Squeezing Dimensions¶

The np.squeeze function removes all size-1 dimensions from an array.

Mental Model

squeeze is the inverse of expand_dims -- it strips away every axis that has only one element, collapsing unnecessary packaging around your data. Think of it like removing single-item wrappers: a shape (1, 3, 1, 5) becomes (3, 5) because those length-1 dimensions carry no information.

np.squeeze Function¶

Removes axes of length one from the array shape.

1. Basic Usage¶

```python import numpy as np

def main(): x = np.random.normal(size=(1, 3, 1, 2, 1, 5, 1)) y = np.squeeze(x) print(f"{x.shape = }") print(f"{y.shape = }")

if name == "main": main() ```

Output:

x.shape = (1, 3, 1, 2, 1, 5, 1) y.shape = (3, 2, 5)

2. All Size-1 Removed¶

Every dimension with size 1 is eliminated from the result.

Selective Squeeze¶

Remove only specific size-1 axes using the axis parameter.

1. Single Axis¶

```python import numpy as np

def main(): x = np.zeros((1, 3, 1, 4)) y = np.squeeze(x, axis=0) print(f"{x.shape = }") print(f"{y.shape = }")

if name == "main": main() ```

Output:

x.shape = (1, 3, 1, 4) y.shape = (3, 1, 4)

2. Multiple Axes¶

```python import numpy as np

def main(): x = np.zeros((1, 3, 1, 4)) y = np.squeeze(x, axis=(0, 2)) print(f"{x.shape = }") print(f"{y.shape = }")

if name == "main": main() ```

Output:

x.shape = (1, 3, 1, 4) y.shape = (3, 4)

Error Handling¶

Squeezing non-singleton dimensions raises an error.

1. ValueError Example¶

```python import numpy as np

def main(): x = np.zeros((1, 3, 1, 4)) try: y = np.squeeze(x, axis=1) except ValueError as e: print(f"Error: {e}")

if name == "main": main() ```

2. Safe Practice¶

Only squeeze axes you know have size 1.

Common Use Cases¶

Squeeze is frequently needed after certain operations.

1. After Slicing¶

Slicing with a single index reduces dimensionality; squeeze cleans up.

2. Model Outputs¶

Neural network outputs often have extra batch or channel dimensions.

3. Broadcasting Result¶

Broadcasting may introduce size-1 dimensions that need removal.

Batch Size 1 Trap

The most common squeeze bug in ML: a model output has shape (1, 10) (batch size 1, 10 classes). Calling np.squeeze removes the batch dimension, producing shape (10,). Later code that expects a 2D batch array breaks silently.

python output = model.predict(single_image) # shape (1, 10) squeezed = np.squeeze(output) # shape (10,) — batch dim gone!

Fix: use selective squeeze (axis=) to remove only the dimensions you intend, never blindly squeeze all size-1 axes.

Exercises¶

Exercise 1. Create an array a = np.zeros((1, 5, 1, 3, 1)). Use np.squeeze to remove all size-1 dimensions and print the resulting shape. Then use selective squeeze to remove only the dimension at axis 0, and print that shape.

Solution to Exercise 1

import numpy as np

a = np.zeros((1, 5, 1, 3, 1))
b = np.squeeze(a)
print(b.shape)  # (5, 3)

c = np.squeeze(a, axis=0)
print(c.shape)  # (5, 1, 3, 1)

Exercise 2. Compute the sum of rows of a = np.array([[1, 2, 3], [4, 5, 6]]) using a.sum(axis=1, keepdims=True). Print the shape of the result. Then squeeze it and verify the shape becomes (2,).

Solution to Exercise 2

import numpy as np

a = np.array([[1, 2, 3], [4, 5, 6]])
row_sums = a.sum(axis=1, keepdims=True)
print(row_sums.shape)  # (2, 1)

squeezed = row_sums.squeeze()
print(squeezed.shape)  # (2,)
print(squeezed)         # [ 6 15]

Exercise 3. Create a = np.array([[42]]) (shape (1, 1)). Squeeze it completely and check the resulting ndim. Then extract the Python scalar using .item() and verify its type is int.

Solution to Exercise 3

import numpy as np

a = np.array([[42]])
b = a.squeeze()
print(b.ndim)   # 0 (scalar array)
print(b.shape)  # ()

val = b.item()
print(val)           # 42
print(type(val))     # <class 'int'>