numpy dot product with missing values

Question

How do you do a numpy dot product where the two vectors might have missing values? This seems to require many additional steps, is there an easier way to do this?:

v1 = np.array([1,4,2,np.nan,3])
v2 = np.array([np.nan,np.nan,2,4,1])
np.where(np.isnan(v1),0,v1).dot(np.where(np.isnan(v2),0,v2))

Answer 1

We can use np.nansum to sum up the values ignoring NaNs after element-wise multiplication -

np.nansum(v1*v2)

Sample run -

In [109]: v1
Out[109]: array([  1.,   4.,   2.,  nan,   3.])

In [110]: v2
Out[110]: array([ nan,  nan,   2.,   4.,   1.])

In [111]: np.where(np.isnan(v1),0,v1).dot(np.where(np.isnan(v2),0,v2))
Out[111]: 7.0

In [115]: v1*v2
Out[115]: array([ nan,  nan,   4.,  nan,   3.])

In [116]: np.nansum(v1*v2)
Out[116]: 7.0

Answer 2

Another solution is to use masked arrays:

v1 = np.array([1,4,2,np.nan,3])
v2 = np.array([np.nan,np.nan,2,4,1])

v1_m = numpy.ma.array(v1, mask=numpy.isnan(v1))
v2_m = numpy.ma.array(v2, mask=numpy.isnan(v2))

numpy.ma.dot(v1_m, v2_m)