Create PyTorch Tensor with Data Types: An Introduction – PyTorch Tutorial

PyTorch supports to create tensors with different data types. In this tutorial, we will introduce you how to do.

Simple Guide to Create a Tensor in PyTorch – PyTorch Tutorial

4 Methods to Create a PyTorch Tensor – PyTorch Tutorial

PyTorch data types

Torch defines 10 tensor types with CPU and GPU variants, here we will list some common used:

NumPy data types vs PyTorch data types

We also can create a tensor using numpy, it means we should know what numpy data can conver to which tensor type. Here is a list.

How to create torch tensors with different data types?

In pytorch, we can set a data type when creating a tensor.  Here are some examples.

Example 1: create a float 32 tensor

import torch

p = torch.tensor([2, 3], dtype = torch.float32)

print(p)
print(p.dtype)

Run this code, we will see:

tensor([2., 3.])
torch.float32

Here we use dype = torch.float32 to set tensor p data type.

Of course, we also can use torch.FloatTensor to create a float32 data.

import torch

p = torch.FloatTensor([2, 3])

print(p)
print(p.dtype)

Run this code, we also can see:

tensor([2., 3.])
torch.float32

Example 2: create a long integer tensor

As to create a float32 tensor, we can create a long integer tensor as follows:

import torch

x = torch.LongTensor([2, 3, 4])
y = torch.tensor([2, 3, 4], dtype=torch.long)

print(x)
print(x.dtype)

print(y)
print(y.dtype)

Run code, we will see:

tensor([2, 3, 4])
torch.int64
tensor([2, 3, 4])
torch.int64