Broadcasting Rules
Understand NumPy's broadcasting rules so you can add a 1-D array to every row of a 2-D array without explicit replication.
Broadcasting Rules is a free Pandas & NumPy Academy lesson on CoddyKit. This is lesson 3 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the Pandas & NumPy Academy learning path, and your progress syncs across the web and the CoddyKit app. The Pandas & NumPy Academy course includes 4 lessons in total.
The Problem Broadcasting Solves
Broadcasting lets NumPy combine arrays of different but compatible shapes without copying data — the smaller one is stretched to fit the larger.
import numpy as np
# Adding a scalar to an array is the simplest broadcast
a = np.array([1, 2, 3])
print(a + 10) # [11 12 13]
# Scalar 10 is 'broadcast' to shape (3,)Broadcasting Rule 1: Prepend 1s
NumPy lines up shapes from the right. If one array has fewer dimensions, it pads the left with 1s — so a (3,) array acts like (1, 3) next to a (4, 3) one.
import numpy as np
m = np.ones((4, 3))
v = np.array([10, 20, 30]) # shape (3,) -> treated as (1, 3)
result = m + v # shape (4, 3)
print(result)
# [[11. 21. 31.]
# [11. 21. 31.]
# [11. 21. 31.]
# [11. 21. 31.]]Broadcasting Rule 2: Stretch Size-1 Dimensions
Any dimension of size 1 can stretch to match the other array — no data is actually copied. Both arrays can stretch their size-1 dimensions at once.
import numpy as np
# (3, 1) + (1, 4) --> (3, 4)
col = np.array([[1], [2], [3]]) # shape (3, 1)
row = np.array([[10, 20, 30, 40]]) # shape (1, 4)
print((col + row).shape) # (3, 4)
print(col + row)Broadcasting Rule 3: Incompatible Shapes
If two dimensions aren't equal and neither is 1, broadcasting fails with a ValueError. Checking shapes first gives you a clear error, not a silent bug.
import numpy as np
a = np.ones(3)
b = np.ones(4)
try:
a + b
except ValueError as e:
print(e)
# operands could not be broadcast together with shapes (3,) (4,)Practical: Subtracting the Column Mean
A classic move: subtract each column's mean from every row to centre your data. With keepdims=True, the mean keeps a shape that broadcasts cleanly.
import numpy as np
data = np.array([[1., 2., 3.],
[4., 5., 6.],
[7., 8., 9.]])
col_mean = data.mean(axis=0) # shape (3,)
centred = data - col_mean # broadcast along axis 0
print(centred)
# [[-3. -3. -3.]
# [ 0. 0. 0.]
# [ 3. 3. 3.]]Practical: Row Normalisation
To make each row sum to 1, divide by its row sum. Use keepdims=True so the sum stays shape (n, 1) and broadcasts across the columns.
import numpy as np
m = np.array([[1., 2., 3.],
[4., 5., 6.]])
row_sums = m.sum(axis=1, keepdims=True) # shape (2, 1)
normed = m / row_sums
print(normed.round(3))
# [[0.167 0.333 0.5 ]
# [0.267 0.333 0.4 ]]Outer Products via Broadcasting
Reshape one array to (n, 1) and another to (1, m), then multiply — broadcasting builds the full outer product matrix for you, no loops needed.
import numpy as np
a = np.array([1, 2, 3])
b = np.array([10, 20, 30, 40])
outer = a[:, np.newaxis] * b[np.newaxis, :]
print(outer)
# [[ 10 20 30 40]
# [ 20 40 60 80]
# [ 30 60 90 120]]Broadcasting with 3-D Arrays
Broadcasting scales to any dimensions. A batch of images shaped (100, 28, 28) can be centred by a per-pixel mean of (1, 28, 28) — NumPy stretches the 1.
import numpy as np
batch = np.random.rand(100, 28, 28) # 100 images
pixel_mean = batch.mean(axis=0, keepdims=True) # (1, 28, 28)
centred = batch - pixel_mean # (100, 28, 28)
print(centred.shape) # (100, 28, 28)np.broadcast_to for Explicit Stretching
np.broadcast_to shows you exactly what a broadcast produces, as a read-only view with no copy. It's perfect for picturing how shapes line up.
import numpy as np
a = np.array([1, 2, 3])
view = np.broadcast_to(a, (4, 3))
print(view)
# [[1 2 3]
# [1 2 3]
# [1 2 3]
# [1 2 3]]
print(view.flags.writeable) # FalseVisualising Compatible Shapes
A handy rule: line shapes up from the right and check each pair. They're compatible if they're equal or one is 1. Predict shapes before you run code!
# Shape compatibility examples:
# (3, 4) + ( 4) -> (3, 4) OK: 4==4, 1 implied
# (3, 4) + (3, 1) -> (3, 4) OK: 4 vs 1, 3==3
# (2, 3, 4) + (3, 4) -> (2, 3, 4) OK
# (3, 4) + (3, ) -> ERROR: 4 vs 3
import numpy as np
print(np.zeros((3,4)).shape) # (3, 4)
print((np.zeros((3,4)) + np.zeros(4)).shape) # (3, 4)Common Broadcasting Mistakes
The top broadcasting trap is forgetting keepdims=True after aggregating, so shapes misalign. When stuck, print(arr.shape) to see what's really going on.
import numpy as np
m = np.arange(6).reshape(2, 3)
row_max = m.max(axis=1) # shape (2,) NOT (2,1)
print(row_max.shape) # (2,)
# m - row_max -> ERROR: shapes (2,3) and (2,) misalign
row_max_col = row_max[:, np.newaxis] # shape (2, 1)
print((m - row_max_col).shape) # (2, 3) OKQuick Check
Test your understanding of NumPy broadcasting rules from this lesson.
Lesson Recap
You've got it! Broadcasting stretches size-1 dimensions, compares shapes from the right, and keepdims=True keeps axes so math lines up. Next: boolean masking.
Frequently Asked Questions
Is the “Broadcasting Rules” lesson free?
Yes — the full text of “Broadcasting Rules” is free to read here on the web. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the Pandas & NumPy Academy course, upgrade to CoddyKit PRO. The Pandas & NumPy Academy course includes 4 lessons in total.
What will I learn in “Broadcasting Rules”?
Understand NumPy's broadcasting rules so you can add a 1-D array to every row of a 2-D array without explicit replication. You practise Pandas & NumPy Academy with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.
Do I need any experience to start Pandas & NumPy Academy?
No prior experience is required. Pandas & NumPy Academy on CoddyKit is structured for beginners through advanced learners, so you can start here or from the beginning and move at your own pace. This is lesson 3 of 4.
How long does the “Broadcasting Rules” lesson take?
Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.
Can I write and run code in this Pandas & NumPy Academy lesson?
Yes. Every Pandas & NumPy Academy lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.