pytorch tensor相关知识点

创建 tensor

• 从list 或者 numpy array 创建 tensor

import torch
import numpy as np
x = torch.tensor([[-1,1],[1, -1]])
x = torch.from_numpy(np.array([[1, -1], [-1, 1]]))

• 全 0 tensor (zero tensor)

x = torch.zeros([2, 2])

• 全 1 tensor (unit tensor)

x = torch.ones([2, 2])

tensor 的常见运算

• squeeze: 移除长度为1的指定维度

x = torch.zeros([1, 2, 3])
print('before',x.shape)
x = x.squeeze(0)
print('after',x.shape)

'''输出：
before torch.Size([1, 2, 3])
after torch.Size([2, 3])
'''

• unsqueeze:拓展一个新的维度，在一个指定维度前面添加一个长度为1的新维度

x = torch.zeros([2, 3])
print('before',x.shape)
x = x.unsqueeze(1)
print('after',x.shape)

```python
'''输出：
before torch.Size([2, 3])
after torch.Size([2, 1, 3])
'''

• Transpose:转置两个指定维度

x = torch.zeros([2, 3])
print('before',x.shape)
x = x.transpose(0, 1)
print('after',x.shape)

```python
'''输出：
before torch.Size([2, 3])
after torch.Size([3, 2])
'''

• cat:连接多个张量

x = torch.zeros([2, 1, 3])
y = torch.zeros([2, 3, 3])
z = torch.zeros([2, 2, 3])
w = torch.cat([x, y, z], dim = 1)
w.shape

'''输出：
torch.Size([2, 6, 3])
'''

• sum加在一起

x = torch.ones([2, 3])
y = x.sum()
y

'''输出：
tensor(6.)
'''

Tensor vs Numpy

属性

Pytorch	Numpy
x.shape	x.shape
x.dtype	x.dtype

• tensor.numel() 查看tensor内元素个数
• trensor.nelment() 查看tensor内元素个数

改变形状

Pytorch	NumPy
x.reshape / x.view	x.reshape
x.squeeze()	x.squeeze()
x.unsqueeze(1)	np.expand_dims(x, 1)

设备

default 情况下，tensor 和 modules 会在CPU上面计算

x = x.to('cpu')
x = x.to('cuda')