I took that very seriously when you said that matrices were important to you. Far from me the idea of forbidding numpy users to use matrices.

My point was the fact that newcomers are confused by the presence of both matrices and arrays. I think that there should be only one matrix/vector/tensor object in numpy. Therefore I would advocate the removal of matrices from numpy.

*But* why not have matrices in a different component? Maybe as a part of scipy? or somewhere else? You would be more than welcome to use them anywhere. Note that I use components outside numpy for my teaching (scipy, sympy, mayavi, nosetest) and I don't have any problems with that.

With my "argument" I endeavoured to explain the potential complications of using matrices instead of arrays when teaching. Perhaps the strongest argument against matrices is that you cannot use vectors. I've taught enough matlab courses to realise the pain that this represents for students. But I realise also that somebody else would have a different experience.

Of course x.T*y should be a 1x1 matrix, this is not an anomaly, but it is confusing for students, because they expect a scalar. That is why I prefer to teach with dot. Then the relation matrix/vector/scalar is crystal clear.