I was trying to debug some code today and found that I had a nan value propagating through some calculations, causing very weird behavior. I figured there must be a quick way to check numpy arrays for nan values. We can use the numpy function isnan:

>>> a = np.array(range(5,20,2))
>>> a
array([ 5,  7,  9, 11, 13, 15, 17, 19])
>>> b = np.append(a,np.nan)
>>> b
array([  5.,   7.,   9.,  11.,  13.,  15.,  17.,  19.,  nan])
>>> np.isnan(a)
array([False, False, False, False, False, False, False, False], dtype=bool)
>>> np.isnan(b)
array([False, False, False, False, False, False, False, False,  True], dtype=bool)

Here we can see that isnan returns a boolean array in the same shape as the input data, with a value of True indicating that the value at that point in the array is a nan. We can use these booleans to slice the arrays to access the nans:

>>> b[np.isnan(b)]
array([ nan])

But that’s not particularly useful, so lets invert the logic using ~ so we can get all of the non nan data:

>>> b[~np.isnan(b)]
array([  5.,   7.,   9.,  11.,  13.,  15.,  17.,  19.])

This performs a unary bitwise inversion on each element in our array, swapping True to False and False to True. Finally, if we are not interested in where the nans are, but just want to know if they are there or not, we can use any() to return a boolean if any value in the array is true:

>>> np.isnan(a).any()
False
>>> np.isnan(b).any()
True

Or if we wanted to check that every value in an array was true, for example if we had all nans, we can use all():

>>> c = np.array([np.nan,np.nan])
>>> c
array([ nan,  nan]
>>> np.isnan(a).all()
False
>>> np.isnan(b).all()
False
>>> np.isnan(c).all()
True

Or to check that every value is not a nan we can use all() and ~ together:

>>> ~np.isnan(a).all()
True
>>> ~np.isnan(b).all()
True
>>> ~np.isnan(c).all()
False