Pytorch---Pytorch的张量基础--1

一、张量的初始化

1.从Python列表初始化张量

代码

import torch
import numpy as np

if __name__ == '__main__':
    t = torch.tensor([1, 2, 3])
    print(t)
    print(t.dtype)

结果

2.默认构造常用的数据类型的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t = torch.FloatTensor([1, 2, 3])  # 默认类型是float32
    print(t)
    print(t.dtype)
    t = torch.LongTensor([1, 2, 3])  # 默认类型是int64
    print(t)
    print(t.dtype)

结果

3.从numpy数组创建Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    np_array = np.arange(12).reshape(3, 4)
    print(np_array)
    print(np_array.dtype)
    t = torch.from_numpy(np_array)  # 直接继承numpy数组的数据类型
    print(t)
    print(t.dtype)

结果

二、张量的基本类型

• 32位浮点型：torch.float32 / torch.float
• 64位浮点型：torch.float64
• 64位整型：torch.int64 / torch.long
• 16位整型：torch.int16
• 32位整型：torch.int32

三、创建指定类型的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t = torch.tensor([1, 2, 3], dtype=torch.float)  # 创建32位浮点型Tensor
    print(t)
    print(t.dtype)
    t = torch.tensor([1, 2, 3], dtype=torch.long)  # 创建64位整型Tensor
    print(t)
    print(t.dtype)

结果

四、随机初始化Tensor

1.创建0~1之间随机值的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t = torch.rand(2, 3)
    print(t)
    print(t.dtype)
    print(t.shape)

结果

2.创建标准正态分布随机值的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t = torch.randn(2, 3)
    print(t)
    print(t.dtype)
    print(t.shape)

结果

3.创建全0的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t = torch.zeros(3, 3)
    print(t)
    print(t.dtype)
    print(t.shape)

结果

4.创建全1的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t = torch.ones(3, 3)
    print(t)
    print(t.dtype)
    print(t.shape)

结果

5.创建与原Tensor形状类型都一样的0~1之间的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t1 = torch.zeros(3, 3)
    print(t1)
    print(t1.dtype)
    print(t1.shape)
    print('-----------------------------------')
    t = torch.rand_like(t1)
    print(t)
    print(t.dtype)
    print(t.shape)

结果

6.创建与原Tensor形状类型都一样的标准正态分布的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t1 = torch.zeros(3, 3)
    print(t1)
    print(t1.dtype)
    print(t1.shape)
    print('-----------------------------------')
    t = torch.randn_like(t1)
    print(t)
    print(t.dtype)
    print(t.shape)

结果

7.创建与原Tensor形状类型都一样的全0的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t1 = torch.rand(3, 3)
    print(t1)
    print(t1.dtype)
    print(t1.shape)
    print('-----------------------------------')
    t = torch.zeros_like(t1)
    print(t)
    print(t.dtype)
    print(t.shape)

结果

8.创建与原Tensor形状类型都一样的全1的Tensor

代码

import torch
import numpy as np

if __name__ == '__main__':
    t1 = torch.rand(3, 3)
    print(t1)
    print(t1.dtype)
    print(t1.shape)
    print('-----------------------------------')
    t = torch.ones_like(t1)
    print(t)
    print(t.dtype)
    print(t.shape)

结果

五、张量的属性

1.返回Tensor的形状

代码

import torch
import numpy as np

if __name__ == '__main__':
    t1 = torch.rand(3, 3)
    print(t1.shape)
    print(t1.size())
    print(t1.size(0))
    print(t1.size(1))

结果

2.返回Tensor的数据类型

代码

import torch
import numpy as np

if __name__ == '__main__':
    t1 = torch.rand(3, 3)
    print(t1.dtype)

结果

3.移动Tensor至显卡设备

代码

import torch
import numpy as np

if __name__ == '__main__':
    print(torch.cuda.is_available())
    t1 = torch.rand(3, 3)
    print(t1.device)
    if torch.cuda.is_available():
        t1 = t1.to('cuda')
        print(t1.device)

结果