4 Methods to Create a PyTorch Tensor

In this tutorial, we will introdue how to create a pytorch tensor. There are some ways to create, we will introduce one by one.

Method 1: Use python data

We can use python list, tuple, float et al to create a pytorch tensor. Here is an example:

import torch
import numpy as np

# python data
v1 = torch.tensor(1)
print(v1)
v2 = torch.tensor([1, 2])
print(v2)
v2 = torch.tensor((2, 2))
print(v2)

Run this code, you will see:

tensor(1)
tensor([1, 2])
tensor([2, 2])

We also can use numpy ndarray to create a pytorch tensor. For example:

v2 = torch.tensor(np.array([1, 2]))
print(v2)

You will see:

tensor([1, 2])

You also can use torch.from_numpy() to convert a numpy data to pytorch tensor.

n = np.ones(5)
v3 = torch.from_numpy(n)
print(v3)

You will get a tensor:

tensor([1., 1., 1., 1., 1.], dtype=torch.float64)

Pytorch has some built-in methods, we can use them to create a new tensor. These methods are:

torch.ones()
torch.rand()
torch.zeros()

et al.

For example:

b = torch.rand((2, 3), dtype=torch.float64)
print(b, b.requires_grad)

We will get a new tensor:

tensor([[0.5199, 0.0872, 0.3165],
        [0.3257, 0.3458, 0.8638]], dtype=torch.float64) False

We can use torch.ones_like() or torch.rand_like() to create a new tensor.

Here is an example:

c = torch.rand_like(b, dtype=torch.float32)
print(c)

You will get:

tensor([[0.7924, 0.9788, 0.2649],
        [0.7819, 0.1700, 0.4190]])