Pytorch基础（一）：Tensor_tensor mask

1.初始化

1.1 pytorch
① list->tensor

data = [[1,2],[3,4]]
x_data = torch.tensor(data)
1
2

② Numpy array->tensor

np_array = np.array(data)
x_np = torch.from_numpy(np_array)
1
2

③ 根据已有张量初始化

x_ones = torch.ones_like(x_data)
x_rand = torch.rand_like(x_data,dtype = torch.float)
1
2

④ 创建指定大小的tensor（开辟新的内存）

shape = (2,3)
x_ones = torch.ones(shape)
x_zeros = torch.zeros(shape)
x_rand = torch.rand(shape)
1
2
3
4

1.2 torchlib
① 数组->Tensor

int data[10] = {3,4,6}
torch::Tensor x_data = torch::from_blob(data,{3},torch::kFloat)
1
2

② 标准库的vector->Tensor：

std::vector<float> std_vector = {3，4，6};
torch::Tensor vector_data = torch::from_brob(std_vector.data(),{3},torch::kFloat);
1
2

③ 根据已有张量初始化

torch::Tensor x = torch::zeros({3,4});
torch::Tensor x_zeros = torch::zeros_like(x);
torch::Tensor x_ones = torch::ones_like(x);
torch::Tensor x_rand = torch::rand_like(x);
//浅拷贝
torch::Tensor y = x
//深拷贝
torch::Tensor z = x.clone();
1
2
3
4
5
6
7
8

④ 创建指定大小的Tensor（开辟新的内存）

torch::Tensor x_ones = torch::ones({3,4});
torch::Tensor x_zeros = torch::zeros({3,4});
torch::Tensor x_eye = torch::eye(4);
torch::Tensor x_full = torch::full({3,4},10);
torch::Tensor x_rand = torch::rand({3,4});
torch::Tensor x_randn = torch::randn({3,4});
torch::Tensor x_randint = torch::randint(0,4,{3,3});
1
2
3
4
5
6
7

2.张量操作

2.1 pytorch
① 索引+切片：(共用一块存储数据的内存)

tensor = torch.rand(4,4)
a = tensor[:,1]
1
2

② 提取指定元素组成新的张量

Tensor[Mask]#即可提取想要的元素
1

2.2 torchlib
① 索引

auto x = torch::rand({3,4});
y = x[1];//选择第0维的第1层
y = x[1][3]//选择第0维的第1层，第2维的第3层对应的元素
1
2
3

② 切片
当切取一层时：

auto x = torch::rand({3,4});
auto y = x.select(0,1)//选择第0维的第1层张量，即矩阵中的第二行
1
2

当连续切取几层时：

auto x = torch::rand({3,4});
auto y = x.narrow(0,2,2)//从第0维的第2层张量开始选两层张量
auto y = x.slice(0,1,3)//第0维中选择第1维到第3-1维的张量
1
2
3

注意：x,y的数值指向同一块内存空间

③ 提取指定元素形成新的张量（关键字index：就代表是提取出来相应的元素组成新的张量）

	std::cout<<b.index_select(0,torch::tensor({0, 3, 3})).sizes();//选择第0维的0，3，3组成新张量[3,3,28,28]
	std::cout<<b.index_select(1,torch::tensor({0,2})).sizes(); //选择第1维的第0和第2的组成新张量[10, 2, 28, 28]
	std::cout<<b.index_select(2,torch::arange(0,8)).sizes(); //选择十张图片每个通道的前8列的所有像素[10, 3, 8, 28]
1
2
3

    Tensor x_data = torch::rand({3,4});
    Tensor mask = torch::zeros({3,4});

    mask[1][1] = 1;
    mask[0][0] = 1;

    Tensor x = x_data.index({ mask.to(kBool) });//index()方法输入参量为布尔值组成的数组，输出参量为对应index的值组成新的张量(新的内存空间)
1
2
3
4
5
6
7

3.张量计算

3.1 pytorch
① 按元素的加减乘除正常使用±*/的运算符即可
② 矩阵乘法：例如xx^T

x = torch.rand((3,4))
y = x @ x.T
y = x.matmul(x.T)
1
2
3

3.2 torchlib
① 按元素的加减乘除正常使用±*/的运算符即可
② 矩阵乘法：例如xx^T

auto x = torch::rand({3,4});
x.mm(x.t());
1
2