Suppose I have a NxN matrix M (lil_matrix or csr_matrix) from scipy.sparse, and I want to make it (N+1)xN where M_modified[i,j] = M[i,j] for 0 <= i < N (and all j) and M[N,j] = 0 for all j. Basically, I want to add a row of zeros to the bottom of M and preserve the remainder of the matrix. Is there a way to do this without copying the data?
I don't think that there is any way to really escape from doing the copying. Both of those types of sparse matrices store their data as Numpy arrays (in the data and indices attributes for csr and in the data and rows attributes for lil) internally and Numpy arrays can't be extended.
Update with more information:
LIL does stand for LInked List, but the current implementation doesn't quite live up to the name. The Numpy arrays used for data and rows are both of type object. Each of the objects in these arrays are actually Python lists (an empty list when all values are zero in a row). Python lists aren't exactly linked lists, but they are kind of close and quite frankly a better choice due to O(1) look-up. Personally, I don't immediately see the point of using a Numpy array of objects here rather than just a Python list. You could fairly easily change the current lil implementation to use Python lists instead which would allow you to add a row without copying the whole matrix.