Missing Data with NaN

Real datasets have gaps. np.NaN (Not a Number) is a special floating-point value that acts as a placeholder for missing or undefined numerical data, so you can keep working with imperfect data.

1Three things to know about NaN

It propagates: any arithmetic involving np.NaN results in np.NaN. This "contagious" nature means missing values are never silently ignored.
It never equals itself: np.nan == np.nan is False. So you can't find NaN with == — use np.isnan().
It's a float: putting a NaN into an integer array converts the whole array to float.

2Play with it

Take the array [1, 2, np.nan, 4] for a spin. Add a number to every element and watch the NaN refuse to change, then use np.isnan() to hunt it down.

Play with NaN

arr = np.array([1, 2, np.nan, 4])

nan

Notice the trailing dot — NumPy prints array floats as 1., 2. (not 1.0). That single nan turned the whole array into floats (dtype: float64), even though we typed whole numbers.

arr + 10

np.isnan(arr)

Because np.nan != np.nan, you can’t find it with ==. Use np.isnan() instead.

playground.py

import numpy as np

arr = np.array([1, 2, np.nan, 4])
print(10 + np.nan)          # nan  (propagates)
print(np.nan == np.nan)     # False (never equals itself)
print(np.isnan(arr))        # the correct way to find NaN

import numpy as np

arr = np.array([1, 2, np.nan, 4])
print(10 + np.nan)          # nan  (propagates)
print(np.nan == np.nan)     # False (never equals itself)
print(np.isnan(arr))        # the correct way to find NaN

Common trap

Never test for missing values with arr == np.nan. Because nothing equals NaN, this compares element-by-element and returns an array of all False — even at the NaN position — so it never finds anything (e.g. [False False False False]). Always reach for np.isnan(arr) instead.

Quick Check

How do you correctly detect NaN values in an array?

← Previous

🧱 Creating Arrays

Introduction to Pandas 🐼