Floating Point Functions in NumPy

Detection Functions

NumPy includes a set of functions that allow you to detect special floating-point values such as NaN (Not a Number), inf (infinity), and check whether values are finite or not. These are crucial for handling edge cases in scientific and numerical computations.

import numpy as np

data = np.array([1.5, np.nan, np.inf, -np.inf, 0.0])

# Detection functions
print("isnan:", np.isnan(data))      # [False  True False False False]
print("isinf:", np.isinf(data))      # [False False  True  True False]
print("isfinite:", np.isfinite(data))# [ True False False False  True]

import numpy as np

data = np.array([1.5, np.nan, np.inf, -np.inf, 0.0])

# Detection functions
print("isnan:", np.isnan(data))      # [False  True False False False]
print("isinf:", np.isinf(data))      # [False False  True  True False]
print("isfinite:", np.isfinite(data))# [ True False False False  True]

Function Breakdown:

• np.isnan(x): Returns True for elements that are NaN (Not a Number).
• np.isinf(x): Returns True for positive or negative infinity values.
• np.isfinite(x): Returns True for elements that are neither NaN nor infinite.

Output

isnan: [False  True False False False]
isinf: [False False  True  True False]
isfinite: [ True False False False  True]

isnan: [False  True False False False]
isinf: [False False  True  True False]
isfinite: [ True False False False  True]

💡 Tip: Use these detection functions before performing operations like division or logarithms to avoid runtime warnings or invalid results.

Rounding Functions

NumPy provides several functions to round floating-point numbers in different ways. These functions operate element-wise and are useful for formatting results, simplifying values, or preparing data for comparisons.

The most commonly used rounding functions include around(), floor(), ceil(), and trunc(), each serving a specific purpose.

import numpy as np

data = np.array([1.7, 2.3, -1.5, -2.7])

print("Original:", arr)
print("around:", np.around(data))  # [ 2.  2. -2. -3.]
print("floor:", np.floor(data))    # [ 1.  2. -2. -3.]
print("ceil:", np.ceil(data))      # [ 2.  3. -1. -2.]
print("trunc:", np.trunc(data))    # [ 1.  2. -1. -2.]

import numpy as np

data = np.array([1.7, 2.3, -1.5, -2.7])

print("Original:", arr)
print("around:", np.around(data))  # [ 2.  2. -2. -3.]
print("floor:", np.floor(data))    # [ 1.  2. -2. -3.]
print("ceil:", np.ceil(data))      # [ 2.  3. -1. -2.]
print("trunc:", np.trunc(data))    # [ 1.  2. -1. -2.]

Function Breakdown:

• np.around(x): Rounds to the nearest integer. By default, uses IEEE 754 rules (round half to even).
• np.floor(x): Rounds each element down to the nearest integer less than or equal to the value.
• np.ceil(x): Rounds each element up to the nearest integer greater than or equal to the value.
• np.trunc(x): Truncates the decimal part — keeps the integer portion of the number.

Output

Original: [1.7, 2.3, -1.5, -2.7]
around: [ 2.  2. -2. -3.]
floor: [ 1.  2. -2. -3.]
ceil: [ 2.  3. -1. -2.]
trunc: [ 1.  2. -1. -2.]

Original: [1.7, 2.3, -1.5, -2.7]
around: [ 2.  2. -2. -3.]
floor: [ 1.  2. -2. -3.]
ceil: [ 2.  3. -1. -2.]
trunc: [ 1.  2. -1. -2.]

💡 Tip: If you're preparing data for indexing or rounding to specific decimal places, np.around(x, decimals=n) allows control over precision.

Floating Point Limits

NumPy provides access to detailed information about floating point types on your system using the np.finfo() function. This is especially useful when you need to understand the numerical limits and precision of floating-point data types such as float32 or float64.

These properties help avoid overflow or underflow issues, and are often used when setting tolerances for numerical comparisons or algorithms.

import numpy as np

# Get floating point information for float64
info = np.finfo(np.float64)

print("Max:", info.max)
print("Min:", info.min)
print("Smallest positive:", info.tiny)
print("Epsilon:", info.eps)

import numpy as np

# Get floating point information for float64
info = np.finfo(np.float64)

print("Max:", info.max)
print("Min:", info.min)
print("Smallest positive:", info.tiny)
print("Epsilon:", info.eps)

Function Breakdown:

• info.max: Largest representable finite float.
• info.min: Smallest representable (most negative) finite float.
• info.tiny: Smallest positive normalized number (not subnormal).
• info.eps: The difference between 1.0 and the next representable float — defines machine precision.

Output (float64)

Max: 1.7976931348623157e+308
Min: -1.7976931348623157e+308
Smallest positive: 2.2250738585072014e-308
Epsilon: 2.220446049250313e-16

Max: 1.7976931348623157e+308
Min: -1.7976931348623157e+308
Smallest positive: 2.2250738585072014e-308
Epsilon: 2.220446049250313e-16

💡 Tip: Use np.finfo(dtype).eps when setting numerical tolerances to compare floating point values safely (e.g., in iterative algorithms or equality checks).

Frequently Asked Questions

What's Next?

Coming up next, we’ll dive into Matrix Multiplication in NumPy — a fundamental operation in linear algebra used across data science, machine learning, and scientific computing.