The Difference Between PyTorch tensor.data and tensor.item() – PyTorch Tutorial

Pytorch tensor.data and tensor.item() can get the value of a tensor. In this tutorial, we will introduce the difference between them.

tensor.data

tensor.data can get a copy of the original tensor, here is the detail.

Understand PyTorch tensor.data with Examples – PyTorch Tutorial

tensor.item()

It will return a real value of an element in a tensor.  We will use an example to show the difference between them.

For example:

import torch
x = torch.tensor([[1.0, 2.0],[2.0, 3.0]], requires_grad = True)
print(x)
print(type(x))

print(x[0][1])
z = x[0][1].item()
print(z)
print(type(z))

Run this code, we will see:

tensor([[1., 2.],
        [2., 3.]], requires_grad=True)
<class 'torch.Tensor'>
tensor(2., grad_fn=)
2.0
<class 'float'>

From this result, we can find:

tensor.item() can be applied on an element, it will return a python type, such as float.

We also can use index to get an element value , for example x[0][1], however, the returned type is tensor.

Look at example below:

z = x[0].item()
print(z)
print(type(z))

Run this code, we will see this error:

ValueError: only one element tensors can be converted to Python scalars

Here we use x[0] to get an item. However, x[0] contains two elements. It also means we only can use tensor.item() on a tensor that only cotains an element.