np.identity(n) is a simpler, more restricted version of np.eye(n): it always produces a square matrix with the ones strictly on the main diagonal, with no equivalent of np.eye()'s M, non-square shape, or k, diagonal offset, parameters. For the common case of just needing a plain identity matrix, np.identity() communicates that specific intent slightly more directly than calling np.eye() with its extra unused parameters.

1Understanding np.identity()

💡

Use np.identity(n) instead of np.eye(n) specifically when you want a plain square identity matrix and nothing more — it's functionally identical to np.eye(n) with no extra arguments, but reads more directly as expressing exactly that intent.

editor.html

import numpy as np

I = np.identity(4)
print(I)

localhost:3000

2Practical Example

Here is a real-world application of np.identity() showing how it is used in production NumPy code.

editor.html

import numpy as np

A = np.array([[2, 0], [0, 3]])
I = np.identity(2)
print(A @ I)

localhost:3000

3Best Practices

Follow these guidelines when working with np.identity():

1. Use np.identity(n) for a plain identity matrix; switch to np.eye() only when you need its extra shape or diagonal-offset options

2. Cast to an integer dtype explicitly (dtype=int) if you need an identity matrix without floating-point representation

3. Use np.identity(n) as the starting point when iteratively building up a matrix via a sequence of linear transformations

⚠️

Tip: Use np.identity(n) instead of np.eye(n) specifically when you want a plain square identity matrix and nothing more — it's functionally identical to np.eye(n) with no extra arguments, but reads more directly as expressing exactly that intent.

editor.html

import numpy as np

I = np.identity(4)
print(I)

localhost:3000

Examples

Example 01Basic Usage

import numpy as np

I = np.identity(4)
print(I)

Example 02Advanced Example

import numpy as np

A = np.array([[2, 0], [0, 3]])
I = np.identity(2)
print(A @ I)

Best Practices

Use np.identity(n) for a plain identity matrix; switch to np.eye() only when you need its extra shape or diagonal-offset options
Cast to an integer dtype explicitly (dtype=int) if you need an identity matrix without floating-point representation
Use np.identity(n) as the starting point when iteratively building up a matrix via a sequence of linear transformations

Interview Question

What's the practical difference between np.identity(n) and calling np.eye(n) with no extra arguments?▼

Hint: Think about what each function is actually capable of producing beyond the simple case.

Called with just n, np.identity(n) and np.eye(n) produce the exact same result — a square identity matrix. The difference is in what each function is designed for: np.eye() additionally supports a non-square shape via its M parameter and a shifted diagonal via its k parameter, while np.identity() is deliberately restricted to only the plain square identity case. Choosing np.identity() when that's genuinely all you need communicates that intent more precisely to a reader than np.eye() with its unused extra capability.

Exercises

MediumPractice using np.identity() in a real scenario.

View Solution

import numpy as np

I = np.identity(4)
print(I)

Frequently Asked Questions

What's the practical difference between np.identity(n) and calling np.eye(n) with no extra arguments?▼

Related Functions

np-eyenp-arraynp-linalg-inv