I have a NumPy matrix like this one (it could have several columns, this is just an example:
array([[nan, nan], [nan, nan], ['value 1', nan], [nan, nan], [nan, nan], [nan, 'value 2']], dtype=object)
I need to merge all columns in this matrix, replacing nan values with the corresponding non-nan value (if exists). Example output:
array([[nan], [nan], ['value 1'], [nan], [nan], ['value 2']], dtype=object)
Is there a way to achieve this with some built-in function in NumPy?
EDIT: if there is more than one non-nan in single row, I will take the first non-nan value.
Values could be string, float or int
Advertisement
Answer
Find the rows where the first column is nan
. This works because nan!=nan
:
rows = arr[:,0] != arr[:,0]
Update the first element of each chosen row with the second element:
arr[rows,0] = arr[rows,1]
Select the first column:
arr[:,[0]]