Compatible Libraries

A list of Python libraries, beyond just numpy and pandas, providing mathematical algorithms that are compatible with grapp’s LinearOperator interface.

scipy

Many of the functions in the scipy.sparse.linalg module are compatible with LinearOperator. Of particular note are:

• The conjugate gradient method scipy.sparse.linalg.cg

• Least squares solver scipy.sparse.linalg.lsqr

• Eigenvalue decomposition via scipy.sparse.linalg.eigs

• The preconditioned eigensolver scipy.sparse.linalg.lobpcg

PyLops

PyLops is a library focused on matrix-free methods, primarily focused on signal processing, but with some general purpose methods as well. The grapp LinearOperator classes are compatible with PyLops. Of particular note are:

• The HutchPP estimator for the trace of a matrix, pylops.utils.estimators.trace_hutchpp

• Combining multiple LinearOperator via pylops.HStack, pylops.VStack, and pylops.BlockDiag

• Kronecker product of LinearOperator via pylops.Kronecker

• Various optimization solvers with forced sparsity, such as pylops.optimization.cls_sparsity.FISTA

spgl1

spgl1 is a regularized least-squares solver (e.g., LASSO) that is compatible with LinearOperator.