PyTorch Pocket Reference

Chapter 2. Tensors

Before we dive deep into the world of PyTorch development, it’s important to familiarize yourself with the fundamental data structure in PyTorch: the torch.Tensor. By understanding the tensor, you will understand how PyTorch handles and stores data, and since deep learning is fundamentally the collection and manipulation of floating-point numbers, understanding tensors will help you understand how PyTorch implements more advanced functions for deep learning. In addition, you may find yourself using tensor operations frequently when preprocessing input data or manipulating output data during model development.

This chapter serves as a quick reference to understanding tensors and implementing tensor functions within your code. I’ll begin by describing what a tensor is and show you some simple examples of how to use functions to create, manipulate, and accelerate tensor operations on a GPU. Next, we’ll take a broader look at the API for creating tensors and performing math operations so that you can quickly reference a comprehensive list of tensor capabilities. In each section, we will explore some of the more important functions, identify common pitfalls, and examine key points in their usage.

What Is a Tensor?

In PyTorch, a tensor is a data structure used to store and manipulate data. Like a NumPy array, a tensor is a multidimensional array containing elements of a single data type. Tensors can be used to represent scalars, vectors, matrices, and n-dimensional ...