Rearrange sparse arrays by swapping rows and columns
CSC format keeps a list of the row indices of all non-zero entries, CSR format keeps a list of the column indices of all non-zero entries. I think you can take advantage of that to swap things around as follows, and I think there shouldn't be any side-effects to it:
def swap_rows(mat, a, b) :
mat_csc = scipy.sparse.csc_matrix(mat)
a_idx = np.where(mat_csc.indices == a)
b_idx = np.where(mat_csc.indices == b)
mat_csc.indices[a_idx] = b
mat_csc.indices[b_idx] = a
return mat_csc.asformat(mat.format)
def swap_cols(mat, a, b) :
mat_csr = scipy.sparse.csr_matrix(mat)
a_idx = np.where(mat_csr.indices == a)
b_idx = np.where(mat_csr.indices == b)
mat_csr.indices[a_idx] = b
mat_csr.indices[b_idx] = a
return mat_csr.asformat(mat.format)
You could now do something like this:
>>> mat = np.zeros((5,5))
>>> mat[[1, 2, 3, 3], [0, 2, 2, 4]] = 1
>>> mat = scipy.sparse.lil_matrix(mat)
>>> mat.todense()
matrix([[ 0., 0., 0., 0., 0.],
[ 1., 0., 0., 0., 0.],
[ 0., 0., 1., 0., 0.],
[ 0., 0., 1., 0., 1.],
[ 0., 0., 0., 0., 0.]])
>>> swap_rows(mat, 1, 3)
<5x5 sparse matrix of type '<type 'numpy.float64'>'
with 4 stored elements in LInked List format>
>>> swap_rows(mat, 1, 3).todense()
matrix([[ 0., 0., 0., 0., 0.],
[ 0., 0., 1., 0., 1.],
[ 0., 0., 1., 0., 0.],
[ 1., 0., 0., 0., 0.],
[ 0., 0., 0., 0., 0.]])
>>> swap_cols(mat, 0, 4)
<5x5 sparse matrix of type '<type 'numpy.float64'>'
with 4 stored elements in LInked List format>
>>> swap_cols(mat, 0, 4).todense()
matrix([[ 0., 0., 0., 0., 0.],
[ 0., 0., 0., 0., 1.],
[ 0., 0., 1., 0., 0.],
[ 1., 0., 1., 0., 0.],
[ 0., 0., 0., 0., 0.]])
I have used a LIL matrix to show how you could preserve the type of your output. In your application you probably want to already be in CSC or CSR format, and select whether to swap rows or columns first based on it, to minimize conversions.