np.expand_dims() and np.squeeze() in NumPy

Introduction

We use np.expand_dims() to add a new dimension to data, and np.squeeze() to remove dimensions of size one. This helps shape data correctly for analysis or machine learning.

When you have a 1D array but need a 2D array for a function that expects rows and columns.

When you want to remove extra single dimensions from data after processing.

When preparing image data that requires a specific number of dimensions.

When reshaping data to match expected input shapes in machine learning models.

When cleaning up data shapes after slicing or selecting specific elements.

Syntax

NumPy

np.expand_dims(array, axis)
np.squeeze(array, axis=None)

axis in expand_dims is where the new dimension is added.

axis in squeeze specifies which single-sized dimension to remove; if None, all single-sized dimensions are removed.

Examples

Adds a new dimension at the front, changing shape from (3,) to (1, 3).

NumPy

import numpy as np
arr = np.array([1, 2, 3])
new_arr = np.expand_dims(arr, axis=0)
print(new_arr.shape)

Removes all single-sized dimensions, changing shape from (1, 3, 1) to (3,).

NumPy

arr = np.array([[[1], [2], [3]]])
squeezed = np.squeeze(arr)
print(squeezed.shape)

Adds a new dimension as a column, changing shape from (3,) to (3, 1).

NumPy

arr = np.array([1, 2, 3])
new_arr = np.expand_dims(arr, axis=1)
print(new_arr.shape)

Removes the first dimension only, changing shape from (1, 3, 1) to (3, 1).

NumPy

arr = np.array([[[1], [2], [3]]])
squeezed = np.squeeze(arr, axis=0)
print(squeezed.shape)

Sample Program

This program shows how to add new dimensions to a 1D array and how to remove single-sized dimensions from a 3D array using np.expand_dims() and np.squeeze().

NumPy

import numpy as np

# Original 1D array
arr = np.array([10, 20, 30])
print(f"Original shape: {arr.shape}")

# Add new dimension at axis 0
expanded = np.expand_dims(arr, axis=0)
print(f"Shape after expand_dims at axis 0: {expanded.shape}")

# Add new dimension at axis 1
expanded2 = np.expand_dims(arr, axis=1)
print(f"Shape after expand_dims at axis 1: {expanded2.shape}")

# Create array with extra single dimensions
arr2 = np.array([[[10], [20], [30]]])
print(f"Shape before squeeze: {arr2.shape}")

# Remove all single dimensions
squeezed = np.squeeze(arr2)
print(f"Shape after squeeze (all single dims removed): {squeezed.shape}")

# Remove single dimension only at axis 0
squeezed_axis0 = np.squeeze(arr2, axis=0)
print(f"Shape after squeeze at axis 0: {squeezed_axis0.shape}")

Important Notes

Adding or removing dimensions does not change the data, only its shape.

Be careful with axis in squeeze: if the dimension is not size one, it will raise an error.

These functions are useful when working with libraries that expect specific input shapes.

Summary

np.expand_dims() adds a new dimension to an array at the specified axis.

np.squeeze() removes dimensions of size one from an array.

Use these to reshape data for analysis or machine learning without changing the data itself.