The special case of Not-A-Number (NaN)

pandas mathematical operators and functions handle NaN in a special manner (compared to NumPy) that does not break the computations. pandas is lenient with missing data assuming that it is a common situation.

To demonstrate the difference, we can examine the following code, which calculates the mean of a NumPy array:

In [54]:
   # mean of numpy array values
   nda = np.array([1, 2, 3, 4, 5])
   nda.mean()

Out[54]:
   3.0

The result is as expected. The following code replaces one value with a NaN value:

In [55]:
   # mean of numpy array values with a NaN
   nda = np.array([1, 2, 3, 4, np.NaN])
   nda.mean()

Out[55]:
   nan

When encountering a NaN value, NumPy simply returns NaN. pandas changes this, so that NaN values are ignored:

In [56]:
   # ignores NaN values
   s = pd.Series(nda)
   s.mean()

Out[56]:
   2.5

In this case, pandas override the mean function of the Series object so that NaN values are simply ignored. They are not counted as a 0 value...