Broadcasting in NumPy

What is broadcasting?

Broadcasting is NumPy’s ability to perform operations on arrays with different shapes by automatically expanding (virtually) smaller arrays.

This is a major reason NumPy code is concise and fast. Nothing is actually copied in memory — NumPy just repeats the smaller array’s values as if it were stretched to fit, without allocating that extra memory.

Simple example: add scalar

scalar

import numpy as np
 
arr = np.array([1, 2, 3])
print(arr + 10)  # [11 12 13]

Here, the scalar 10 is broadcast to match the shape (3,).

Example: add vector to matrix

vector-matrix

import numpy as np
 
mat = np.array([
    [1, 2, 3],
    [4, 5, 6]
])
vec = np.array([10, 20, 30])
 
print(mat + vec)

vec (shape (3,)) broadcasts across each row.

sketch Broadcasting a row vector across a matrix p5.js Pause Restart

The (3,) vector is virtually stretched down to match every row of the (2, 3) matrix — no extra memory is actually allocated.

Broadcasting rules (must know)

When operating on two arrays, NumPy compares shapes from the trailing dimension (rightmost first) and works backward.

Two dimensions are compatible when:

1. They are equal, OR
2. One of them is 1

If dimensions are incompatible → broadcasting error.

Example: column vector + matrix

column

import numpy as np
 
mat = np.array([
    [1, 2, 3],
    [4, 5, 6]
])
col = np.array([100, 200]).reshape(2, 1)
 
print(mat + col)

col has shape (2, 1) and broadcasts across columns.

Common broadcasting error

error

import numpy as np
 
mat = np.zeros((2, 3))
vec = np.array([1, 2])
 
# mat + vec -> ValueError (shapes (2,3) and (2,) not compatible)

Fix by reshaping vec to a column vector if that’s what you intend:

fix

vec = vec.reshape(2, 1)
print(mat + vec)

From the book

McKinney treats broadcasting as one of NumPy’s most powerful — and most confusing — features. The rule to memorize: compare shapes from the last axis backward; each pair of axis sizes must be equal or one of them must be 1. When you hit a ValueError, it’s almost always a shape mismatch on a non-trailing axis that needs a reshape or [:, None].

diagram Broadcasting compatibility check mermaid

  flowchart TD
  A["Compare shapes,
trailing axis first"] --> B{"Axis sizes equal
or one is 1?"}
  B -- "Yes, for every axis" --> C["Broadcast succeeds
virtually stretch the smaller array"]
  B -- "No, some axis mismatches" --> D["ValueError:
shapes not aligned"]

NumPy compares shapes from the trailing dimension inward; each axis pair must match or be 1.

Why broadcasting is useful in analytics

• Normalize columns: X / X.max(axis=0)
• Center data: X - X.mean(axis=0)
• Apply weights: X * weights

Next

Continue to: NumPy Arithmetic Operations for element-wise math and matrix operations.

🧪 Try It Yourself

Exercise 1 – Broadcast a Scalar

pch.exerciseTag pch.exerciseTitle

Exercise 2 – Broadcast a Row Vector Across a Matrix

pch.exerciseTag pch.exerciseTitle

Exercise 3 – Fix a Shape Mismatch

pch.exerciseTag pch.exerciseTitle