RFC deprecate numpy.matrix

[lorentzenchr]

It has been mentioned here and there, but I don't find any "official" document about it. Is there a plan to deprecate and finally remove numpy.matrix?

A clear plan would also help with the scipy.sparse transition from sparse matrices to sparse arrays, see scipy/scipy#24802 and scikit-learn/scikit-learn#31177 (in the sense when to remove sparse matrices at the latest).

[charris]

Other way round, we are waiting on scipy.

[lorentzenchr]

The scipy roadmap says

SciPy sparse matrices are being replaced by sparse arrays. The sparse matrix formats are mostly feature-complete, however their main issue is that they act like numpy.matrix (which will be deprecated in NumPy at some point).

Does the snake bite its own tail?

[ngoldbaum]

I think Chuck is saying that there shouldn't be anything blocking SciPy from deprecating their uses of numpy.matrix, but we probably won't follow suit with numpy.matrix itself until SciPy has done so.

[rkern]

Yes, we're definitely intending to deprecate numpy.matrix. That intention, formal or not, motivates the move in scipy.sparse. On the numpy side, we didn't want to schedule anything until we knew there was a clear path to sparse arrays, and we'll likely pull the trigger after scipy does; we're at the base of the stack and don't want to topple anyone. We're keeping numpy.matrix only reluctantly until this main use of it is out of the way.

But please, yes, full steam ahead. We're behind you all the way.

[jakevdp]

Does the snake bite its own tail?

scipy.sparse depends on numpy.matrix, but not vice-versa (for example, spmatrix.todense() returns a np.matrix instance)

Because of this, NumPy could not remove np.matrix without breaking SciPy, unless SciPy first removes these uses.

The blocker for SciPy is not NumPy, but rather the many downstream consumers of scipy sparse matrices.