Skip to content

Lesson 2: PyTorch Basics

Introduction

In this lesson, we'll explore the fundamental data structures in PyTorch that form the building blocks of deep learning models.

Scalars

A scalar is a single number. In PyTorch, you can create a scalar tensor:

scalar = torch.tensor(7)

Scalars are 0-dimensional tensors, often used for loss values or simple numerical operations.

Vectors

A vector is a 1-dimensional array of numbers. Think of it as a line of values:

vector = torch.tensor([1, 2, 3, 4])

Vectors are commonly used for representing features or single data points.

Matrices

A matrix is a 2-dimensional array of numbers, organized in rows and columns:

matrix = torch.tensor([[1, 2, 3],
            [4, 5, 6],
            [7, 8, 9]])

Matrices are essential for representing images, feature maps, or layers of neural networks.

Tensors

Tensors are the generalization of vectors and matrices to higher dimensions. They're the core data structure in PyTorch:

# 3D tensor
tensor = torch.tensor([[[1, 2], [3, 4]],
            [[5, 6], [7, 8]]])

Key Tensor Properties

  • Shape: Describes the dimensions of a tensor
  • Rank: Number of dimensions
  • Device: Where the tensor is stored (CPU/GPU)
# Example of tensor properties
t = torch.tensor([[1, 2], [3, 4]])
print(f"Shape: {t.shape}")
print(f"Rank: {len(t.shape)}")
print(f"Device: {t.device}")

Practice Exercise

Try creating a Tensor to represent a common use case in your life. For example use a matrix to represent a list of students, where each inner array is 1 class. Then use a Tensor where each inner array represents a grade and the array inside that a class.