Pytorch张量操作

1 #type check
2 a=torch.randn(2,3)       
3 print(a.type())                            #torch.FloatTensor
4 print(type(a))                             #<class ‘torch.Tensor‘>  这种情况比较少
5 print(isinstance(a, torch.FloatTensor))    #True

1 #标量 dimension 0/rank 0（常用于loss）
2 b=torch.tensor(1.3)
3 print(b)                 #tensor(1.3000)
4 print(b.shape)           #torch.Size([])  成员
5 print(len(b.shape))      #0
6 print(b.size())          #torch.Size([])  成员函数

 1 #张量 dimension 1（常用于bias）
 2 print(torch.tensor([1.1]))    #tensor([1.1000]) 指定具体数据，可以是N维
 3 print(torch.FloatTensor(1))   #tensor([9.6429e-39]) 指定具体维度，随机初始化
 4 data=np.ones(2)
 5 print(torch.from_numpy(data)) #tensor([1., 1.], dtype=torch.float64) 从numpy引入
 6 
 7 c=torch.ones(2)
 8 print(c.shape)                #torch.Size([2])
 9 
10 #张量 dimension 2
11 d=torch.randn(2,3)
12 print(d)
13 #tensor([[-1.8543, -0.7280,  0.6671],
14 #        [ 1.1492, -0.6379, -0.4835]])
15 print(d.shape)          #torch.Size([2, 3])
16 print(d.shape[0])       #2
17 print(d.size(1))        #3    
18 
19 #张量 dimension 3
20 f=torch.rand(1,2,3)
21 print(f)
22 #tensor([[[0.3690, 0.5702, 0.2382],
23 #         [0.3130, 0.5591, 0.3829]]])
24 print(f.shape)         #torch.Size([1, 2, 3])
25 print(f[0])            #取第一个维度
26 #tensor([[0.4535, 0.4307, 0.6469],
27 #        [0.1591, 0.0778, 0.4489]])
28 print(f[0][1])         #tensor([0.1591, 0.0778, 0.4489])

1 a=torch.rand(2,3,28,28)
2 print(a.numel())       #4704 全部元素个数
3 print(a.dim())         #4 维度数量 

 1 #方法一：import from numpy
 2 a=np.ones([2,3])
 3 print(torch.from_numpy(a))
 4 #tensor([[1., 1., 1.],
 5 #        [1., 1., 1.]], dtype=torch.float64)
 6 
 7 #方法二：import from List
 8 print(torch.tensor([[2.,3.2],[1.,22.3]]))
 9 #tensor([[ 2.0000,  3.2000],
10 #        [ 1.0000, 22.3000]])

1 #设置默认类型
2 print(torch.tensor([1.2,3]).type())               #torch.FloatTensor（不设定的话，默认值）
3 torch.set_default_tensor_type(torch.DoubleTensor)
4 print(torch.tensor([1.2,3]).type())               #torch.DoubleTensor

1 #rand [0,1]
2 a=torch.rand(2,2)
3 print(a)
4 print(torch.rand_like(a))         #根据a的形状生成，也可用dtype指定新类型
5 
6 #randint [min,max)不包含max
7 print(torch.randint(1,10,[3,3]))  #第1个参数是min，第二个参数是max，第三个参数是shape

1 #正态分布
2 print(torch.randn(3,3))         #N(0,1)
3 print(torch.normal(mean=torch.full([10],0),std=torch.arange(1,0,-0.1)))    #normal得到的维度为1
4 #tensor([ 0.1445, -0.5133, -0.5565,  0.0831,  0.1350,  0.1023, -0.6264, -0.1651, 0.2856,  0.0187])

 1 #ones全1，zeros全0，eye对角，ones_like
 2 print(torch.full([2,3],7))             #第一个参数是shape，第二个参数是value
 3 #tensor([[7., 7., 7.],
 4 #        [7., 7., 7.]])
 5 print(torch.full([], 7))               #tensor(7.) 生成标量
 6 
 7 #生成 [0,n-1]的等差数列
 8 print(torch.arange(0,10,2))            #tensor([0, 2, 4, 6, 8])
 9 
10 #[0,10]等间距切割成steps份
11 print(torch.linspace(0, 10, steps=6))  #tensor([ 0.,  2.,  4.,  6.,  8., 10.])
12 print(torch.logspace(0, 1, steps=5))   #tensor([ 1.0000,  1.7783,  3.1623,  5.6234, 10.0000])
13 
14 
15 #randperm
16 print(torch.randperm(10))             #tensor([8, 5, 2, 4, 7, 1, 3, 9, 6, 0])
17 
18 a=torch.rand(2, 3) 
19 idx=torch.randperm(2)
20 print(idx)                            #tensor([1, 0])
21 print(a[idx])
22 #tensor([[0.6159, 0.9362, 0.5502],
23 #        [0.9774, 0.6772, 0.2909]])

1 #切片与索引
2 a=torch.rand(4,3,28,28)
3 print(a[0].shape)               #torch.Size([3, 28, 28])
4 print(a[0,0].shape)             #torch.Size([28, 28])
5 print(a[0,0,2,4])               #tensor(0.7309)
6 
7 print(a[:2,:1].shape)           #torch.Size([2, 1, 28, 28])  等价于a[:2,:1,:,:].shape
8 print(a[:,:,::2,::2].shape)     #torch.Size([4, 3, 14, 14])

1 #select by specific index
2 print(a.index_select(2, torch.arange(8)).shape)  #torch.Size([4, 3, 8, 28])  第1个参数表示维度，第2个参数表示索引号（Tensor类型）
3 #...表示任意多的维度
4 print(a[0,...].shape)           #torch.Size([3, 28, 28])
5 print(a[:,1,...].shape)         #torch.Size([4, 28, 28]) 

1 #select by mask
2 x=torch.randn(3,4)
3 mask=x.ge(0.5)                  #greater equal大于等于
4 print(mask)
5 #tensor([[False,  True, False, False],
6 #        [ True, False, False,  True],
7 #        [ True, False,  True, False]])
8 torch.masked_select(x, mask)    #tensor([1.1361, 0.5963, 0.6286, 0.5664, 0.5152]) 维度为1

1 #select by flatten index
2 src=torch.tensor([[4, 3, 5],[6, 7, 8]])
3 print(torch.take(src,torch.tensor([0, 2, 5])))  #tensor([4, 5, 8])

1 a=torch.rand(4,1,28,28)
2 print(a.view(4,28*28).shape)        #torch.Size([4, 784])
3 print(a.reshape(4*28,28).shape)     #torch.Size([112, 28])

 1 a=torch.rand(4,1,28,28)
 2 print(a.unsqueeze(0).shape)         #torch.Size([1, 4, 1, 28, 28]
 3 
 4 b=torch.tensor([1.2, 2.3])          #torch.Size([2])
 5 print(b.unsqueeze(0))               #tensor([[1.2000, 2.3000]])   torch.Size([2, 1])
 6 print(b.unsqueeze(-1)) 
 7 #tensor([[1.2000],
 8 #        [2.3000]])
 9 
10 x=torch.rand(32)
11 y=torch.rand(4,32,14,14)
12 x=x.unsqueeze(1).unsqueeze(2).unsqueeze(0)      #[32]->[32,1]->[32,1,1]->[1,32,1,1]
13 print(x.shape)                                  #torch.Size([1, 32, 1, 1])) 再进行扩展即可计算x+y 

1 a=torch.rand(1,32,1,1)
2 print(a.squeeze().shape)             #torch.Size([32])  不指定维度就挤压全部
3 print(a.squeeze(0).shape)            #torch.Size([32, 1, 1])
4 print(a.squeeze(-1).shape)           #torch.Size([1, 32, 1])
5 print(a.squeeze(1).shape)            #torch.Size([1, 32, 1, 1])  size不等于1，删减不了

1 x=torch.rand(32)
2 y=torch.rand(4,32,14,14)
3 x=x.unsqueeze(1).unsqueeze(2).unsqueeze(0)      #[32]->[32,1]->[32,1,1]->[1,32,1,1]
4 print(x.shape)                                  # torch.Size([1, 32, 1, 1])
5 
6 print(x.expand(4,32,14,14).shape)               #torch.Size([4, 32, 14, 14])
7 print(x.expand(-1,32,3,-1).shape)               #torch.Size([1, 32, 3, 1])  -1指该维度不变
8 
9 print((x.expand(4,32,14,14)+y).shape)           #torch.Size([4, 32, 14, 14])

1 a=torch.rand(1,32,1,1)
2 print(a.repeat(4,32,1,1).shape)                 #torch.Size([4, 1024, 1, 1])
3 print(a.repeat(4,1,1,1).shape)                  #torch.Size([4, 32, 1, 1])

1 a=torch.rand(3, 4)
2 print(a.t().shape)                             #torch.Size([4, 3])

1 a=torch.rand(4,3,32,32)
2 b=a.transpose(1, 3).contiguous().view(4,3*32*32).view(4,32,32,3).transpose(1, 3)               
3 #[b,c,h,w]->[b,w,h,c]->[b,whc]->[b,w,h,c]->[b,c,h,w]  展开时按原来的顺序展开whc对应32,32,3
4 print(torch.all(torch.eq(a,b)))                       #tensor(True)

1 b=torch.rand(4,3,28,32)
2 print(b.transpose(1, 3).shape)                      #torch.Size([4, 32, 28, 3])
3 print(b.transpose(1, 3).transpose(1, 2).shape)      #torch.Size([4, 28, 32, 3])
4 
5 print(b.permute(0,2,3,1).shape)                     #torch.Size([4, 28, 32, 3]