Pytorch对tensor操作的常用方法汇总

Tensor(张量)

什么是tensor?

• scalar(标量)：一个数值
• vector(向量)：一维数组
• matrix(矩阵)：二维数组
• tensor(张量)：大于二维的数组，即多维数组

Tensor的类型：一共9种类型

• numpy.float64
• numpy.float32
• numpy.float16
• numpy.int64
• numpy.int32
• numpy.int16
• numpy.int8
• numpy.uint9(无符号整型)
• numpy.bool

1.torch.ones：默认值为1

2.torch.zeros：默认值为0

import torch
y = torch.zeros(3,4)
print(y)
    //控制台结果
    tensor([[0., 0., 0., 0.],
           [0., 0., 0., 0.],
           [0., 0., 0., 0.]])

3.torch.full：填充

4.torch.empty：为空

5.torch.rand：随机数，均匀分布在0和1之间

import torch
x = torch.rand(10)
print(x)
    //控制台输出
    tensor([0.6338, 0.8150, 0.3313, 0.2341, 0.9809, 0.2198, 0.9884, 0.2101, 0.4671, 0.6889])

6.torch.randn：正态分布，均值为0，方差为1

import torch
x = torch.randn(10)
print(x)
    //控制台输出
    tensor([ 0.1908, -0.0746, -0.3406, -1.4599,  0.7656,  1.2863, -0.3496,  1.5926, 0.7925,  3.1737])

7.torch.randint：创建随机整数

import torch
x = torch.randint(1,99,(3,3))
print(x)
    //控制台输出
    tensor([[35, 56, 53],
           [22, 67, 85],
           [31, 28, 58]])

8.torch.randperm：选择随机数，下面的例子为输出0到9的10个随机数

import torch
x = torch.randperm(10)
print(x)
    //控制台输出
    tensor([9, 8, 2, 1, 4, 3, 0, 6, 5, 7])

9.torch.argmin()：特别的在dim=0表示二维中的列，dim=1在二维矩阵中表示行

import torch
x = torch.randint(1,99,(3,3))
print(x)
print(torch.argmin(x, dim=0))
    //控制台结果
    tensor([[32, 75, 31],
           [ 1, 36, 74],
           [92,  9, 15]])
    tensor([1, 2, 2])

10.torch.argmax()：

import torch
x = torch.randint(1,99,(3,3))
print(x)
print(torch.argmax(x, dim=1))
    //控制台结果
    tensor([[40,  7, 31],
           [60, 12, 85],
           [71, 51, 89]])
    tensor([0, 2, 2])

11.torch.ones_like：属性相似

12.torch.randn_like：属性相似

13.torch.add：加法

14.torch.tensor：创建

15.torch.as_tensor：创建

16.torch.from_numpy：从numpy转为tensor

import numpy as np
import torch
x = np.array([3,4,5,6,7])
print(x)
y = torch.from_numpy(x)
print(y)
    //控制台结果
    array([3 4 5 6 7])
    tensor([3, 4, 5, 6, 7], dtype=torch.int32)

17.torch.reshape：改变shape

18.torch.numel：统计张量里有几个元素

import torch
x = torch.rand(2,3)
print(x)
print(torch.numel(x))
    //控制台结果
    tensor([[0.2432, 0.2156, 0.5053],
           [0.7344, 0.4466, 0.5718]])
    6   

19.torch.view：视图

20.torch.arange：生成一个区间的数，下面例题是生成从10开始到30结束之间步长为4的数，左闭右开

import torch
x = torch.arange(10,30,4)
print(x)
    //控制台输出
    tensor([10, 14, 18, 22, 26])

21.torch.linspace：切分，把2到10切分成5等分

import torch
x = torch.linspace(2,10,5)
print(x)
    //控制台结果
    tensor([ 2.,  4.,  6.,  8., 10.])

22.torch.eye：对角线为1的tensor

import torch
x = torch.eye(3,3)
print(x)
    //控制台结果
    tensor([[1., 0., 0.],
           [0., 1., 0.],
           [0., 0., 1.]])

23.torch.cat：连接，把两个张量堆叠起来,dim=0为纵轴堆叠，dim=1为横轴堆叠

import torch
x = torch.randint(1,10,(2,3))
print(x)
y = torch.cat((x,x)，dim=0)
print(y)
    //控制台结果
    tensor([[3, 9, 9],
           [7, 9, 6]])
    tensor([[3, 9, 9],
           [7, 9, 6],
           [3, 9, 9],
           [7, 9, 6]])

import torch
x = torch.randint(1,10,(2,3))
print(x)
y = torch.cat((x,x),dim=1)
print(y)
    //控制台结果
    tensor([[9, 9, 2],
           [5, 2, 8]])
    tensor([[9, 9, 2, 9, 9, 2],
           [5, 2, 8, 5, 2, 8]])

 24.torch.index_select：根据索引选择，.index_select中的参数0为横向选，1为纵向选

import torch
x = torch.randint(1,10,(4,4))
print(x)
indices = torch.tensor([0,2])
y = torch.index_select(x,0,indices)
print(y)
    //控制台结果
    tensor([[8, 1, 7, 8],
           [3, 1, 6, 1],
           [9, 2, 7, 9],
           [2, 1, 7, 8]])
    tensor([[8, 1, 7, 8],
           [9, 2, 7, 9]])

25.torch.narrow：缩小

26.torch.t / torch.transpose：转置， torch.transpose中0表示行，1表示列，下面例子为行列交换

import torch
x = torch.tensor([[1,2,3],[4,5,6]])
y1 = x.t()
y2 = x.transpose(1,0)
print(y1)
print(y2)
    //控制台结果
    tensor([[1, 2, 3],
           [4, 5, 6]])
    tensor([[1, 4],
           [2, 5],
           [3, 6]])
    tensor([[1, 4],
           [2, 5],
           [3, 6]])

27.torch.take：根据索引获取元素

28.torch.split：分割

import torch
x = torch.tensor([1,2,3,4,5,6,7])
y = x.split(2)
print(y)
    //控制台结果
    (tensor([1, 2]), tensor([3, 4]), tensor([5, 6]), tensor([7]))

29.torch.is_tensor(x)：检测x是否为张量

import torch
x = [2,8,3,3,2,1,2]
y = torch.rand(1,2)
print(torch.is_tensor(x))
print(torch.is_tensor(y))
    //控制台结果
    False
    True

30.torch.chunk：把张量切块，dim=0为横向切，dim=1为纵向切

import torch
x = torch.randint(1,10,(3,3))
y = torch.chunk(x,2,dim=0)
print(x)
print(y)
    //控制台结果
    tensor([[4, 1, 9],
           [6, 2, 4],
           [7, 7, 3]])
    (tensor([[4, 1, 9],
            [6, 2, 4]]), tensor([[7, 7, 3]]))

import torch
x = torch.randint(1,10,(3,3))
y = torch.chunk(x,2,dim=1)
print(x)
print(y)
    //控制台结果
    tensor([[9, 1, 4],
           [1, 2, 5],
           [3, 9, 8]])
    (tensor([[9, 1],
            [1, 2],
            [3, 9]]), tensor([[4],
                              [5],
                              [8]]))