PyTorch Tensor的基本操作

发布于 2024-04-27  55 次阅读


Please refresh the page if equations are not rendered correctly.
---------------------------------------------------------------

import torch
import numpy as np

torch.__version__
'2.2.1+cu121'

Create a torch tensor from Numpy

arr = np.array([1, 2, 3, 4, 5])
print(arr)
[1 2 3 4 5]
arr.dtype
dtype('int64')
type(arr)
numpy.ndarray
x = torch.from_numpy(arr)
print(x)
tensor([1, 2, 3, 4, 5])
x.dtype
torch.int64
type(x)
torch.Tensor
torch.as_tensor(arr)
tensor([1, 2, 3, 4, 5])

Note here any changes to arr will change the value of x

arr[0] = 99
x
tensor([99,  2,  3,  4,  5])

2D array

arr2d = np.arange(0.0, 12.0)
arr2d = arr2d.reshape(4, 3)
arr2d
array([[ 0.,  1.,  2.],
       [ 3.,  4.,  5.],
       [ 6.,  7.,  8.],
       [ 9., 10., 11.]])
x2 = torch.from_numpy(arr2d)
x2
tensor([[ 0.,  1.,  2.],
        [ 3.,  4.,  5.],
        [ 6.,  7.,  8.],
        [ 9., 10., 11.]], dtype=torch.float64)

torch.tensor(), torch.Tensor() and torch.FloatTensor()

my_arr = np.arange(10)
my_tensor = torch.tensor(my_arr)
my_tensor
tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
my_other_tensor = torch.from_numpy(my_arr)
my_other_tensor
tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
my_arr[0] = 999
print(my_tensor)
print(my_other_tensor)
tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
tensor([999,   1,   2,   3,   4,   5,   6,   7,   8,   9])
my_tensor[0] = 998
my_tensor
tensor([998,   1,   2,   3,   4,   5,   6,   7,   8,   9])
new_arr = np.array([1, 2, 3], dtype=np.int32)
print("Data type of new_arr: ", new_arr.dtype)

my_tensor = torch.tensor(new_arr)
print(f"Data type of my_tensor: {my_tensor.dtype}")

my_Tensor = torch.Tensor(new_arr)
print(f"Data type of my_Tensor: {my_Tensor.dtype}")

my_FloatTensor = torch.FloatTensor(new_arr)
print(f"Data type of my_FloatTensor: {my_FloatTensor.dtype}")
Data type of new_arr:  int32
Data type of my_tensor: torch.int32
Data type of my_Tensor: torch.float32
Data type of my_FloatTensor: torch.float32
Everything not saved will be lost.
最后更新于 2024-05-27