ax + b

nn.Linear(inputSize,outputSize,bias=True)

输入(N, inputSize)

输出(N, outputSize)
lin = nn.Linear(5,3) data = autograd.Variable(torch.randn(2, 5)) print(lin(data))
Variable containing: -0.1838 -0.1833 -0.6425 0.2675  0.0263  0.0482 [torch.FloatTensor of size 2x3]
Non-Linearities

非线性，常用的函数有 tanh(x),σ(x),ReLU(x) 这些都是激励函数

在pytorch中大部分激励函数在torch.functional中
data = autograd.Variable( torch.randn(2, 2) ) print(data) print (F.relu(data))#relu函数是小于零是0，大于零就是它本身
Variable containing: -2.0620  1.4252 0.5694  0.2251 [torch.FloatTensor of size 2x2] Variable containing: 0.0000  1.4252 0.5694  0.2251 [torch.FloatTensor of size 2x2]
Softmax and Probabilities

softmax是x_i/sum(x)
data = autograd.Variable( torch.randn(5) ) print(data) print(F.softmax(data)) print(F.softmax(data).sum()) print(F.log_softmax(data))
Variable containing: 0.6861 0.1695 -0.4775 -2.0097 0.7039 [torch.FloatTensor of size 5] Variable containing: 0.3340 0.1992 0.1043 0.0225 0.3400 [torch.FloatTensor of size 5] Variable containing: 1 [torch.FloatTensor of size 1] Variable containing: -1.0967 -1.6133 -2.2604 -3.7925 -1.0789 [torch.FloatTensor of size 5]
点击“阅读原文”
继续阅读
阅读原文