How can I check for NaN values?

In the realm of Python

Python, the dragon language of programming, offers its own ways to sense the presence of NaNs.

Detecting NaN in Python standard library

Python brings its own touch with math.isnan(). This function is your faithful scout, looking for NaN values in the palm trees of your code.

import math

# Meet the NaN catcher of the Python realm
value = float('nan')
print(math.isnan(value))  # Gotcha! I mean... True


Good to remember: a NaN is the introvert of number land - it doesn't even equal itself! This quirk can be used to detect a NaN.

is_shy = float('nan')
print(is_shy != is_shy)  # True. Told you, it's shy!


NaN in pandas - the King of dataville

In the kingdom of data manipulation known as pandas, NaN is viewed as part of the royal family - the missing king. Here’s how to find him:

import pandas as pd

df = pd.DataFrame([1, 2, np.nan, 4])
print(df.isna())  # Assembling the royal search party


A NaN detector, tailored for you

Living with NaN means occasional encounters, and creating a function like isNaN could be your royal guard against frequent NaN ambushes:

# NaN detector at your service
def isNaN(num):
    return num != num

# Putting the detector to the test
print(isNaN(float('nan')))  # Affirmative! NaN detected!


Those NaN quirks that make them unique

Looks can deceive

NaN might look like a number but it follows its own rules. But remember, NaN doesn't equal even itself:

nan_number = float('nan')
print(nan_number == nan_number)  # False. You got pranked!


Consistency? Check!

Detecting NaN might differ in terms of function calls, but math.isnan(), np.isnan(), and pd.isna() all agree when faced with a NaN:

# The NaN suspect
nan_suspect = float('nan')

# The detective squad
print(math.isnan(nan_suspect))  # True. Fingerprints match!
print(np.isnan(nan_suspect))    # True. DNA confirmed!
print(pd.isna(nan_suspect))     # True. The hat fits perfectly!


In the name of performance

In this speed-obsessed world, simplicity trumps complexity. For NaN detection, x != x might outrun other methods. Time it, I dare you:

import timeit

# The race is on, and...it's a simple check by a nose!
simple_check_time = timeit.timeit('x != x',
                                  setup='x = float("nan")',
                                  number=1000000)

math_check_time = timeit.timeit('math.isnan(x)',
                                setup='import math; x = float("nan")',
                                number=1000000)

# Catch a breath and print the times
print(f"Simple check: {simple_check_time}")
print(f"math.isnan(): {math_check_time}")
