Broadcasting

Broadcasting is how NumPy lets you do operations on arrays of different shapes — like adding a vector to every row of a matrix — without writing an explicit loop. Once it clicks, it changes how you write array code.

The rule

NumPy aligns the shapes right-to-left. For each axis:

• if the sizes are equal → fine
• if one of them is 1 → it stretches to match the other
• otherwise → error

That’s the whole rule.

A:    (5, 4, 3)
B:       (4, 3)    → aligns as (1, 4, 3) → broadcasts to (5, 4, 3) ✓

A:    (5, 4)
B:    (5,)         → aligns as (1, 5)   → can NOT match (5, 4) ✗
                                              (the trailing axis differs)

The 3 patterns you’ll use 90% of the time

1. Subtract a row from every row (centering)

(click Run)

The mean has shape (3,), X has shape (4, 3). NumPy lines them up as (1, 3) vs (4, 3) — the 1 broadcasts, and the subtraction runs on every row.

This is exactly how you “center” your features before PCA or linear regression. No loop required.

2. Scale each column independently

(click Run)

3. Outer product — column × row

(click Run)

This is the outer product pattern. You see it in attention mechanisms, in computing pairwise distances, and in countless physics simulations.

When broadcasting fails

(click Run)

v[:, np.newaxis] (or equivalently v.reshape(-1, 1)) is how you tell NumPy “treat this 1D array as a column.” This little trick fixes 80% of “could not broadcast” errors.

A non-obvious use: distance matrix

Compute pairwise squared distances between every pair of points — without writing a loop:

(click Run)

What you just wrote is the inner loop of k-nearest-neighbors. The “broadcast a pair of new-axis insertions” pattern is the cleanest way to do all-pairs operations in NumPy.

Quick check