Is your feature request related to a problem? Please describe.

np.matrix is deprecated, and is slowly wiped from the public code. New users, I hope, don't learn what np.matrix is, and that's to the better.

At the same time, sparse matrices have a public interface for creating numpy matrices, which is actually used quite a lot in scipy tests. A new user who is unaware about the existence of numpy matrix type does not have a clear way to anticipate that a matrix is returned, or that it's deprecated. So a possible scenario (from my personal experience, even though I'm not a new user) is that:

• The user calls spmatrix.todense()
• The following code either emits a deprecation warning for doing matrix operations, or does the usual confusing stuff with matrix vector multiplication.
• The user realizes they should have used spmatrix.toarray(), switches to it and carries on.

Describe the solution you'd like

I propose to move towards making .todense an alias for .toarray through a deprecation cycle. This could be done by:

0. Eliminating .todense from scipy own codebase
1. Adding a keyword-only argument return_array=False to .todense and emitting a deprecation warning if it isn't set to True.
2. Making return_array=True a default
3. Eventually eliminating it and making .todense an alias to .toarray.

Describe alternatives you've considered

Other options are:

• Keeping todense around until numpy removes it.
• Making it an alias right away and hoping nobody will notice 🙃 idk