Pytorch/Numpy之squeeze

目录

对于Pytorch而言

对于Numpy而言

代码

输出

---

对于Pytorch而言

squeeze(torch.tensor,axis)  返回值与原tensor共享内存，修改返回值中元素值对原tensor也有影响

• 未指定axis，将张量size()中所有为1的维度去除
• 指定axis，如果该axis为1则去除，反之保留原来的size不变

对于Numpy而言

squeeze(np.ndarray,axis)  返回值与原ndarray共享内存，修改返回值中元素值对原ndarray也有影响

• 未指定axis，将ndarray shape中所有为1的维度去除
• 指定axis，如果该axis为1则去除，反之则报错

 

代码

t = torch.rand(1,2,1,3)
print(t.size(),'\n')

t1 = torch.squeeze(t)
print(t1.size(),'\n')

t2 = torch.squeeze(t,0)
print(t2.size(),'\n')

t3 = torch.squeeze(t,1)
print(t3.size(),'\n')

n = np.zeros((1,2,1,3))
print(n.shape,'\n')

n1 = np.squeeze(n)
print(n1.shape,'\n')

n2 = np.squeeze(n,0)
print(n2.shape,'\n')

n3 = np.squeeze(n,1)
print(n3.shape,'\n')

 

输出

torch.Size([1, 2, 1, 3]) 

torch.Size([2, 3]) 

torch.Size([2, 1, 3]) 

torch.Size([1, 2, 1, 3]) 

 

(1, 2, 1, 3) 

(2, 3) 

(2, 1, 3) 

    return squeeze(axis=axis)
ValueError: cannot select an axis to squeeze out which has size not equal to one