Skip to content

scipy.sparse API

Sparse matrix class

  • bsr_matrix

    Block Sparse Row format

  • coo_matrix

    Coordinate format.

    No support for arithmetic operators, and no slicing.

    Usually used to construct, then changed to csr/csc for arithmetic.

    coo_matrix(D)
coo_matrix((m, n), dtype=np.int32)
coo_matrix((data, (I, J)), shape=[M, N])

coo.toarray()
coo.shape/ndim/dtype
coo.tocsr() # csr_matrix(coo)

  • csr_matrix

    Compressed Sparse Row format

    efficient row slicing and arithmetic operations between CSRs.

  • csc_matrix

    Compressed Sparse Column format

  • dia_matrix

    Diagonal storage format

    dia_matrix((data, offsets=0), shape=(M,N))

  • dok_matrix

    Dictionary Of Key format

  • lil_matrix

    Linked List format

Functions

eye(m[, dtype, format="dia"])
identity(n[, dtype, format])
diags(diagonals[, offsets, shape, format, dtype])
rand(m, n[, density, format, dtype])
issparse(x) # isspmatrix(x)
isspmatrix_csc(x)

linalg

inv(A)
expm(A) # exp using Pade approximation

eigs(A[, k, M, which])
    '''
    k: calculate the first k eig
    M: generalized eigenproblem Ax=wMx
    which: which first k eig to find
        "LM": largest magnitude
        "SM": smallest magnitude
        "LR": largest real part
        "LI": largest imaginary part
    '''
eigsh(A[, k]) # fast for real sym mat