aa is an n by 1 array (aa.shape yields (n,1)). bb is an n array (bb.shape yeilds (n,)). cc = aa - bb. cc is an n by n array (cc.shape yields (n,n). cc should be n by 1 at most. Is this how numpy is supposed to work? Is it a bug? It's difficult to avoid getting an n by nothing array.
here's the MRE you posted in the comment, reformatted:
>>> aa=np.array([[1],[2],[3],[4]])
>>> bb=np.array([1,2,3,4])
>>> aa - bb
array([[ 1, 0, -1, -2], [ 2, 1, 0, -1], [ 3, 2, 1, 0]])
What are you trying to do here? the problem is that numpy interprets each nested string as a new dimension, so aa has 4 rows and 1 column, while bb has 1 row. When you subtract them, numpy interprets it as subtracting each value in bb from each row in aa. this means that the result will have as many rows as aa, with the value of each column being the difference between that row of aa with the corresponding value of bb.
if you just want to subtract them as two 1-dimensional vectors, then you need to initialize aa differently or use the flatten method:
>>> aa=np.array([[1],[2],[3],[4]])
>>> bb=np.array([1,2,3,4])
>>> aa = aa.flatten()
>>> aa - bb
array([0, 0, 0, 0])