That depends on what you want to do. The ‘omitnan’ or 'omitmissing' flags simply delete the NaN values before performing the calculation. If you are doing row-wise or column-wise calculations, that would likely be your best option as well, the doing the operation on each row or column in a loop.
If you are doing matrix calculations, removing the NaN values destroys the matrix architecture, so that is not an option. The best approach in that instance might be to interpolate the missing values. The interpolation method depends on where the NaN values are in the matrix (if they are at the row or column ends, 'nearest' would liely be the best option), and what you want to do with them.
