Indexing

Just as with other functionality, indexing resembles that of NumPy a lot.

You can access and modify individual elements of a bsparse matrix using the usual square bracket syntax. Unlike with scipy.sparse, changes to the sparsity structure are allowed!

>>> bcoo = bsp.zeros((3, 2))
>>> bcoo[0, 0]
COO(shape=(1, 1), nnz=0, dtype=float64)

With this empty matrix of bshape == (10, 10), you can even modify the row- and column-sizes of the matrix by setting a block.

>>> bcoo[1,0] = np.ones((2,3))
>>> bcoo
BCOO(bshape=(3, 2), bnnz=1 | shape=(4, 4), nnz=6)

When calling bsp.zeros(), all block-sizes are set to 1 automatically. Since there are no elements that constrain the sizes, we can just update the them to accommodate the block.

We can also use slicing syntax to get a sub-block of the matrix.

>>> bcoo[1:, :2]
BCOO(bshape=(2, 2), bnnz=1 | shape=(3, 4), nnz=6)
>>> bcoo[1:, 2::-1]
BCOO(bshape=(2, 2), bnnz=1 | shape=(3, 4), nnz=6)

Another fun thing to do is mixing dense and sparse arrays.

>>> bcoo[0,-1] = bsp.sparse.random((2,2), density=0.9)
>>> bcoo.toarray()
array([[0.        , 0.        , 0.        , 0.53923029, 0.91937085],
       [0.        , 0.        , 0.        , 0.69792673, 0.        ],
       [1.        , 1.        , 1.        , 0.        , 0.        ],
       [1.        , 1.        , 1.        , 0.        , 0.        ],
       [0.        , 0.        , 0.        , 0.        , 0.        ]])