Why np.expand_dims() and np.squeeze() in NumPy? - Purpose & Use Cases

The Big Idea

What if you could reshape your data perfectly every time without guessing or errors?

The Scenario

Imagine you have a list of numbers representing data, but your analysis tool expects data in a specific shape, like a table with rows and columns. You try to add or remove these extra dimensions manually by counting commas and brackets, which quickly becomes confusing and messy.

The Problem

Manually reshaping data by adding or removing dimensions is slow and error-prone. You might add too many brackets or remove the wrong ones, causing your program to crash or give wrong results. It's like trying to fold a map perfectly by guessing where the creases should be.

The Solution

Using np.expand_dims() and np.squeeze() lets you easily add or remove single dimensions from your data arrays. These functions handle the shape changes cleanly and safely, so you don't have to worry about counting brackets or messing up your data structure.

Before vs After

✗ Before

data = [[1, 2], [3, 4]]  # Trying to add a dimension manually
new_data = [[[1, 2], [3, 4]]]

✓ After

import numpy as np
data = np.array([[1, 2], [3, 4]])
new_data = np.expand_dims(data, axis=0)

What It Enables

It enables smooth and error-free reshaping of data arrays, making complex data processing and machine learning workflows much easier.

Real Life Example

When preparing images for a machine learning model, you often need to add a batch dimension to a single image array. np.expand_dims() adds this dimension effortlessly, while np.squeeze() removes unnecessary ones after processing.

Key Takeaways

Manually changing array shapes is confusing and risky.

np.expand_dims() adds a new dimension safely.

np.squeeze() removes unwanted single dimensions easily.

Practice

1. What does np.expand_dims() do to a NumPy array?

easy

A. Adds a new dimension at the specified axis

B. Removes all dimensions of size one

C. Flattens the array into one dimension

D. Changes the data type of the array

Solution

1. Step 1: Understand the function purpose

   np.expand_dims() is used to add a new axis to an array, increasing its dimensions.

2. Step 2: Compare with other options

   Removing dimensions or flattening is done by other functions like np.squeeze() or reshape(). Changing data type is unrelated.

3. Final Answer:

   Adds a new dimension at the specified axis -> Option A

4. Quick Check:

   np.expand_dims() = Adds dimension [OK]

Hint: Expand dims adds axis, squeeze removes axis [OK]

Common Mistakes:

• Confusing expand_dims with squeeze
• Thinking it changes data values
• Assuming it flattens the array

2. Which of the following is the correct syntax to add a new axis at position 1 using np.expand_dims() on array a?

easy

A. np.expand_dims(a, axis=0)

B. np.expand_dims(a, 0)

C. np.expand_dims(a, axis=1)

D. np.expand_dims(a, axis=2)

Solution

1. Step 1: Check the function signature

   np.expand_dims(array, axis) requires the axis parameter to specify where to add the new dimension.

2. Step 2: Identify correct axis value

   Axis=1 adds the new dimension at position 1. np.expand_dims(a, 0) (positional) and np.expand_dims(a, axis=0) add the dimension at position 0. np.expand_dims(a, axis=2) causes an AxisError since valid axes are -2 to 1 for a 1D array.

3. Final Answer:

   np.expand_dims(a, axis=1) -> Option C

4. Quick Check:

   Use axis=1 to add dimension at position 1 [OK]

Hint: Use axis= to specify position clearly [OK]

Common Mistakes:

• Omitting axis keyword and confusing position
• Using wrong axis number
• Mixing axis=0 and axis=1

3. What is the shape of the array after applying np.squeeze() to arr = np.array([[[1], [2], [3]]])?

medium

A. (3,)

B. (1, 3, 1)

C. (3, 1)

D. (3, 1, 1)

Solution

1. Step 1: Check original shape

   arr has shape (1, 3, 1) because it is a 3D array with one element in first and last dimension.

2. Step 2: Apply np.squeeze()

   np.squeeze() removes all dimensions of size 1, so the first and last dimensions are removed, leaving shape (3,).

3. Final Answer:

   (3,) -> Option A

4. Quick Check:

   squeeze removes size 1 dims = (3,) [OK]

Hint: Squeeze removes all size 1 dims from shape [OK]

Common Mistakes:

• Assuming squeeze removes only one dimension
• Confusing original shape with squeezed shape
• Thinking squeeze flattens the array

4. Given arr = np.array([1, 2, 3]), which code will cause an error?

medium

A. np.expand_dims(arr, axis=-1)

B. np.expand_dims(arr, axis=0)

C. np.squeeze(arr)

D. np.expand_dims(arr, axis=3)

Solution

1. Step 1: Check original array shape

   arr has shape (3,), so it has 1 dimension.

2. Step 2: Validate axis values for expand_dims

   Valid axis values for adding a new dimension are between -2 and 1 (inclusive) because adding a dimension increases rank by 1. Axis=3 is out of bounds and causes an error.

3. Final Answer:

   np.expand_dims(arr, axis=3) causes error -> Option D

4. Quick Check:

   Axis out of range causes error [OK]

Hint: Axis must be in valid range for expand_dims [OK]

Common Mistakes:

• Using axis larger than array ndim + 1
• Confusing axis indexing with array shape
• Assuming squeeze can cause error on 1D arrays

5. You have a 2D array data with shape (5, 1). You want to prepare it for a machine learning model that expects input shape (5,). Which code correctly reshapes data using np.squeeze() and np.expand_dims()?

hard

A. np.expand_dims(np.squeeze(data), axis=1)

B. np.squeeze(data)

C. np.squeeze(np.expand_dims(data, axis=-1))

D. np.expand_dims(data, axis=-1)

Solution

1. Step 1: Understand the original shape and target shape

   data shape is (5, 1). The model expects (5,), so we need to remove the extra dimension.

2. Step 2: Use np.squeeze() to remove size 1 dimension

   np.squeeze(data) removes the dimension of size 1, resulting in shape (5,). Adding dimensions again is unnecessary and would change shape incorrectly.

3. Final Answer:

   np.squeeze(data) -> Option B

4. Quick Check:

   Squeeze removes extra dims to match model input [OK]

Hint: Squeeze to remove size 1 dims for model input [OK]

Common Mistakes:

• Adding dimensions instead of removing
• Using expand_dims unnecessarily
• Not matching model expected input shape