How to access the metadata of a tensor in PyTorch?

In this article, we are going to see how to access the metadata of a tensor in PyTorch using Python.

PyTorch in Python is a machine learning library. Also, it is free and open-source. It was firstly introduced by the Facebook AI research team. A tensor in PyTorch is similar to a NumPy array. But it doesn’t have any knowledge of deep learning, graphs, etc. It is considered a normal n-dimensional array that can be used for mathematical computation. It differs from a Numpy in terms of its execution platform. Unlike Numpy array, a tensor is capable of running on CPU or GPU.

Example 1:

In this example, each of the statements used in the source code has been discussed below,

The first step is to import the torch library. We need to create a tensor. For example, we have created a tensor of dimension 5 X 3. Now to access metadata that is, the size and shape of the tensor we have used the .size() and .shape method. We have used the torch.numel() method. It gives us the total number of elements in the created tensor. Finally, we are printing these data to the console.

Python3

# Python Program demonstrate how
# to access meta-data of a Tensor

# Import necessary libraries
import torch

# Create a tensor of having dimensions 5 X 3
tensor = torch.Tensor([[5,1,7],[7,2,9],[4,7,9],
                       [8,12,14],[2,4,7]])
print("tensor:\n", tensor)

# Get the meta-data of tensor
# Get the size of tensor
tensor_size = tensor.size()
print("The size of tensor:\n", tensor_size)

# Applying .shape method to get the tensor size
print("The shape of tensor:\n", tensor.shape)

# Compute the number of elements in the tensor
size = torch.numel(tensor)
print("Total number of elements in tensor:\n", size)

Output:

tensor:
 tensor([[ 5.,  1.,  7.],
        [ 7.,  2.,  9.],
        [ 4.,  7.,  9.],
        [ 8., 12., 14.],
        [ 2.,  4.,  7.]])
The size of tensor:
 torch.Size([5, 3])
The shape of tensor:
 torch.Size([5, 3])
Total number of elements in tensor:
 15

Example 2:

In this example, each of the statements used in the source code has been discussed below,

The first step is to import the torch library. We need to create a tensor. For example, we have created a tensor of numbers from 1 to 9 (inclusive). Now to access metadata that is, the size and shape of the tensor we have used the .size() and .shape method. We have used the torch.numel() method. It gives us the total number of elements in the created tensor. Finally, we are printing these data to the console.

Python3

# Python Program demonstrate how to
# access meta-data of a Tensor

# Import the library
import torch

# Creating a tensor
data = [1, 2, 3, 4, 5, 6, 7, 8, 9]
tensor = torch.tensor(data)

# Printing tensor
print(tensor)

# Get the meta-data of tensor
# Get the size of tensor
tensor_size = tensor.size()
print("The size of tensor:\n", tensor_size)

# Applying .shape method to get the tensor size
print("The shape of tensor:\n", tensor.shape)

# Compute the number of elements in the tensor
size = torch.numel(tensor)
print("Total number of elements in tensor:\n", size)

Output:

tensor([1, 2, 3, 4, 5, 6, 7, 8, 9])
The size of tensor:
 torch.Size([9])
The shape of tensor:
 torch.Size([9])
Total number of elements in tensor:
 9