While accessing elements by their numerical position (like array[2] or array[1, 3]) and slicing ranges of data (like array[1:5]) are useful, often you need to select data based on its value or some condition. This is where boolean indexing comes into play. It's a powerful technique that allows you to filter arrays using True/False conditions.

Creating Boolean Arrays

The foundation of boolean indexing is the boolean array. You typically create these arrays by applying comparison operators directly to a NumPy array. NumPy performs the comparison element-wise, returning a new array of the same shape filled with True or False values.

Let's see an example:

import numpy as np

# Create a simple 1D array
data = np.array([1, 5, 2, 8, 3, 7, 4, 6])
print(f"Original array: {data}")

# Apply a condition
is_greater_than_4 = data > 4
print(f"Boolean array (data > 4): {is_greater_than_4}")

Executing this code will output:

Original array: [1 5 2 8 3 7 4 6]
Boolean array (data > 4): [False  True False  True False  True False  True]

Notice how is_greater_than_4 has the same number of elements as data. Each element in is_greater_than_4 is True if the corresponding element in data is greater than 4, and False otherwise.

You can use any standard comparison operators: >, <, >=, <=, == (equal to), and != (not equal to).

Selecting Data with Boolean Arrays

Once you have a boolean array, you can use it directly inside the square brackets [] of the original array. This operation selects only the elements from the original array where the corresponding value in the boolean array is True.

# Use the boolean array to select elements
selected_data = data[is_greater_than_4]
print(f"Selected elements (where data > 4): {selected_data}")

# You can also apply the condition directly inside the brackets
selected_directly = data[data < 5]
print(f"Selected elements (where data < 5): {selected_directly}")

Output:

Selected elements (where data > 4): [5 8 7 6]
Selected elements (where data < 5): [1 2 3 4]

The result is a new 1D array containing only the elements that satisfied the condition. This works regardless of the original array's dimensions.

Let's try it with a 2D array:

matrix = np.array([[1, 2, 3],
                   [4, 5, 6],
                   [7, 8, 9]])

# Select elements greater than 5
selected_matrix_elements = matrix[matrix > 5]
print(f"\nOriginal 2D array:\n{matrix}")
print(f"Selected elements (where matrix > 5): {selected_matrix_elements}")

Output:

Original 2D array:
[[1 2 3]
 [4 5 6]
 [7 8 9]]
Selected elements (where matrix > 5): [6 7 8 9]

Important: Notice that even when applying boolean indexing to a 2D array, the result is often a 1D array containing the selected elements. NumPy flattens the output because the True values might be scattered across different rows and columns, not necessarily forming a rectangular sub-array.

Scatter plot showing original data points (gray) and points selected using the condition data > 4 (blue).

Combining Multiple Conditions

You often need to filter data based on several conditions simultaneously. You can combine boolean arrays using logical operators:

& (logical AND): Selects elements where both conditions are True.
| (logical OR): Selects elements where at least one condition is True.
~ (logical NOT): Inverts the boolean array (changes True to False and False to True).

Important: When combining conditions with & or |, you must use parentheses () around each individual condition. This is because of operator precedence rules in Python and NumPy. The bitwise operators & and | have higher precedence than comparison operators like < or >. Without parentheses, Python might try to evaluate something like data > 3 & data which leads to an error or unexpected results.

data = np.array([1, 5, 2, 8, 3, 7, 4, 6, 9, 0])

# Condition 1: data > 3
# Condition 2: data < 8
selected_and = data[(data > 3) & (data < 8)]
print(f"Elements where (data > 3) AND (data < 8): {selected_and}")

# Condition 1: data is even (data % 2 == 0)
# Condition 2: data > 5
selected_or = data[(data % 2 == 0) | (data > 5)]
print(f"Elements where (data is even) OR (data > 5): {selected_or}")

# Condition: data is NOT equal to 5
selected_not = data[~(data == 5)]
print(f"Elements where data is NOT equal to 5: {selected_not}")

Output:

Elements where (data > 3) AND (data < 8): [5 7 4 6]
Elements where (data is even) OR (data > 5): [2 8 7 4 6 9 0]
Elements where data is NOT equal to 5: [1 2 8 3 7 4 6 9 0]

Modifying Data with Boolean Indexing

Boolean indexing isn't just for selection; it's also incredibly useful for modifying array elements that meet certain criteria. You can use a boolean index on the left side of an assignment operation to change the values of all elements where the condition is True.

# Create an array with positive and negative numbers
arr = np.array([1, -2, 3, -4, 5, -6])
print(f"Original array: {arr}")

# Replace all negative numbers with 0
arr[arr < 0] = 0
print(f"Array after replacing negatives with 0: {arr}")

# Set all elements greater than 3 to a specific value (e.g., 100)
arr[arr > 3] = 100
print(f"Array after setting elements > 3 to 100: {arr}")

Output:

Original array: [ 1 -2  3 -4  5 -6]
Array after replacing negatives with 0: [1 0 3 0 5 0]
Array after setting elements > 3 to 100: [  1   0   3   0 100   0]

This technique is very common in data cleaning and preparation, for example, replacing invalid sensor readings or capping extreme values.

Boolean indexing provides a flexible and readable way to filter and manipulate your NumPy arrays based on the data itself, forming a significant part of efficient data analysis workflows.