Flattening Arrays in NumPy tutorial with Examples

Flattening an array in NumPy means converting a multi-dimensional array into a one-dimensional array.

This is commonly used when reshaping data, especially in machine learning, image processing, and data analysis, where you often need to format data into a single row or column.

In this tutorial, we will cover:

1. Basic Flattening with flatten()
2. Flattening with ravel()
3. Using reshape() to Flatten Arrays
4. Difference Between flatten() and ravel()
5. Flattening Higher-Dimensional Arrays

Let’s go through each method with examples.

1. Basic Flattening with flatten()

The flatten() method in NumPy converts an array into a one-dimensional array.

It returns a new copy of the array in a flattened form.

Example 1: Flattening a 2D Array

import numpy as np

# Create a 2D array
array_2d = np.array([[1, 2, 3], [4, 5, 6]])

# Flatten the 2D array
flattened_array = array_2d.flatten()
print("Flattened Array:\n", flattened_array)

Output:

Flattened Array:
 [1 2 3 4 5 6]

• Explanation: flatten() creates a new 1D array from the 2D array, copying the data in a row-major (C-style) order by default.

Example 2: Flattening in Column-Major Order

You can specify the order in which to flatten: ‘C' (row-major), ‘F' (column-major), or ‘A' (depends on memory layout).

# Flatten in column-major (Fortran-style) order
flattened_array_fortran = array_2d.flatten(order='F')
print("\nFlattened Array (Column-Major Order):\n", flattened_array_fortran)

Output:

Flattened Array (Column-Major Order):
 [1 4 2 5 3 6]

• Explanation: With order='F', the array is flattened by columns, reading elements in a column-first order.

2. Flattening with ravel()

The ravel() function is similar to flatten(), but it returns a flattened view of the array if possible.

This means that ravel() often does not create a new array but instead provides a reference to the original array, making it more memory-efficient.

Example 3: Flattening a 2D Array with ravel()

# Flatten the array with ravel
raveled_array = array_2d.ravel()
print("Raveled Array:\n", raveled_array)

Output:

Raveled Array:
 [1 2 3 4 5 6]

• Explanation: ravel() provides a 1D view of array_2d, which is generally more memory-efficient than flatten().

Example 4: Column-Major Order with ravel()

# Flatten with column-major (Fortran-style) order
raveled_array_fortran = array_2d.ravel(order='F')
print("\nRaveled Array (Column-Major Order):\n", raveled_array_fortran)

Output:

Raveled Array (Column-Major Order):
 [1 4 2 5 3 6]

• Explanation: Like flatten(), ravel() also allows specifying order='F' to flatten the array by columns.

3. Using reshape() to Flatten Arrays

The reshape() function can also be used to flatten an array by setting its shape to (-1,), where -1 instructs NumPy to infer the dimension.

Example 5: Flattening a 2D Array with reshape()

# Flatten the array using reshape
reshaped_array = array_2d.reshape(-1)
print("Flattened Array using reshape:\n", reshaped_array)

Output:

Flattened Array using reshape:
 [1 2 3 4 5 6]

• Explanation: reshape(-1) flattens the array into a 1D array by inferring the length of the dimension.

Example 6: Column-Major Order with reshape()

# Flatten the array using reshape in column-major order
reshaped_array_fortran = array_2d.reshape(-1, order='F')
print("\nFlattened Array using reshape (Column-Major Order):\n", reshaped_array_fortran)

Output:

Flattened Array using reshape (Column-Major Order):
 [1 4 2 5 3 6]

• Explanation: reshape(-1, order='F') reshapes the array into a flattened 1D array in column-major order.

4. Difference Between flatten() and ravel()

While flatten() and ravel() may appear similar, there is an important difference between them:

• flatten() always returns a copy of the array, which means modifications to the flattened array do not affect the original array.
• ravel() returns a view of the array when possible, which is more memory-efficient. Modifying the output of ravel() may affect the original array.

Example 7: Modifying a Flattened Array with flatten() and ravel()

# Modify the flattened copy with flatten
flattened_copy = array_2d.flatten()
flattened_copy[0] = 99
print("Original Array after flatten modification:\n", array_2d)

# Modify the raveled view with ravel
raveled_view = array_2d.ravel()
raveled_view[0] = 99
print("\nOriginal Array after ravel modification:\n", array_2d)

Output:

Original Array after flatten modification:
 [[ 1  2  3]
 [ 4  5  6]]

Original Array after ravel modification:
 [[99  2  3]
 [ 4  5  6]]

• Explanation: Changing flattened_copy (from flatten()) does not affect array_2d, while modifying raveled_view (from ravel()) changes the original array since ravel() returns a view when possible.

5. Flattening Higher-Dimensional Arrays

The same methods (flatten(), ravel(), and reshape()) can be used to flatten arrays of any dimensionality into a 1D array.

Example 8: Flattening a 3D Array

# Create a 3D array
array_3d = np.array([[[1, 2], [3, 4]], [[5, 6], [7, 8]]])

# Flatten the 3D array
flattened_3d = array_3d.flatten()
print("Flattened 3D Array:\n", flattened_3d)

Output:

Flattened 3D Array:
 [1 2 3 4 5 6 7 8]

Example 9: Column-Major Order Flattening of a 3D Array

# Flatten the 3D array in column-major order
flattened_3d_fortran = array_3d.flatten(order='F')
print("\nFlattened 3D Array (Column-Major Order):\n", flattened_3d_fortran)

Output:

Flattened 3D Array (Column-Major Order):
 [1 5 3 7 2 6 4 8]

• Explanation: Flattening higher-dimensional arrays in column-major order changes the order in which elements are accessed and listed in the flattened array.

Summary of Array Flattening Techniques in NumPy

Conclusion

In this tutorial, we explored various ways to flatten arrays in NumPy, including:

• Flattening with flatten(), which returns a new flattened array.
• Flattening with ravel(), which returns a flattened view of the original array when possible.
• Flattening with reshape(), which converts an array to 1D with shape (-1,).
• Column-major order flattening to control the order of elements in the flattened array.