Lesson 1 of 16 · Vectors
Scalars and Vectors
Understand scalars and vectors two ways, as arrows and as lists of numbers, plus components, dimensions, feature vectors and embeddings, with pictures and NumPy.
- 7 min read
- Intermediate
- Reviewed
- The difference between a scalar and a vector
- The two ways to picture a vector: an arrow and a list of numbers
- What components and dimensions are
- How every data point in machine learning becomes a vector
- How to create and use vectors with NumPy
Why linear algebra?
Machine learning works with lots of numbers at once: every pixel of an image, every feature of a customer, every number inside a neural network. Linear algebra is the maths of handling whole groups of numbers together, and its two main tools are vectors and matrices. This first module is about vectors.
Scalars: single numbers
Scalars have a size, and nothing else. "25 degrees" does not point anywhere.
Vectors: numbers with a direction, or a list of numbers
The two pictures are the same thing seen two ways:
- As an arrow: "go 3 to the right and 2 up". Physics and geometry like this view.
- As a list:
[3, 2], two numbers in a fixed order. Data science and programming like this view.
Moving an arrow does not change the vector. [3, 2] drawn from (1, 1) is the same vector as [3, 2] drawn from the origin: same length, same direction. Usually we draw vectors starting from the origin.
Components and dimension
We can draw vectors in 2 and 3 dimensions. Beyond that we cannot draw them, but the maths works exactly the same in 100 or 10,000 dimensions. That is the great power of thinking of a vector as a list.
Notation you will see
Books and papers write vectors in several ways. They all mean the same thing:
| Notation | Meaning |
|---|---|
| v (bold) or v⃗ (arrow) | the vector called v |
[3, 2] or (3, 2) | a vector written as a row |
| a tall column of numbers | a column vector, the most common form in maths books |
| vᵢ | the i-th component of v |
Data points are vectors
Here is the idea that connects linear algebra to machine learning:
house = [120, 3, 2, 15] # size, bedrooms, bathrooms, age
print("Dimension:", len(house))
print("Size (first component):", house[0])
print("Age (last component):", house[-1])A photo of 28 × 28 pixels becomes a vector with 784 components. A customer becomes a vector of their age, spending and visits. Once everything is a vector, the same maths works for houses, photos and customers alike.
Vectors in NumPy
A Python list can hold numbers, but it cannot do vector maths. NumPy's array can:
import numpy as np
v = np.array([3, 2])
w = np.array([1, 0, 4])
print(v)
print("Dimension of v:", v.shape[0])
print("Dimension of w:", w.size)
print("Second component of w:", w[1])Common mistakes
Describe customers as feature vectors and read their components.
import numpy as np # Three customers as feature vectors: [age, visits per month, average spend] ada = np.array([34, 12, 450]) ravi = np.array([28, 3, 120]) mei = np.array([45, 8, 300]) # 1. Print the dimension of ada's vector. # 2. Print the number of visits of each customer (the second component), on one line. # 3. Print the age of the oldest customer. # Output must be exactly: # Dimension: 3 # Visits: 12 3 8 # Oldest: 45
A dataset describes each flat with 5 numbers. What are the flats in linear algebra terms?
Summary
- A scalar is one number. A vector is an ordered list of numbers, pictured as an arrow.
- Each number is a component, and the number of components is the dimension.
- Every data point becomes a feature vector, and words become embeddings.
- In NumPy,
np.array([...])makes a vector, and.shapegives its size.
Next: adding vectors and stretching them.
Finished reading? Tick it off to track your progress.