mario.set_linear_solver(solver: str)

Set the default backend used by solve-based demand-driven formulas.

Parameters:

solver – Linear solver backend name: "scipy" (the default) or "umfpack". UMFPACK requires the optional scikit-umfpack package and its native SuiteSparse libraries.

Notes

This setting only affects solve-based IOT/SUT formulas. It has no effect on the explicit-inverse path used to build the w/wcc/waa Leontief inverse matrices directly.

UMFPACK performs sparse direct LU solves with float64 precision and reuses factors within a resolution request. Set the linear strategy to "direct" or "auto"; "iterative" is incompatible with this backend. Numerical failures raise an error without a pseudo-inverse or least-squares fallback. Sparse indices use 64 bits to avoid the native int32 factor allocation limit. The solver compares fill-reducing orderings before factorization. Singular components are checked before factorization and reported with their node labels. By default they raise an error. An explicit mario.set_singular_policy() of "partial" can instead return independent coefficients with a warning and NaN at affected nodes. Results satisfy a relative residual threshold of 1e-8, applied only to retained equations when a partial result is requested. Install with conda install -c conda-forge scikit-umfpack. With SuiteSparse already installed, pip install mariopy[umfpack] is also supported.

Examples

>>> mario.set_compute_method("solve")
>>> mario.set_linear_solver("umfpack")
>>> mario.set_linear_strategy("direct")