Numpy matrix multiplication returns nan

Question

I have two 2-D matrices, and I want to multiply these two matrices to get a new matrix. The first matrix A has dimension 943 x 1682, and it is shown below:

[[ 5.  3.  4. ...,  0.  0.  0.]
 [ 4.  0.  0. ...,  0.  0.  0.]
 [ 0.  0.  0. ...,  0.  0.  0.]
 ..., 
 [ 5.  0.  0. ...,  0.  0.  0.]
 [ 0.  0.  0. ...,  0.  0.  0.]
 [ 0.  5.  0. ...,  0.  0.  0.]]

And another matrix B has dimension 1682 x 20, and shown below:

[[ 0.          0.          0.         ...,  0.          0.          3.        ]
 [ 0.          0.57735027  0.57735027 ...,  0.          0.          3.        ]
 [ 0.          0.          0.         ...,  0.          0.          1.        ]
 ..., 
 [ 0.          0.          0.         ...,  0.          0.          2.        ]
 [ 0.          0.          0.         ...,  0.          0.          1.        ]
 [ 0.          0.          0.         ...,  0.          0.          1.        ]]

However, when I try A.dot(B), or np.matmul(A,B), I got a new matrix whose values are all nan, as shown below:

[[ nan  nan  nan ...,  nan  nan  nan]
 [ nan  nan  nan ...,  nan  nan  nan]
 [ nan  nan  nan ...,  nan  nan  nan]
..., 
 [ nan  nan  nan ...,  nan  nan  nan]
 [ nan  nan  nan ...,  nan  nan  nan]
 [ nan  nan  nan ...,  nan  nan  nan]]

I figure this might be a result of multiplying 0. But why would it return nan at every position? And how should I deal with this so that I can get numbers instead of nan?

Thank you very much for help!

Answer 1

A single nan column in the first matrix, and\or a single nan row in the second matrix, could cause this issue. A way to verify that indeed all values are valid in both matrices is to filter out the nans and see if the shape remains the same:

a_shape_before = A.shape
a_shape_after = A[numpy.logical_not(numpy.is_nan(A))].shape
assert a_shape_before == a_shape_after

And likewise for B.